Dr. Ian Cutress: Custom PCs that actually are custom!
We talk through our custom PCs with Dr. Ian Cutress, covering bespoke heatsinks, our custom Magnetar V Series and rack mount options – not forgetting our Threadripper Pro Magnetar workstations. Find out what we chatted about on Ian’s video!
Why Armari are the workstation specialists you need
Statement from Dan Goldsmith Technical Director
For our first 18 years of business from 1994 to around 2012, Armari was a traditional system builder. We built systems using all off the shelf components. As time went on, particularly for workstation builds we were finding we had to make compromises. We would pick the best motherboard and cooler we wanted to use, but found these wouldn’t fit into the chassis the customer had the space for, so we would end up making a compromise somewhere. Most of the off the shelf PC chassis on the market today are designed for the DIY, build at home market. There are very few good workstation chassis on the market right now that are great for type of workstations customers are looking for today, especially at the high end. 90% of off the shelf chassis are made from very thin steel to cut down on cost and shipping weight. They then tempered glass side panels and a few cheap fans thrown in. None of them are UKCA or CE EMC emission compliant, which is important for certain customers.
When you’re dealing with the latest high end workstation processors we have in the market today, you can realistically be looking at CPU that puts out 500Watts and then you want to drop in a couple of 350Watts GPUs and you’re got 1,200Watts of internal heat to manage. If you run these flat out 24/7 in a case that was designed for a single GPU and gaming CPU and you have a lot of thermal problems, which can partly address with noisy fans, buts not a good experience for the customer.
So in 2012, we decided to get into low volume chassis manufacture, giving us the ability to design a chassis around the components we found to be best in the market, so we wouldn’t have to compromise. Manufacturing these in low volumes, means we can tweak and evolve our designs between batches, in a similar way that Tesla is constantly evolving their EV cars throughout the year, they don’t wait for a yearly model change to fix a problem. Having this vertical integration approach means, if NVIDIA drops a new graphics cards with an oversized cooler like happened with the RTX 3000 series, we can quickly change the cooling arranging inside our chassis so we manage the heat from the GPU so it doesn’t adversely limit clock frequency boost on the CPU.
As high-end workstation can get expensive these days, we wanted to make sure we had a strong structure but still keep it as light as possible, which is why we use a lot of aluminium in our designs. This also makes our chassis almost 100% recyclable as we don’t use plastics. We can ship any of Magnetar workstations globally in a moderately sized box and know when it arrives it will arrive working. We don’t need to rely on massive shipping boxes to protect a fragile PC during shipping as this is hugely expensive and wasteful today, as the strength is inherent in the chassis design itself. Every Magnetar chassis for example has adapter card transit support, so GPUs don’t come loose in transit.
We believe this vertical integration approach really does deliver the best experience, highest performance and best reliability for our customers. Yet it still allows us to be very nimble and transition to the latest technologies pretty much at launch. In the end we ultimately don’t have compromise, so we’re free to always choose from the best components the market has to offer for our customers.
We are now taking orders for Threadripper Pro 5000 Workstations, with delivery in 2 weeks. Experience the next level of performance in our 4 new Threadripper workstations.
Our Threadripper Pro 5000 workstations are more than 10% faster than competing solutions from either Lenovo and Dell and we are only the workstation company with a 2U Threadripper Pro 5000 workstation!
June 16th – AMD launches 3 new Zen2 based Ryzen processors, offering a boost in frequency at the same price as the previous X parts. The Ryzen 9 3900XT and Ryzen 7 3800XT now boost up to 4.7GHz, with the Ryzen 5 3600XT boosting up to 4.5GHz.
The existing Ryzen 9 3900X, Ryzen 7 3800X and Ryzen 5 3600X still remain available to order now.
Anandtech – The Armari Magnetar X64T Workstation OC Review: 128 Threads at 4.0 GHz, Sustained!
Dr. Ian Cutress of Anandtech puts one of the Armari Magnetar X64T workstations, currently the fastest workstation available on the market in 2020 thought its paces. With 128 Threads running up to 4.0GHz sustained, the Magnetar X64T is between 7 and 10x faster in real world workstations applications than a typical Intel Skylake Xeon E5v3 system which is used to get a baseline score of 1.
Blitzing around a race track in a fast car only ever convinces you of one thing: I need to go around the track even faster. I need a better car, I need a better engine, better brakes, or better tires. I need that special go faster juice, and I need to nail the perfect run. The world of professional computing works the same, whether it comes down to rendering, rapid prototyping, scientific compute, medical imaging, weather modelling, or something like oil and gas simulation, the more raw horsepower there is, the more can be done. So enter the new Armari Magnetar X64T – an overclocked 64-core Threadripper 3990X that holds the new SPECworkstation3 world record. We got hold of one. It’s really fast.
Playing with Performance
AMD’s Threadripper 3990X is one of those crazy processors. It comes at you with some of the best of any processor statistics: it has 64 cores and 128 threads, it has 256 MB of L3 cache, it has a TDP of 280 W, which allows for a 2.9 GHz base frequency up to a 4.3 GHz turbo. It is overclockable, and so with the right system those frequencies can go even higher. With the best binned 7nm chiplets, paired with quad-channel DDR4-3200 memory, for multithreaded workloads it is one of the ultimate powerhouses anyone can build in a single socket with a socketable processor.
In our initial review of the Threadripper 3990X, it blitzed any software that could take advantage of all those threads – the nearest competitors were the 32-core Threadrippers, or Intel’s 28-core Xeon-W processors. We even put it up against two of Intel’s $10000 28-core Xeons, and it won pretty much everything by a large margin.
So what happens when we overclock it? There are those that want more, and not just those overclocking for fun – workstation customers, like animation studios, are always looking for ways in which they can rapidly render frames for upcoming projects. If a cooling system can be built to withstand it, and the power is available, then there’s always scope to get more out of the hardware that comes from the big players. This is what the Armari Magnetar X64T Workstation is designed to do – get more.
To that end, today AMD and SPEC is announcing that the Magnetar X64T workstation, a system that you can buy, will off-the-shelf give the best performance in SPECworkstation3 ever seen.
The Magnetar X64T: Performance Reimagined
The key highlight from this review, should you not read any further, is that this system is built to blitz workloads. The Threadripper 3990X is usually fast enough in its own right, but Armari have gone above and beyond. The goal of this system is to be an off-the-shelf powerhouse that requires very little setup from its customers.
Armari, perhaps a lesser well known system integrator, is a company that has in recent years focused on building systems for 3D animation, video editing, and scientific research. With over 20 years of experience, Armari’s hardware has gone into high performance computing solutions and clusters that have featured in the TOP500 lists, as well as rendering server farms for the top animation, VFX, and CGI studios in Soho, London.
These are clients who want the best performance, and Armari positions itself not so much as a boutique system builder, but something between the big OEMs (like Dell/HP) and the main retailers to offer custom solutions by leveraging its network of cooling and hardware contacts around the world. This enables the company to build custom chassis, obtain optimized memory, order power supplies with custom connector configurations, and ensure consistency from batch-to-batch when ordering from its partners. In speaking to Armari’s Technical Director Dan Goldsmith, he mentioned that working with partner companies for so long has enabled them to get access to rapid prototyping and component consistency with continual feedback with partners such as EKWB, ASRock, Tyan, and many other ODM companies that Armari leverages on a regular basis.
The Magnetar X64T, I was told, leverages the strong relationship Armari has with AMD. The Opteron was a popular range a decade ago, and that partnership has been maintained through today. The goal of the Magnetar project was to create a system that offers the best that Threadripper has to offer while still enabling the under-the-desk workstation platform. This project has been slightly delayed due to COVID, and AMD now has Threadripper Pro, but those processors are not overclockable – for those that want raw performance, AMD and Armari believe they are on a winner.
The key to the system is in how Armari is cooling the processor, and the choice of components. The Magnetar X64T features a custom water cooling loop, which is perhaps not anything new in its own right, however the company has created a component chain to ensure consistency in its design, as well as using some of the most powerful options available.
