Scan 3XS GWP-ME Q116C
Delays, chip shortages, and stuck on a 14nm manufacturing process. Should Intel be worried?
PRICE £2,500 | DEVELOPER Scan | WEBSITE www.scan.co.uk
With this workstation, we’re getting our first chance to test out AMD’S new RDNA graphics architecture in a professionalgrade Radeon Pro W5700. This is a mid-range card with 8GB of memory, comparable to Nvidia’s Quadro RTX 4000.
The entire specification is an example of what we think the average person would want from a rendering workstation without breaking the bank. The Radeon Pro W5700 is what we’re most interested in here, though. At the back AMD has generously included five Displayport connectors as well as USB-C, so potential display connectivity is superb, supporting five 4K displays at 60Hz. But disappointingly, unlike the small
single-slot
Quadro RTX 4000, AMD has had to go with a dual-slot approach, with a very chunky cooler taking up a whole slot.
It’s also a precursor to a major issue we had with it. The fan on the cooler in our review sample was poor, and when the GPU came under load, it emitted a whining sound that is simply not fun to work with.
That aside though, the Radeon Pro W5700 has indeed managed to just about bring AMD up to Nvidia’s currentgeneration performance levels, which is yet another fantastic achievement for the company.
The RDNA architecture does it thanks to the same major advantage AMD has with its processors – the use of a 7nm manufacturing process for superior efficiency while, like Intel, Nvidia is currently stuck on 14nm. RDNA uses less power than previous AMD cards, resulting in lower TDPS.
But this is a mid-range $799 card, so we’re looking at general efficiency and performance improvements in this price bracket, rather than any mind-blowing specifications. AMD’S previousgeneration, high-end and more expensive Vega 10 cards – Radeon Pro WX 8200 and WX 9100 – will comfortably outperform the W5700.
The devil is in the detail though. W5700 uses 8GB of cheaper GDDR6 rather than HBM2 memory in the Vega cards, with a narrower 256-bit memory bus but at a much higher 1,750 MHZ memory clock frequency, so the overall memory bandwidth is comparable to the high-end Vega cards.
If you’ve digested all that jargon, you’re probably interested to see how the Radeon Pro W5700 does in real tests. Blender is a good place to start, and in the BMW GPU test, the W5700 took four minutes, seven seconds while the RTX 4000 paired with the same processor took three minutes, 26 seconds.
A reasonable showing for the W5700 but in Specviewperf, it beat the RTX 4000 in a few tests, edging ahead slightly in Maya, but slightly behind in 3ds Max. And with the Opencl result, the W5700 and Quadro RTX 4000 were neck and neck, with only a 0.5% difference in scores.
So it’s a good showing and a great first effort at RDNA for AMD. That said, the small differences and extra noise with AMD means we’d opt for a mid-range workstation with an Nvidia card at this price point.
But given the close performance gap, we’re more excited about the launch of further next-generation professional graphics cards from AMD than we have been in years, considering they will use the same architecture but with beefier hardware.
VERDICT
14 years ago Intel launched its first ‘Conroe’ Core 2 Duo processor, a dual-core chip that blew away the performance of anything AMD had at the time. It then quickly followed up with the world’s first quad-core chip, which was an astonishingly powerful processor, and commentators asked whether AMD should be worried.
We are not quite seeing a direct reversal of that situation now. Today’s mainstream Intel line-up still has a (very slim) edge when it comes to single-threaded performance, with processors that run at higher boost clock frequency. The 8-core Core i9-9900k processor runs at up to 5GHZ for example.
Single-threaded performance matters a lot. In games, on the Windows desktop and when running any 3D software that depends on graphics card performance, faster single-threaded CPU performance squeezes more polygons and pixels out of the GPU. But this advantage is indeed shrinking with the improvements in each new generation of AMD processors.
This is evident from previous benchmark results over the last few months, where we can see that when paired with the same graphics card, the Core i9-9900k pushes ahead of the 16-core 4.7GHZ Ryzen 3950X in GPU tests, in Specviewperf, Blender and slightly ahead in the most detailed Hotel Luxmark Opencl test.
Rumours are pointing to some big changes to the specification of Intel’s forthcoming mainstream processors, Core-i5 with six CPU cores across the range, while the top-end Core i9 will have ten CPU cores and a whopping 5.3GHZ ‘velocity’ clock speed. We expect these processors will slightly extend Intel’s single-threaded performance advantage.