The water block is probably the best place to start, because this is a completely custom-for-Armari design built in partnership with EK Water Blocks. This block is specifically built for this one motherboard, the ASRock TRX40 Taichi, and applies cooling to both the processor and the power delivery. The block works in conjunction with the highest-quality thermal paste pads on the market, to ensure a flat connection with the water block. As it also covers the power delivery, Armari worked with ASRock to enable a consistent z-height of all the power delivery components, something that can vary during manufacturing, and maintain that consistency on a batch-by-batch basis. Pair this up with Armari’s custom FWL liquid cooling pump, reservoir, tubing, 3x140mm radiator, and fan combinations (many of which are custom from their respective ODMs), and we have a cooling capacity in excess of 700 W. The coolant is a special long-life coolant designed for 24/7 over three years, and the standard warranty comes with service during those three years, including collection and return, at no extra cost.
Now, the ASRock TRX40 Taichi isn’t the top Threadripper motherboard on the market, and Armari fully admits that, however it points out that the best motherboard available costs twice as much. In working with ASRock, they were able to co-ordinate what was needed within the discrete motherboard component lists as well as enable a custom BIOS implementation for additional control. One of the tradeoffs I was told about is that a cheaper motherboard might mean slightly cheaper components, however Armari says that their cooling system and setup were co-operatively tuned to meet its customers’ demands.
With this cooling arrangement, Armari have fired up the overclock. In our initial review of the Threadripper 3990X, we were observing ~3450 MHz during our sustained running with the CPU reaching its full 280 W. For the Armari Magnetar X64T, we have an all-core frequency from 3950-4100 MHz, depending on the workload. Users might scoff at the +400-550 MHz lift, but bearing in mind this is across all of the 64 cores simultaneously, and the cooling is built such that this frequency is sustained for renders or simulations that might take days. Further details of frequency and power later in the review.
While having the overclocked CPU is great, the Magnetar X64T system we were delivered also had memory, graphics, and storage.
Armari Magnetar X64T as shipped (X64T-RD1600G3-FWL)
AMD Ryzen Threadripper 3990X
Overclocked to ~4.0 GHz All-Core Turbo
Custom Armari FWLv2 Liquid Cooling Loop
Custom CPU+VRM Monoblock
420x45mm EK Coolsense Radiator
3 x EK-Vardar 140ER EVO 140mm fans
High Performance Pump
Clear Coolant, Designed for 3yr operation
PNY NVIDIA Quadro RTX 6000 24 GB
ASRock TRX40 Taichi
256 GB of DDR4-3200
1600W 80PLUS Gold 93%, rated to 50ºC
0% fan under 40% load
9x PCIe connections
ASRock Hyper Quad M.2 PCIe 4.0 x16 add-in card
1 x Corsair MP600 PCIe 4.0 x4 1 TB Boot Drive
2 x Corsair MP600 PCIe 4.0 x4 1TB Striped Array
Realtek RTL8125 2.5 GbE (motherboard)
Intel I211-AT 1 GbE (motherboard)
Intel AX201 Wi-Fi 6 module (motherboard)
Onboard Realtek ALC1220 + ALC4050H
3 x EK 140mm for radiator
2 x Noctua 140mm for internal airflow
1 x SanAce 80mm low noise for DRAM
Price as Built
£10790 + tax
(~$14200 + tax)
Special launch price for September
1 Year RTB
3 Year Parts+Labor
One service/coolant replacement, inc collection/pickup
Loaner systems available if bigger issues occur
The system as shipped came with an PNY NVIDIA RTX6000 graphics card, which is essentially an RTX 2080 Ti on steroids with 24 GiB of GDDR6, and the system can be configured with two of them. As Threadripper is not an ECC-qualified platform, the X64T comes with the peak configuration supported, 256 GB, but with custom SPD profiles to run up to DDR4-3600. Unfortunately due to how quickly this system was rebuilt for this review, the system I was sent was using DDR4-3200 at CL20, as some of the original memory was accidentally splashed with coolant, and Armari wanted to ensure I wouldn’t have any issues with the system.
Storage comes in two forms, both of which are PCIe 4.0. As shipped, we were specified with a boot drive to the tune of a Corsair MP600 1 TB PCIe 4.0 x4 drive. Another two of these drives were provided inside an ASRock Hyper M.2 PCIe 4.0 card, plugged into one of the PCIe 4.0 slots. Armari says that as newer and bigger PCIe 4.0 drives come to market beyond the Phison E16 solutions, this should expand to higher capacity drives or faster drives as required.
The power supply is a fully custom 1600W 80PLUS Gold unit, rated to run at 50 ºC with 93% efficiency. It has a custom fan profile directly from the OEM, and is set to only stir up the fans if the power required goes above 40% (640 W). The fully modular PSU has nine 8-pin connections and five 6-pin connections, providing 14 total, should any customer want to go above and beyond. The PSU on its own has a 10-year warranty.
The motherboard has a 2.5 GbE wired network port and a 1 GbE wired network port, and Armari does offer a 10G upgrade (space permitting based on the PCIe slots). Wi-Fi 6 support comes as standard, as does the ALC1220 audio configuration.
The chassis is the last custom part to discuss, with the system featuring the Magnetar naming on the front with the Armari logo. The chassis is big, but quite standard for a high-end workstation platform: 53cm x 22cm x 57cm (20.9-in x 8.7-in x 22.4-in), with a typical single GPU weight of 18 kg (39.7 lbs).
The chassis comes with handles on top that fold away, making the system easy to move around as required. I love these.
Inside there is lots of ‘space’ for directed airflow. The pump and reservoir is found in the bottom of the case, underneath the standard drive-bays, while the 3x140mm double thick radiator is at the top built into the side of the chassis. This is a special hinged mount, which makes the side panel easy to remove and the cooling apparatus easy to inspect.
There is a PCIe retention bracket for any add-in card installed, and in the base of the chassis is the power supply, hidden away. The insides weren’t necessarily built to look aesthetically pleasing, however the system as provided by Armari has a nice clean look.
Due to a couple of issues with arranging this system for review, I was told that normally Armari adds in some custom sealant to help with the liquid cooling loop, however as it requires 24 hours to set, they weren’t able to in this instance. The liquid cooling loop is pre-tested for every system they build at over 1 bar of pressure, along with full stability testing and thermal testing before shipping. For any reason if a system needs to be returned for warranty, Armari can supply a loaner system if required. As mentioned above, the standard warranty includes one full service and inspection, and the coolant can be replaced in order to give the customer another 3 years of ‘hassle free’ operation.
The News Today: World Records
Today AMD and Armari are announcing that the new Magnetar X64T has set a new world record in the SPECworkstation 3 benchmark. The system that achieved this test is, by and large, the system I am testing today (it was stripped down and rebuilt with an updated water block). For the customers that Armari typically services this one of the primary benchmarks they care about, and so getting a new world record for a commercially available system should put Armari’s offerings high on their list.
Our testing, as shown over the next few pages, is similarly impressive. We already saw that the Threadripper 3990X with no overclock was a formidable beast in almost every one of our rendering and compute workloads. The only real comparison point we have to compare against is our W-3175X workstation that was provided when we reviewed that system.
The Magnetar X64T-RD1600G3 FWL (the full name) system in our testing is ~£10790 ($14200) excluding tax . This includes a Windows 10 Professional 64-bit license, and Armari’s 3 year premium workstation warranty, with 1-year on site and 2/3rd year parts and labor, along with a loaner system for the duration of any repairs.
With its new 64-core Ryzen Threadripper 3990X, AMD has completely transformed the design viz workstation. Ray trace rendering and CAD have never been such good bedfellows, writes Greg Corke
A few years ago, it was almost unthinkable for a CPU to have 64 cores. And, even if it could, the frequency of those cores would be so low you wouldn’t really want one inside your workstation. Lots of cores are great for rendering, but if that comes at the expense of single threaded performance, which is what makes CAD and many other design and engineering applications tick, then it’s a compromise few would be willing to make.
Until recently, advances in CPU technology had become quite predictable, but it’s amazing how quickly things can change. In summer 2017 AMD launched Ryzen Threadripper. The first-generation CPUs featured up to 16 cores and were great for multi-threaded workflows but lacked the all-important single threaded performance to make them a serious threat to Intel. Now two and a half years later, and with AMD’s 3rd Gen Threadripper rollout complete, that couldn’t be further from the truth.
Early February saw the release of Ryzen Threadripper 3990X, a 64-core, 128-thread beast of a CPU which hardly compromises on frequency at all. It has a base clock of 2.9GHz and a max boost of 4.3GHz, but with sufficient cooling in place it can even get close to 4.0GHz on all 64 cores. Viz specialists, architects, engineers and product designers can really have their cake and eat it too. It’s a phenomenal proposition for anyone that uses a CPU renderer.
As far as multi-threaded performance is concerned, Intel simply can’t compete. The closest it has is the Core i9-10980XE (18-cores) ($1,000) and the Xeon W-3175X (28-cores) ($2,999). If you want more cores, you’d need the server-focused Xeon Platinum 9282 (56- cores) ($50,000) or two Xeon Platinum 8280 (28-cores) ($20,000). And you still wouldn’t be able to beat Threadripper 3990X.
Of course, with limited competition on the desktop workstation front, AMD can charge a premium, and the Threadripper 3990X doesn’t come cheap. The 64-core CPU will set you back $3,990, matching its model number precisely. This might seem expensive but, when you consider the huge impact it can have on design viz workflows, many will consider it money well spent. After all, there’s only so many coffee breaks you can have in one day, as you wait for a render to complete.
The render god
Armari has a long history of developing high-performance workstations that are both extremely well-built and well-tuned. The UK firm was one of the first to get on board with 1st Gen Threadripper and now offers AMD CPUs across its entire Magnetar range, from single socket Ryzen and single socket Threadripper, to single and dual socket EPYC, which is AMD’s official enterprise CPU.
For the Threadripper 3990X, Armari has designed and manufactured a completely bespoke chassis to handle its extreme demands. The 3990X is rated at 280W Thermal Design Power (TDP) but, unlike AMD EPYC, it can actually be pushed much higher. And when more power gets pumped into the CPU, the brakes come off and Threadripper really starts to fly in multi-threaded workflows.
To do this, Armari uses AMD Precision Boost Overdrive, which will essentially continue to push the frequency of the CPU as long as the workstation can cool it adequately. And the Magnetar X64T-G3 certainly can. Its Full Water Loop (FWL) cooling system is impressive and comes with a giant radiator with nearly three times the surface area of those used in its other workstations.
On average, Armari reckons it can sustain 550 – 650 Watts of power in real world applications, with momentary boosts in excess of 800 Watts. In practice, this means the machine can maintain very high clock speeds over long periods of time; not just in single threaded workflows, but when rendering as well.
We left it rendering in the design viz focused KeyShot for well over an hour and it maintained a phenomenal 3.90GHz on all 64 cores. Fan noise was noticeable, but not too distracting. However, it’s important to note that this is a prototype system and, when the machine hits production, Armari says the radiator fans will be tuned back to around 50-60% duty cycle maximum.
The machine completed our 4K, 128 pass test render in a record breaking 38 secs. That’s nearly twice as fast as a 32-core Threadripper 3970X and more than ten times as fast as the six-core Intel Xeon E-2176G, the kind of CPU you’d typically find in a CAD workstation. It was also streets ahead of the competition in the V-Ray NEXT benchmark with a result of 93,436 ksamples.
But this isn’t just about numbers on charts. A CPU like this can have huge impact on workflow. With high-quality 1,280 x 720 resolution renders literally taking a few seconds and 1,920 x 1,080 resolution renders not much longer, there’s no more stop and start. You really can iterate in real time without having to compromise on quality or resolution. Although, naturally, this depends on the complexity of your scene.
Memory also plays a very important role in rendering. 3rd Gen Threadripper can support up to 256GB, spread across eight slots, which is important if you work with very large scenes. This is the theory, at least. Armari tells us compatible 32GB modules are currently quite expensive, which is one of the reasons why our test machine was configured with 64GB (4 x 16GB Corsair Vengeance LPX DDR4-3600 C18 SDRAM modules) – the other being that 64GB is a good amount for mainstream viz workflows.
But it’s not just about capacity. 3rd Gen Threadripper also features a new memory architecture which gives every single core fast and equal access to memory. In contrast, with 2nd Gen Threadripper not all cores had direct access to memory, so sometimes had to ask other cores for data and then wait for it to arrive. Chaos Group’s CTO Vladimir Koylazov explains this in more detail below.
Chaos Group on 3rd Gen Threadripper
Chaos Group’s CTO Vladimir Koylazov shares his thoughts on the Threadripper 3990X and what 64-cores and the new memory architecture of 3rd Gen Threadripper means for rendering.
The third generation Threadripper CPUs are great for ray tracing – and there is one crucial breakthrough that makes it possible. The Threadripper 3990X CPU implements uniform memory access for all cores, which gives a huge performance boost for rendering. Here is a short explanation.
Usually the main bottleneck for many-core machines is RAM – especially with ray tracing, different cores usually need different parts of the scene geometry or shaders. Scenes these days can be very large, measuring hundreds of gigabytes. Making sure that each CPU core gets the data that it needs from the system RAM as quickly as possible is a fairly difficult task.
To make it somewhat easier for hardware manufacturers, the so called “NUMA” architecture was introduced (where NUMA stands for “Non-Uniform Memory Access”). For multi-CPU systems, this means that each CPU only has access to a portion of the system RAM directly, and if it needs data from the other portions, it needs to ask another CPU to fetch it. Within a single CPU it means that only certain cores have direct access to the memory, and other cores must ask them to fetch the data they need. For ray tracing specifically, this adds quite a bit of overhead and typically NUMA configurations affect performance in quite a big negative way. Unfortunately, there is no easy way to optimise the software around this hardware peculiarity. This is one of the reasons why many-core dual-CPU systems have sometimes not performed as expected for our customers, especially with large scenes that are far larger than the CPU caches. We have profiled many such systems with CPUs from different manufacturers and, barring bugs or other multithreading problems, we have found that invariably the main bottlenecks occur when the CPU cores wait for data to arrive from the system RAM. Anything that slows that operation, like a NUMA architecture, has an adverse effect on performance.
In the newest [3rd Gen] Threadripper CPUs, all cores have equal access to the system RAM without additional delays like asking another core to fetch the data. This allows each CPU core to move through the compute operations a lot faster than previous NUMA-based architectures. What you get are 64 CPU cores operating closer to their maximum potential – which has not been possible with any other CPUs previously. This means that, on the whole, we have not had to do much to optimise the V-Ray and Corona code for the new Threadrippers.
We did have one piece of code in V-Ray (the light cache calculations) that was limited to 64 threads and which we had to rework a little bit in the latest V-Ray builds so that we can use all 128 logical threads, but beyond that, we only had to make sure that each CPU core can operate as independently as possible from the rest.
Of course, designers, engineers and architects aren’t only interested in ray trace rendering. Putting BIM and CAD to one side for a moment, there are several multithreaded tools that can benefit from multiple CPU cores, although very few apart from video encoding and editing that can use all 64 cores as efficiently as a ray trace renderer. Many simulation and point cloud processing tools, for example, are limited to a dozen or so cores, or offer very little additional benefit if your workstation CPU has more. With some applications, processing times can even go up once you hit the CPU core sweet spot for the software or dataset you’re working on.
With this in mind, unless you know for certain that your design and engineering focused software will benefit from 64 cores, we wouldn’t really recommend the Threadripper 3990X for anything other than ray trace rendering. Instead, money would probably be better spent on the 24-core AMD Ryzen Threadripper 3960X, which is less than half the price, or even the 16-core AMD Ryzen 9 3950X.
In saying this, it’s important to note that Threadripper beats Ryzen hands down when it comes to memory bandwidth and cache, both of which can be really important for memory intensive workflows like point cloud processing or simulation. 3rd Gen Threadripper can also support much more memory – 256GB compared to 128GB in 3rd Gen Ryzen.
For single threaded applications like CAD or BIM the 64-core 3990X is never going to beat a top-end eight core Intel CPU like the Core i9 9900K. But it’s not that far behind. It completed our Solidworks 2020 IGES export test in 84 secs, only 9 seconds slower than an overclocked 4.9GHz Core i9 9900K, which is still one of the best CPUs out there if you want a workstation that is 100% focused on CAD.