Currently Intel does have workstation-grade ‘Cascade Lake’ Xeon products that offer a generous number of processor cores that can match some of AMD’S chips. But the pricing of these products is far from generous, and AMD’S line-up offers more cores for less. For example, the 16-core Intel Xeon W-3245 retails for £1,900, while AMD’S 32-core Threadripper 3970X retails for £1,769. Intel’s 28-core W-3275M is listed at £7,086.
With that enormous price difference, for any kind of multi-threaded application, it’s currently a no-brainer to go with Threadripper, or the more affordable Ryzen 3950X processor over Intel. Not only do you get more cores for less money, but crucially the 3950X can be used in any mainstream AM4 motherboard, unlike both Xeon-w and Threadripper, which both require a more expensive platform.
Another major advantage AMD is enjoying right now is its support for PCIE 4.0, bringing greater bandwidth and transfer speeds to PCIE M.2 SSDS. This will certainly come to Intel’s platform, but it has not been confirmed yet whether we should expect it with the forthcoming 10thgeneration chips.
Intel may be ready to deploy a different weapon in its battle with AMD though, in the form of price cuts. Some chips have launched at half the price their predecessors did, but it’s not confirmed yet whether Intel will extend these reduced prices to its mainstream desktop line-up. Rumours and leaks suggest not, but anything is possible later on, especially if Intel loses market share to AMD this year.
“ANOTHER MAJOR ADVANTAGE IS AMD’S SUPPORT FOR PCIE 4.0, BRINGING GREATER BANDWIDTH AND TRANSFER SPEEDS”
Afew issues ago we crowned the 16-core AMD Ryzen 3950X paired with an Nvidia Quadro RTX graphics card as the current all-round best recommendation for a desktop rendering specification. And that recommendation has not changed, even after testing a workstation with a 32-core Threadripper processor, one with an Intel processor and both new AMD and Nvidia graphics cards.
Our advice is pretty simple. The 3950X is the best processor on the market today. For a mid-range rig, pair it with a Quadro RTX 4000. If you want more graphics performance, bump that up to an RTX 5000, with either 32GB or 64GB of memory, and the biggest SSD you can afford.
Anyone who really does work with the most high-end workloads is probably rushing to order a Threadripper-based workstation now, but we think there are only a slim amount of rendering tasks that would need a 32-core chip, when 16 cores isn’t enough.
Besides the obvious price increase, the other small differences with Threadripper such as the extra noise caused by the beefy cooling requirements and the physical chip size, make us consider that for all but the heaviest rendering workloads. For most of us, 16 cores is still a tremendous leap over quadcore or eight-core chips that the vast majority of mid-range workstations still run on. Performance will indeed be about 4x that of a quad-core chip, which means renders complete in 1/4 the time. That can shave hours off a final project. Things will be even faster with a 32-core chip of course, but if 16 cores can get a four-hour render to complete in one hour, is the extra noise and expense worth it to bring that down slightly further?
If you’re working on broadcast-quality animations that will be rendered in 4K or even higher, then both your personal rig and the render farm you use will certainly benefit from $2,000 32-core processors, and even more from the $4,000 64-core, 128-thread AMD Threadripper 3990X processor that has only recently been announced at the start of 2020.
To complicate things, with more than 16 cores it seems Windows and some 3D software do not quite behave themselves as expected. Our results found that in some tests, you don’t get a linear performance improvement by doubling core counts, and this problem may not go away any time soon. You’ll absolutely need to be using the latest versions of any software if you’re planning to upgrade.
AMD was recently asked whether a 128-core processor was in the works, and the response was that there are scaling efficiency issues at this level that may need to be addressed first.
For a more affordable Threadripper system than Scan’s high-end specification, it can be paired with a midrange graphics card, which may better suit your personal workload, and you could also opt for the
$1,299 24-core Threadripper 3960X. And following the news of the forthcoming 64-core chip, rumours point to a forthcoming 48-core Threadripper 3980X to fill in the gap.
In graphics, we are eagerly awaiting the launch of more RDNA cards, following some very impressive results from the Radeon Pro W5700. However we also know Nvidia is rumoured to be preparing its
Geforce RTX 3080 generation of GPUS, possibly based on 7nm, for launch by the summer.
It’s simply exciting to be able to even discuss this kind of rendering performance and know it’s not simply a pipe dream any more. With its new affordable massively multicore processors AMD has taken the entire computing industry forward, achieved world firsts with some of them, and subsequently, has moved rendering to a new era.
“IT’S EXCITING TO KNOW THIS KIND OF RENDERING PERFORMANCE IS NOT SIMPLY A PIPE DREAM”
Lisa Su’s leadership, new technology and fresh thinking have rebooted AMD’S fortunes