Inside the Magnetar X64T-G3
Armari’s 64-core Threadripper workstation is a serious piece of engineering and one of the heaviest workstations we’ve ever reviewed, thanks in part to its hefty cooling system. To make it easier to carry there are two handles on top that flip around 180 degrees, so they sit flush when not in use. Next to the front handle you’ll also find two USB 3.1 Gen 1 ports and (on the production version, at least) a USB 3.2 Gen2x2 port. There are plenty more USB ports on the rear of the machine, as well as two Ethernet ports (2.5Gb/s and 1Gb/s).
To handle the big power demands of the 3990X, Armari uses the ASRock TRX40 motherboard. With limited on-board M.2 sockets, it comes with a Hyper Quad M.2 PCIe add in board that can host up to four M.2 NVMe SSDs. In our review machine it’s populated with a pair of 1TB Corsair MP600 M.2 NVMe SSDs configured as a 2TB RAID 0 array.
The MP600 is based on PCIe 4.0, which offers twice the bandwidth of PCIe 3.0, so is already a fast SSD. It boasts sequential read and write speeds of 4.95GB/sec and 4.25GB/sec respectively but configuring it as RAID 0 takes it to new levels. In the CrystalDiskMark benchmark it clocked 9,062MB/sec read and 8,298MB/sec write and copied a 90GB zip file in just over 50 secs.
Of course, SSD performance like this will only truly benefit those working with colossal datasets typically used in workflows such as high-end viz, 8K video, point cloud or simulation. Anyone with more mainstream viz workflows will likely be more than happy with a single MP600, backed up by up one to four 3.5/2.5-inch SATA/SAS HDDs/SSDs.
If you’ve forked out $4k for a 64-core CPU, the chances are you’ll only really want to use the GPU for 3D graphics or VR. The Magnetar X64T-G3 came with a single AMD Radeon Pro W5700 (8GB), which is a decent choice for mainstream design viz. We review this pro GPU in detail in this article.
For more demanding 3D workflows, the machine can handle one or two Nvidia Quadro RTX 5000, 6000 or 8000 GPUs, but if you want three or four GPUs, perhaps for GPU rendering, then you’re best off talking to Armari. With a different motherboard (the ASRock TRX40 Creator) and a 2000W PSU (instead of our test machine’s 1300W EVGA SuperNOVA G2 GOLD Modular) the workstation can support four GPUs. However, it may not be able to deliver the same power to the Threadripper 3990X, so all core clock speeds may be lower.
In all the years I’ve been reviewing workstations I can’t remember a CPU ever impressing me as much as the 64-core Threadripper 3990X. It really is a phenomenal feat of engineering, giving the best of both worlds for single threaded CAD and multi-threaded ray trace rendering. If you use a design viz tool like V-Ray or KeyShot then it’s completely untouchable. Intel has nothing that gets remotely close.
But Intel Core or Intel Xeon aren’t the only competitors to AMD Threadripper. In the last couple of years, the GPU has also become a serious challenger for rendering. This is especially true for Nvidia RTX GPUs which feature dedicated cores for ray tracing and AI denoising, and more memory on the high-end Quadros. RTX is also supported by an increasing number of viz tools including V-Ray, KeyShot and Enscape.
GPU rendering has certainly been gaining traction, but now with AMD Threadripper delivering genuinely huge leaps in performance and offering quick access to up to 256GB memory, the battle is far from over.
Of course, a 64-core CPU isn’t for everyone. Designers, architects and engineers who only use CAD or BIM software will likely fare better on Intel, which still offers faster single threaded performance with an eight core CPU like the Core i9-9900K. But if you’re into ray tracing in any shape or form, any one of the 3rd Gen Threadrippers, including the 24-core 3960X and the 32-core 3970X, should serve you well. And Armari is proving to be one of the best to get the most out of this exciting new platform.
» AMD Ryzen Threadripper 3990X (2.9GHz, 4.3GHz Boost) (64 cores) CPU
» 64GB (4 x 16GB) Corsair Vengeance LPX DDR4-3600 C18 SDRAM
» ASRock TRX40 Taichi motherboard
» 2 x 1TB Corsair MP600 PCIe 4 M.2 NVMe SSD (RAID 0)
» AMD Radeon Pro W5700 GPU (8GB)
» Full Water cooled Loop system (FWL upgrade) – includes 1 Free service (Coolant change, O-Ring inspection)
» Microsoft Windows 10 Pro for Workstations 64-Bit
» Armari Magnetar S/M/R/X Series – 3 Year RTB workstation warranty
» £6,664 (Ex VAT)
Greg Corke test drives AMD’s new affordable $399 pro graphics card that delivers real time viz and VR capabilities for the price of a typical CAD-focused GPU
For years, the sub £400 graphics card has been a mainstay of the CAD and BIM workstation. Each new generation delivers more power, but because of the nature of most CAD and BIM software, better performance on paper might not necessarily translate to a better experience for the end user.
The fact is, the majority of 3D design software is very CPU-limited. In other words, 3D performance is bottlenecked by the CPU, not the GPU. So, even if you got the most powerful workstation graphics card in the world, your 3D model still wouldn’t move more smoothly. With top-end cards costing close to £5,000 you could end up wasting lots of money on graphics processing power you never use.
But workflows are changing. In architecture in particular, the use of real-time visualisation and Virtual Reality (VR) is on the rise. These applications are extremely GPU intensive and are not bottlenecked by the CPU in the same way that CAD software typically is. Tools like Enscape, Lumion, Twinmotion, Unity, Unreal Engine, IrisVR, VR4CAD, eDrawings Professional and more will happily devour all the graphics processing power you throw at them.
If you currently own a sub £400 professional graphics card you probably won’t be getting the performance you need for real-time viz, especially at 4K resolution. And you can forget about VR. To use these GPU hungry applications effectively you’d have had to invest in a more powerful and more expensive GPU. The ‘VR Ready’ Nvidia Quadro RTX 4000 and AMD Radeon Pro W5700 would certainly get you there but will set you back around £700 and typically take the cost of your workstation beyond £2,000.
Now with the release of the AMD Radeon Pro W5500 those with more limited budgets can also get on board. At $399, it might be priced like a professional GPU for CAD or BIM, but it’s significantly more powerful and can also be used for VR.
The Radeon Pro W5500 follows on from the Radeon Pro W5700 to become AMD’s second professional GPU based on its new 7nm ‘Navi’ RDNA architecture. The single slot graphics card features 8GB GDDR6 memory and four DisplayPort 1.4 outputs to drive up to four 4K displays.
The maximum power consumption is 125W and the board requires an external 6-pin connector, which is fed from the PSU. With this spec, it should be compatible with almost all desktop workstations, apart from the Small Form Factor (SFF) models, which tend to take low profile graphics cards.
The Radeon Pro W5500 is PCIe Gen 4 compatible, which offers double the bandwidth of PCIe Gen 3. The new interface standard is currently only available in workstations with 3rd Gen AMD Ryzen or AMD Ryzen Threadripper CPUs, but AMD’s graphics card will still work perfectly well in PCIe Gen 3-based Intel systems. This isn’t really a big deal for CAD or design viz centric workflows, however, as having more bandwidth is unlikely to make any real difference to performance.
In terms of positioning, the Radeon Pro W5500 sits below the double height 205W Radeon Pro W5700. We see it as the natural replacement for both the CAD-focused Radeon Pro WX 5100 (8GB) and the entry-level ‘VR Ready’ Radeon Pro WX 7100 (8GB), both of which are over three years old and are based on AMD’s 14nm Graphics Core Next (GCN) design.
In terms of price point, the natural competitor to the Radeon Pro W5500 is the Nvidia Quadro P2200. This CAD-focused GPU launched in Spring 2019 as an incremental update to 2017’s Quadro P2000. It has since become a mainstay in many CAD-focused workstations.
Both the P2000 and P2200 are based on Pascal, and not the newer Turing architecture which powers Nvidia’s higher-end Quadro RTX GPUs. Fitted with 5GB GDDR5X, they feature less memory but have a max power consumption of 75W, 50W less than the Radeon Pro W5500. However, AMD claims a workstation powered by the W5500 actually consumes less electricity at the plug, in certain CAD-centric workflows.
To test out the Radeon Pro W5500 GPU, UK firm Armari provided one of its excellent Magentar X Series workstations with the following specification:
AMD Ryzen Threadripper 3970X CPU (32-cores)
128GB of DDR4 memory
1TB Corsair MP600 PCIe 4 M.2 NVMe SSD
ASRock TRX40 Creator motherboard
Microsoft Windows 10 Pro
AMD Radeon Pro Driver version – Enterprise 20.Q1
Nvidia Quadro driver version – 442.19
When using CAD or BIM software, being able to move models quickly and accurately into position is important for creative flow. But with most CAD software being bottlenecked by the CPU and single threaded performance in CPUs advancing at a very slow rate, this is not always possible when working with large models. As a result, software developers have had to look for other ways to improve 3D performance.
The answer has been to temporarily simplify the model when it’s in motion, then fill in the details once it stops. Different software deals with this in different ways. In Solidworks, for example, detailed geometry is turned into simplified blocks. In Autodesk Revit, lines, shadows and small objects are removed. In Autodesk Inventor, you can set a minimum frame rate between 0 and 20 Frames Per Second (FPS) and in order to maintain it, smaller objects disappear.
To accurately test the relative performance of GPUs we disable all performance optimisations, so the model is always displayed in full detail. Observing how the Radeon Pro W5500 and other GPUs work with Revit and Inventor exposes just how CPU limited these applications are, with all GPUs giving near identical FPS scores when rotating a model in a consistent way using a 3Dconnexion SpaceMouse Pro.
In general, we find anything around 20 FPS and above to be adequate for 3D modelling. Even lower frames can be acceptable to some but going below 10 FPS is undesirable. Our Revit hotel model is relatively small and while the Mastenbroek heavy machinery model in Inventor is larger, architects and engineers are certainly working on much more complex models. And as model complexity increases, frame rates will go down, increasing the reliance on the software’s model simplification.
To illustrate just how CPU bottlenecked these applications are, we used the freely downloadable utilities GPU-Z, and CPU-Z. When rotating a model in Inventor and Revit, CPU utilisation hit 100%, but the Radeon Pro W5500 hardly got out of first gear, using only around 10-20% of its resources. In CAD/BIM applications like these, of which there are many, you could use a much lower specced GPU and still get the same 3D performance.
Solidworks used to suffer from similar problems in some viewing modes, but for the 2020 release it got a brand-new graphics engine that is much less CPU limited. In our siister publication DEVELOP3D we looked at the OpenGL 4.5-based graphics engine when it was still in beta in Solidworks 2019. You can read the in-depth article here.
In Solidworks 2020 we tested with a range of models and viewing modes – shaded with edges and with the more realistic RealView, with shadows and Ambient Occlusion enabled. We observed GPU utilisation to be significantly higher, often hitting 100%. It means users can get a real benefit from more powerful GPUs, as shown in our benchmark charts.
For our small to medium sized models, the Black Owl PC (295 components, 3.6M triangles) and production machine (591 components, 5.7M triangles) the Radeon Pro W5500 was well above the minimum ideal 20 FPS, even at 4K resolution. However, with our largest model, the colossal MaunaKea Spectroscopic Explorer telescope (8,000 components, 59M triangles), it fell below. If working with models of this size you’d probably want to boost performance by enabling Level of Detail (LoD). The Nvidia Quadro P2200 performed much better and only went below 20 FPS when RealView was enabled. It looks like AMD has some optimisation work to do here.
Autodesk Inventor, like many CAD applications, is very CPU limited, hence the very similar scores observed with all the GPUs we tested, in both shaded with edges mode (above) and realistic mode (below).
The Radeon Pro W5500 is slower than the Quadro P2200 but it easily delivers more than 20 FPS with this 3.6 million triangle Black Owl PC model, even at 4K and with RealView enabled (below)
The Radeon Pro W5500 struggles with this colossal 59 million triangle telescope model. The Quadro P2200 has a clear advantage but you need a Quadro RTX 4000 to hit 20 FPS+ when RealView is enabled.
Performance in real time viz
Performance in CAD and BIM software may have been mixed, but the Radeon Pro W5500 starts to show its true colours in real time viz. It doesn’t hit the heights of a design viz focused GPU like the Radeon Pro W5700 or Nvidia Quadro RTX 4000, but it does offer a significant benefit over the Nvidia Quadro P2200.
In the AEC-focused LumenRT it was between 25% to 31% faster. We experienced similar with our automotive test model in Autodesk VRED Professional, although the Quadro P2200 took a slight lead when anti-aliasing was enabled.
In Enscape, where we use a very large architectural scene of a museum and its surrounding area, the W5500’s lead got even bigger at 4K resolution. This probably isn’t down to the raw power of the GPU, however, rather available GPU memory. At 4K, the model needs around 7.5GB, which far exceeds the Quadro P2200’s 5GB.
For CAD, 5GB is usually plenty, but because real time visualisation uses more realistic materials and lighting it often needs more.
But what do these results show us? In general, the Radeon Pro W5500 looks well suited to real time work at FHD resolution but may struggle with some scenes at 4K and above. If you need to maintain frame rates at these higher resolutions, you’d really need to look at a more powerful GPU.
In the AEC focused LumenRT the Radeon Pro W5500 does well at FHD resolution (above), but slows down significantly at 4K (below)
The Enscape museum model is very complex but the Radeon Pro W5500 still delivers more than 20 FPS at FHD resolution
At 4K, this large Enscape model needs 7.5GB of GPU memory which explains why the 5GB Quadro P2000 and P2200 really stutter.
In automotive visualisation tool Autodesk VRED Professional at FHD resolution the Radeon Pro W5500 performed well, even with anti-aliasing enabled
Even at 4K the Radeon Pro W5500 did OK, but it slowed down considerably when AA was enabled, losing out to the Quadro P2200 (see below).
Performance in GPU rendering
In addition to real time 3D, the AMD Radeon Pro W5500 can also be used for ray trace rendering. However, its use is limited by application support.
As you’d expect, it works with applications that use AMD Radeon ProRender, either directly or via a plug-in. These include Autodesk 3ds Max, PTC Creo, Modo, Solidworks Visualize, Cinema 4D, Acca Software and others. It’s not compatible with Nvidia Iray or Nvidia RTX which is found in Luxion KeyShot 9, V-Ray NEXT, Unreal Engine, Enscape and many others.
Solidworks Visualize 2020 supports both Nvidia Iray and AMD Radeon ProRender — but it’s not possible to compare performance directly because there are visual differences between all render engines.
In the world of Radeon ProRender we were quite surprised by what we saw, with the Radeon Pro W5500 standing shoulder to shoulder with the more powerful Radeon Pro W5700. The GPU was also significantly faster than the Radeon Pro WX 5100 and WX 7100.
In GPU rendering, the Radeon Pro W5500 surprisingly stands shoulder to shoulder with the more powerful Radeon Pro W5700
Performance in VR
The Radeon Pro W5500 is the first sub $400 professional GPU that can be used for pro VR, but it’s really for entry-level VR workflows. A quick run through the VRMark benchmark shows it to be around 13% to 23% faster than the Radeon Pro WX 7100 but significantly slower than the Radeon Pro W5700 and Nvidia Quadro RTX 4000.
Of course, VR for design, architecture and engineering is a complex matter and performance is not only influenced by the application, but the size of the dataset, the complexity of the geometry (number of triangles), how well the geometry has been optimised, as well as the lighting and textures.
We tested it out on some real-world CAD data using an Oculus Rift. In Enscape, we were pleasantly surprised to get a generally good experience with our colossal museum model, although for it to be completely flicker free from all angles we needed to dial down quality to low. However, for the excellent Enscape application, low quality is still pretty good.
In Autodesk VRED Professional we fired up the automotive Genesis sample model (5.7 million triangles). It worked fine with AA set to off, but flickered as soon as Anti-Aliasing was enabled, even on low. In summary, the W5500 isn’t the kind of GPU you’d use for automotive styling, but it should do a pretty good job with less visually rich models, and in AEC workflows – unless you’re working with really complex datasets. Here,
An 8GB professional GPU for CAD, VR and real time viz, all for $399
In VR using Autodesk VRED Professional, the Radeon Pro W5500 could handle this 5.7million triangle Genesis model, but only with AA disabled
The Radeon Pro W5500 performed well in VR with this huge Enscape model that includes a detailed museum as well as several surrounding buildings.
it’s important to note that the Oculus Rift is an entry-level VR headset with a per eye resolution of 1,080 × 1,200. With more modern headsets, which have higher resolution displays, so more pixels to render, performance may go down. This includes the HTC Vive Pro (1,440 x 1,600) Oculus Rift S (1,280 × 1,440) and HP Reverb (2,160 x 2,160).
Finally, with the W5500, you don’t have to be tethered to your workstation to use VR. With AMD ReLive for VR, which comes with the Radeon Pro driver, users can get a professional wireless VR capability by streaming data from the GPU to an HTC Vive Focus Plus headset. We haven’t tested it out ourselves with the W5500, but AMD has had ReLive for VR working with Unreal Engine and Solidworks eDrawings Professional. You can read more about it on DEVELOP3D.com.
AMD has made a lot of noise about the multi-tasking capabilities of the Radeon Pro W5500 and how it compares favourably to the Quadro P2200. The Radeon Pro team specifically highlights how its new GPU can maintain 3D performance even when the CPU is being taxed heavily on a multithreaded task like ray trace rendering.
We put this to the test, rendering a scene in KeyShot using the CPU, while modelling in Solidworks at the same time. Using the Black Owl PC model, the Radeon Pro W5500 maintained a solid 80 FPS, virtually the same as it did when KeyShot was not in use. Meanwhile, the Quadro P2200 went down from 141 FPS to 18 FPS, a significant slowdown.
There are workarounds for this that effectively tune your workstation. Setting CPU affinity, for example, which binds a process to specific CPU cores, can help reduce the performance impact on the Quadro P2200. With our 32-core Threadripper 3970X workstation we set Solidworks and the Nvidia Quadro driver to run on cores 1-4 and KeyShot to run on cores 5-32 and got a fast 139 FPS out of the P2200.
Of course, with a 32-core CPU you have the luxury of losing a few cores and not impacting render times significantly. But in a standard quad core or six core workstation you’d miss out on a lot by going down this route.
The AMD Radeon Pro W5500 also performs well when handling multiple GPU tasks at the same time. When rendering a scene on the GPU using Solidworks Visualize we got 27 FPS out of the Black Owl PC model in Solidworks. With the Quadro P2200 it went down to an unusable 6 FPS.
For the Quadro P2200, the extent of the performance impact does depend on how demanding the 3D application is. As mentioned earlier, Solidworks 2020 uses 100% of the GPU. With Autodesk Inventor, which uses far less, the Nvidia Quadro P2200 went down from 41 FPS to an acceptable 24 FPS and in Revit from 19 FPS to 12 FPS.
It’s important to note here that with Nvidia’s Turing architecture, the foundation for Quadro RTX, Nvidia has improved the multitasking capabilities of its GPUs considerably and from past experiences we’ve found the Quadro RTX 4000 can handle concurrent GPU rendering and real time 3D tasks very well.
The Radeon Pro W5500 has a clear advantage over the Quadro P2200 when multi-tasking on the GPU or when the workstation’s CPU is being hammered on a multithreaded task like ray trace rendering.
One of the less publicised features of the AMD Radeon Pro W5500 is out of the box support for AMD Remote Workstation technology. As the name suggests, it allows users to access their physical workstation remotely, from ‘almost any device’ – PC, laptop or tablet. AMD says you get the same workstation experience you’d expect to get in the design office, although this will depend on the quality of your internet / network connection, both in terms of latency and bandwidth.
This feature should resonate with those that need to work from home on occasion (think flexible working, boiler service or sick child) or finish off a design in the evening. AMD doesn’t charge a licence fee for the technology, which works with Microsoft Remote Desktop Connection and Citrix Virtual Apps and Desktops.
We tested it out using Microsoft Remote Desktop Connection on our local network and it worked very well, even over wireless. Setup is remarkably easy. In Windows, simply ‘allow’ remote connections to the host workstation then, on the client device, use the Remote Desktop Connection app to connect remotely.
Remote Desktop automatically uses the Radeon Pro GPU on the host workstation, so you get full 3D acceleration in your CAD application. In our tests everything was so responsive it felt exactly as if we were working locally. Nvidia Quadro GPUs also work with Microsoft Remote Desktop Connection and we got a similarly good experience with the Quadro P2200.
On a local network, this feature could be great for getting 3D acceleration into a meeting room without having to physically move a workstation, but the real power of the technology comes when accessing workstations from different locations. We didn’t try this out as it’s a bit more complex, involving port forwarding on the router and connection over virtual private network (VPN), but this should be bread and butter work for most IT depts.
With the Radeon Pro W5500, AMD has filled a bit of a gap in the pro graphics market, offering a GPU that goes beyond CAD into real-time viz and VR at a very competitive price point. Previously, a professional VR Ready GPU would have set you back around £700 + VAT. The W5500 is half that price.
Product designers and engineers will certainly take note, but the new GPU looks to be a particular attractive proposition for architects who increasingly want to augment their 3D design process with rich interactive visualisation or immersive VR. Client presentations are one thing but there is huge potential to use VR during the design process, dipping in and out of the virtual world to get a much better sense of scale and proportion for your evolving building. Previously, this simply wasn’t possible with the typical workstation that sits on the desks of most architects.
The W5500 isn’t for everyone. Those who simply use CAD, especially CPU limited applications like Revit and Inventor, could certainly get away with a lower spec GPU. In addition, those who take their real-time viz or VR seriously would be much better suited with the significantly more powerful Radeon Pro W5700, Quadro RTX 4000 or even the Quadro RTX 5000. But if you want to dip your toes into viz or VR, or give your entire design team the capability to do so, then the $399 price tag is much easier to swallow.
Of course, we don’t expect Nvidia to stay quiet for long. Its current competitive card, the Quadro P2200, is based on its older Pascal architecture and there’s a huge performance gap between the CAD-centric GPU and the powerful Quadro RTX 4000, which is built on the more modern Turing architecture.
AMD has said the Radeon Pro W5500 will be available from Dell in in the first half of 2020. We’d be surprised if other workstation manufacturers didn’t follow suit.
Armari Magnetar R64ER-RS1280/LC review: AMD raises the bar yet again | IT PRO
by: James Morris
Just when you thought the pressure from AMD on Intel’s market dominance might be about to abate, yet more competition is building. The Zen 2 architecture is paying dividends across all levels of the processor business, from mobile to server. But it’s at the high end that Zen 2 particularly makes its mark. We’ve already taken a deep dive into what the latest generation of AMD EPYC has to offer for servers with its ability to deliver up to 64 cores per socket, but it has great potential for the workstation market too. Here, we take a first look at what 64 cores can do for content creation users with a system from British vendor Armari, catchily named the Magnetar R64ER-RS1280/LC.
Armari Magnetar R64ER-RS1280/LC review: Processor and memory
To show what the new EPYC really has to offer, our review sample arrived with the top-end AMD EPYC 7742. This is a processor aimed at a dual-socket configuration, and you would be more likely to purchase this system with the much cheaper 7702P aimed at single sockets only, since this is a single-socket workstation. There’s not a huge difference between the two, apart from the significant 2,000 cost reduction for this system with the 7702P instead of the 7742. The 7742 has a 2.25GHz base clock and 3.4GHz turbo mode, where the 7702P has a 2GHz base with 3.35GHz turbo mode.
These nuances are overshadowed by the fact that both CPUs offer a whopping 64 cores and 128 threads. Putting this in perspective, Intel’s current Xeon Scalable range tops out at 28 cores, and those cost twice as much as the EPYC 7742, or more than three times as much as the 7702P. Intel has announced a 56-core Xeon Platinum 9282, but that’s currently only available through select datacentre- oriented server OEMs rather than for workstations, and likely to be an order of magnitude more expensive still.
So the 64-core EPYC is an incredible achievement for AMD, and putting this even further in perspective is the power envelope it operates within. The 7742 has a 225W TDP, and the 7702P requires 200W. But the 56-core Intel Xeon Platinum 9282 has a 400W TDP whilst the fastest 28-core 8284 is 240W. On paper, therefore, the EPYC is incredibly power-efficient, so your electricity bill will be lower too.
However, things aren’t quite as simple as that. Although Intel processors have clearly defined frequencies that different numbers of cores can run at – two at the top frequency, then four at the next one down etc – AMD’s processors are much more complicated. The consumer-grade processors and Threadrippers offer Precision Boost. This clever system learns the capabilities of your motherboard power supply and CPU cooling to dynamically control the frequency of each core according to what the system can handle without throttling, so if you have really meaty water cooling you can get better performance.
The EPYC range doesn’t have Precision Boost in such an aggressive way, but it does have something called cTDP which acts in a similar fashion and has to be turned on in the BIOS. Because the AMD EPYC range is more concerned with stability, if you have meaty cooling you can enable a higher TDP and more cores will run at a higher clock frequency. For the 7742, the regular TDP is 225W, but the cTDP is 240W, which Armari has enabled here. The 7702P, however, only offers a 200W setting, so will generally run at a lower clock compared to the 7742.
Armari’s R80 chassis is designed to fit 240W-capable liquid cooling so that CPUs like AMD’s can get the best performance possible, and this comes in the form of an Enermax Liqtech unit. During testing, we saw all 128 threads hit 3.35-3.4GHz consistently, and this was still a surprisingly quiet system. Another advantage of the new 2nd-generation AMD EPYC is that it now officially supports 3,200MHz DDR4 memory. The architecture sticks with eight channels, but this is still two more than Intel’s Xeon Scalable’s six channels, although the recently announced Cascade Lake-AP version sports 12 channels. Armari has equipped the R64ER-RS1280/LC with eight 16GB modules of 3,200MHz DDR4 ECC Registered SDRAM. This takes up all the available slots, but 128GB of RAM isn’t going to need an upgrade anytime soon anyway.
Armari has played it safe with graphics acceleration. Since this system is more likely aimed at rendering than constant modelling, an NVIDIA Quadro RTX 4000 has been supplied rather than the higher-end 5000. The RTX 4000 takes up the usual slot for Quadro 4000-series GPUs as the affordable high-end choice.
The RTX version sports a commendable 2,304 CUDA cores running at a base 1,005MHz with a 1,545MHz boost. There’s 8GB of GDDR6 frame buffer on board, operating at an effective 13,000MHz, providing a hefty 416GB/sec of bandwidth. Despite the powerful GPU and fast memory, this is still a 160W TDP card, so is reasonably frugal on power consumption as well. It should still provide some very usable 3D modelling ability and a decent level of CUDA or OpenCL compute if needed.
Armari Magnetar R64ER-RS1280/LC review: Storage
Although this system has plenty of room for storage, of which more later, Armari has chosen to supply just a boot drive in the shape of a Samsung PM981 1TB NVMe M.2 SSD. One of the benefits of the new EPYC is that this system supports PCI Express 4.0, but the choice of NVMe storage with a PCI Express 4.0 interface is not yet that great, so Armari has opted for the Samsung due to known dependability. This drive still managed sequential reading at 3,269MB/sec in CrystalDiskMark 6, with sequential writing at 2,378MB/sec, so it’s hardly a slouch. However, we’ve seen PCI Express 4 NVMe SSDs achieving 50% faster throughput.
The R64ER-RS1280/LC is built around Armari’s custom-designed R80 workstation chassis, which is a very solid steel box that can also be configured as a rackmount, although you can’t use water cooling in rackmount configuration due to the location of the radiator and vents on the case. Befitting the serious focus of this system, a 1,280W modular power supply is integrated that can simply be slid out the back if it needs to be changed.
This chassis also has a couple of cold-swap 3.5in drive bays accessible from the rear, plus two 2.5in cold-swap bays at the front. By default, these are for SATA or SAS drives, but can optionally be reconfigured with NVMe U.2 backplanes. There aren’t any more bays inside the case, but with two PCIe Gen4-capable M.2 slots on the ASRock ROMED8-2T motherboard as well, you’ve got space for up to six drives, so plenty of room for workstation-level capacity and RAID 0 or 1 configurations if desired. Unsurprisingly, there’s no room for a 5.25in optical drive at all.
The front of the case sports two USB 3.1 Gen1 ports, with minijacks for headphone and microphone. The motherboard supplies a healthy selection of rear ports, including two 10Gbit LAN connections and another RJ45 aimed at IPMI management. There’s a VGA port plus two USB 3.1 Gen 1 and a single USB 3.1 Gen2 Type C. There’s even a COM port for hardware management systems that still use this interface.
The two obvious comparisons in our recent reviews are Armari’s own Magnetar X200 and the PC Specialist Axiom. However, the Intel Xeon W-3175X in the X200 appears to have been discontinued already, and the Axiom costs more than twice as much as this system. So this should put our performance results in some perspective.
In our Media Benchmarks, where the X200 managed 678, the R64ER-RS1280/LC achieved an overall score of 618, with an image result of 176, video encoding result of 581, and multi-tasking score of 790. Image editing is the weakest here, held back by the single-core clock speed that can’t keep up with processors that have fewer cores but run them a lot faster. However, considering that you can’t get the X200 anymore, the fact that the R64ER-RS1280/LC is second fastest to it is moot. The Axiom only managed 471 overall, with lower scores in every category, so for everyday tasks other than image editing, this is the fastest workstation you can currently buy that we’ve tested.
The results in the various flavours of Maxon Cinebench’s CPU benchmark really hammer this home. This is a test that can really benefit from lots of cores. The R64ER-RS1280/LC managed an incredible 8,857 in R15 and a huge 21,729 in R20. The X200 only managed 6,299 in R5 and 15,080 in R20, whilst the Axiom achieved 7,170 in R15. A workstation based around the 32-core AMD Ryzen Threadripper 2990WX (which uses the previous-generation architecture) only manages around 5,700 in R15. As you’d hope for a 64-core system, the EPYC 7742 totally canes multi-threaded activities.
However, this does come with a caveat that became evident when we tried some other CPU-focused tests. The Blender Gooseberry 3D render took 564.69 seconds, which is noticeably faster than a 32-core Ryzen Threadripper, but not by as much as you’d expect, and the X200 finished this render in 409.7 seconds. It’s clear that Blender isn’t giving the full benefit of 128 threads, and the Geekbench 4 scores of 4,398 single-threaded or 35,492 multi-threaded also illustrates a similar theme. This makes sense compared to the Axiom, which has a faster single-core clock but fewer cores, so has a better single-threaded result and slower multi-threaded. But the X200 managed 80,509 multi-threaded with only 56 threads, so this test clearly doesn’t get the full benefit of lots of cores.
Graphics aren’t the focus of this system, but you will want to know that it doesn’t lag in this department. The R64ER-RS1280/LC’s NVIDIA Quadro RTX 4000 managed a very respectable 178.69 in Maxon Cinebench R15’s viewport test, and SPEC Viewperf 13 results were also decent. You will see better scores than 197.73 in the 3dsmax-06 test from workstations with faster single-core frequencies, but this is still a great result, as is 260.06 in maya-05. Likewise, 244.86 in catia-05, 37.74 in creo-02 and 327.39 in snx-03 are good, but 133.98 in sw-04 is merely mediocre. Nevertheless, this system will be perfectly capable for 3D animation modelling, CAD and industrial design. The Quadro RTX 4000 also achieved 5,993 in Luxmark 3.1, so there’s a decent amount of CUDA or OpenCL-based GPU compute available too.
Armari Magnetar R64ER-RS1280/LC review: Verdict
Overall, it’s the multi-threaded performance that is the real benefit here, thanks to those 64 cores and 128 threads. However, you will need to ensure you use software that can take full advantage of them. If not, a workstation with higher single-core clocks, based around Intel’s recently-released Core i9 10980XE, the AMD Ryzen 3000 series or soon-to-arrive third-generation Threadripper might be more suitable. But that takes nothing away from what has been achieved here. AMD has packed a gobsmacking amount of compute power into a single socket, and for tasks that can make use of all the threads, this is an awesome amount of performance for the money.
The Armari Magnetar R64ER-RS1280/LC showcases just how much compute power AMD’s latest EPYC has to offer workstation users – if you have the right software
2.25GHz EPYC 7742
128GB 3,200MHz DDR4
8GB GDDR6 NVIDIA Quadro RTX 4000
1TB Samsung PM981 M.2 NVMe SSD
Windows 10 Professional 64-bit
3 years (1st year onsite, 2nd and 3rd year RTB parts and labour)
The phrase custom workstation manufacturer means different things to different people. There are many firms that simply take off-the-shelf PC components, throw in a professional graphics card and call it a workstation. But there are very few that take customisation to the level that Herts-based Armari does.
Armari is well known for designing its own custom chassis — the slimline VR-focused Magnetar V25 that we reviewed in 2017, for example, features a dedicated pocket for the HTC Vive link box. But you may not know that the company goes even deeper into customisation, making bespoke components for its workstations such as PCIe riser boards, heat sinks and PSU cable looms.
In its brand new Magnetar R80 workstation it is looking to get the most out of two power-hungry Xeon Scalable CPUs by letting them run at full frequency when all cores are being used. Liquid cooling is the obvious answer but while there are many off the shelf liquid coolers for consumer CPUs, when it comes to the new Xeon they simply don’t exist, as the volumes are not there. As a result, Armari is developing its own custom liquid cooling system so the CPUs can run at turbo speed on all cores, rather than just the usual one or two.
Unfortunately, the custom cooling solution is still being manufactured, so Armari sent in a pre-production machine with air cooling instead. This means that instead of the two Intel Xeon Gold 6154 processors running all of their 18 cores at 3.70GHz (which Armari says it will achieve once the custom cooling system is ready), clock speeds on this air-cooled system will be a touch slower. But not that much slower. On test, it managed to maintain around 3.5GHz even after we had been rendering a complex KeyShot scene for over an hour.
Rendering performance is virtually identical to the dual 32-core AMD Epyc-based Boston workstation. The Magnetar R80 completed the V-Ray benchmark in 21 seconds and our KeyShot test render in 94 seconds.
However, because it attains this level of performance with fewer CPU cores running at a higher frequency, it beats the Boston machine hands-down in a single threaded race. With a single core clocked at 3.70GHz, it was 32% faster in our SolidWorks IGES export test.
■ 2 x Intel Xeon Gold 6154 (3.0GHz, 3.7GHz Turbo) (18 cores, 36 threads)
■ 192GB (12 x 16GB) DDR4-2666 ECC Registered
■ AMD Radeon Pro WX 9100 (16GB) (17.Q4 driver)
■ 480GB Intel Optane 900P NVMe SSD + 4TB Western Digital Gold SATA-6Gb HDD
■ Supermicro X11DAi-N
■ Microsoft Windows 10 Pro
■ 1 year on-site maintenance contract (2 and 3 year options available)
CPU benchmarks (single threaded) – Seconds (smaller is better)
SolidWorks 2015 IGES export: 116
CPU benchmarks (multi-threaded) – Seconds (smaller is better)
GPU compute benchmark – Seconds (smaller is better)
V-Ray render benchmark (GPU): 21
3D graphics benchmarks (3D CAD) – Score (bigger is better)
SPECapc for SolidWorks 2015 (shaded + edges): 4.97
SPECapc for SolidWorks 2015 (RealView + shadows): 8.78
SPECapc for SolidWorks (RealView + shadows + AO): 25.42
SPECapc for PTC Creo 3.0 (shaded + edges): N/A
SPECapc for PTC Creo 3.0 (reflection): N/A
3D graphics benchmarks (design viz) – Frames Per Second (FPS) (bigger is better)
LumenRT (hotel model) (FPS): 14 (17.Q4 driver) 51 (17.Q4.1 driver)
LumenRT (roundabout model): 10 (17.Q4 driver) 26 (17.Q4.1 driver)
Autodesk LIVE (Villa Enhanced model): 71 (17.Q4 driver) 128 (17.Q4.1 driver)
Autodesk VRED Professional (AA off): 60
Autodesk VRED Professional (AA medium): 26
Autodesk VRED Professional (AA ultra high ): 11
3D graphics benchmarks (VR) – Frames Per Second (FPS) (bigger is better)
VR Mark (Orange): 138.16 (17.Q4 driver) 212.6 (17.Q4.1 driver)
VR Mark (Blue): 40.24 (17.Q4 driver) 41.34 (17.Q4.1 driver)
VR Mark (Cyan): 147.72 (17.Q4 driver) 148.39 (17.Q4.1 driver)
Notably, it was also only 17% slower than the Fujitsu Celsius W570power+, a typical CAD workstation with a quad core Intel Core i7-7700 CPU (3.6GHz up to 4.2GHz) and Nvidia Quadro P4000 GPU. Keeping clock speeds relatively high also benefits 3D graphics performance in CPU limited applications like SolidWorks. In shaded with edges mode, for example, the machine was only a touch slower than the Fujitsu and is almost certainly powerful enough to handle anything you can throw at it as far as 3D CAD is concerned.
In GPU hungry applications like Autodesk VRED Professional, however, it’s the high-end Vega-based AMD Radeon Pro WX 9100 that does all the heavy lifting, delivering solid scores on par with the Quadro P4000-based BOXX APEXX S3.
The combination of 3.70GHz CPU and Radeon Pro WX 9100 GPU also makes this machine capable of powering VR on the HTC Vive or Oculus Rift. In the VRMark benchmark it excelled in the DirectX 12-based Cyan test, delivering a score over and above the Nvidia Quadro P5000. This is good news for those using the Unreal or Unity game engines for viz or VR as these applications are likely to move to DirectX 12 soon. However, in the orange room it was slower than we expected, as it was in other DirectX 11 game engine applications — LumenRT and Autodesk Revit Live.
To explore if this was a driver issue, we changed the 17.Q4 driver for the 17.Q4.1 driver, which is the one recommended for the similar AMD Radeon Vega Frontier Edition. Performance immediately rose significantly in Autodesk Revit Live, LumenRT and VRMark orange room. However, as 17.Q4.1 is not the recommended driver for the Radeon Pro WX 9100 we would not advocate its use in a professional environment. It looks like AMD might need to do some driver tweaks here. Interestingly, we did not experience anywhere near the same kind of slow down when using the Radeon Pro WX 9100 with the 17.Q4 driver on an Intel Core i5-4690K-based workstation so it may be a chipset compatibility issue (the Intel C621 chipset is very new).
In VR itself, it performed well, working perfectly fine in Revit Live, an application the Radeon Pro WX 7100 struggles with. In Autodesk VRED Professional it could handle all of our test models, but with anti-aliasing enabled we only got a flicker free experience with smaller models.
The Radeon Pro WX 9100 is also well suited to GPU rendering, but it depends on the application. In the V-Ray benchmark, which uses an experimental OpenCL mode, performance was significantly slower than a Quadro P4000. However, in Radeon ProRender, a free GPU OpenCL renderer developed by AMD, it edges out a Quadro P5000.
To boost GPU rendering performance, you can add a double and a single slot GPU. In our test machine, however, it’s limited to a second WX 9100 or two WX 7100s, as the fourth slot is taken by a high performance 480GB Intel Optane 900P PCIe SSD. There’s also a 4TB Western Digital Gold Datacenter HDD for storing the bulk of your data. The test machine came configured with 192GB of memory, filling 12 out of the 16 DIMM slots with 16GB DDR4-2666 ECC modules to make the most of the Xeon’s six-channel memory architecture.
For a machine with such high computational potential, the chassis is surprisingly small. And despite having to cool two 200W CPUs and a 230W GPU (as well as memory and storage) fan noise is really not too distracting. We imagine it will get even quieter with liquid cooling.
The chassis has some nice touches, such as the long and slender rack server class 1,280W power supply, which pulls out from the rear of the machine, making it very easy to change in case of a failure. There are also two front-loading 2.5-inch drive bays that can be accessed easily.
Overall, the Magnetar R80 looks to be a phenomenal machine for digital artists, offering incredible performance for ray trace rendering without having to sacrifice performance in single threaded workflows. There’s also a powerful VR capable GPU. And while it looks like there is some work to do in DirectX performance, one must not forget that this is a pre-production machine.
Incredibly, there’s more to come. When Armari starts shipping the machine with liquid cooling (scheduled for Feb 2018) and all core speeds rise to 3.70GHz, it looks destined to set a new DEVELOP3D benchmark record for desktop rendering.
Request for us to contact you
Ask a question