2016 is going to be yet another big year for GPU acceleration for Digital Content creation!. From render engines like Redshift and Octane, GPU powered viewports in Maya and MARI, and video acceleration across the majority of video applications, having a powerful GPU is becoming a necessity of you need to create pretty pictures.
But which GPU to get? Over the past couple of months we have been comparing two powerful workstation class GPU’s against each other in day to usage in a Windows 10 environment.
The cards are the NVIDIA Quadro M5000 (thanks to Escape Technology and NVIDIA for the use of the M5000) and my AMD W8100, both cards have 8gb of VRAM and both sit one stop under the ‘ultimate’ cards in their respective range.
|AMD W8100||NVIDIA QUADRO M5000|
|8 GB,GDDR5 memory||8 GB,GDDR5 memory|
|512-bit,memory interface||256-bit memory interface|
|320 GB/s memory bandwidth||211 GB/s,memory bandwidth|
|2,560 stream processors||2,048 CUDA Cores|
|Four discrete DisplayPort 1.2a outputs||Four,discrete DisplayPort 1.2 Multi-Stream,outputs|
|Maximum,DisplayPort 1.2a resolution 4096x2160||Maximum,DisplayPort 1.2 resolution 4096x2160|
While both cards can be used across OpenCL apps, the NVIDIA cards still use CUDA, NVIDIAs proprietary acceleration toolset. While most developers seem committed to using OpenCL, there is no getting over the fact that CUDA is still the dominant architecture, especially in GPU rendering.
So even though the AMD card is cheaper, the fact that it may not be able to run your creative software of choice needs to be measured against the potential lack of compatibility. Then again why invest in a workstation card at all, aren’t gaming cards cheaper?
Well yes and on spec's they may appear faster, but having used workstation cards for 6 months on Windows after a year of gaming cards and before that the closed environment of Mac OSX. Workstation cards on the PC do make everything run ‘better’, from the smoothness of the UI, the definite stability increases and the generally quieter operation and better support. To cut it short always go for the workstation card!
In fact I have found a good mix is to run a workstation card as my main driver and keep a couple of gaming cards in my PC for ‘dumb’ acceleration when more GPU’s are needed. This can work with either card, but on the whole it’s easier to manage with the NVIDIA Card.
Even though in terms of on paper spec's the AMD W8100 card may appear to have more grunt with 2,560 stream processors vs the Quadro’s M5000’s 2,048 CUDA cores, and 320GB/s of bandwidth vs 211GB/s, the AMD card is based on an older architecture than the Quadro and is also more of a power hog and require two 6-pin power connectors against the Quadro’s singular power port which is also a lot less power hungry so therefore can run in lesser spec’d systems.
In our tests and in general usage the NVIDIA makes up for the seeming lack of power with a newer architecture and generally wider adoption by developers. We ran benchmarks for Autodesk Maya, Adobe Premiere CC, MODO and MARI. I would have loved to have run benchmarks in NUKE but at the moment The Foundry does not support AMD cards on Window’s….did I mention that software compatibility for AMD was a problem…? :sigh:
In our Premier benchmark we took 6 streams of the same 8 second 4k footage from a GoPro ran a different Red Giant Universe sketch plugin on each stream and used Media Encoder to spit out a 1920 by 1080 clip using H264. This test saturated the GPU memory (we tried it on a technically more powerful NVIDIA Gaming card with 3gb of RAM, it fell over pretty much straight away…).
For Maya and MODO we used a 16 million polygon model and set it off on a 1000 frame tumbling animation and used FRAPS to get an average framerate over 60 seconds, for MARI we used the same model but applied a 16k oil texture in the spec and diffuse channels and then tumbled the UI, again we got an average with FRAPS. For Cinema 4D performance we used the ever reliable CInebench R15.
As we can see, other than in Cinebench, the NVIDIA Quadro M5000 decimated the AMD W8100. It is worth noting that all the newer GPU accelerated workflows such as Autodesk Maya's and MARI’s viewport and the Adobe Premiere CC plugins work far better on the newer architecture from NVIDIA. MODO's viewport is a bit more puzzling as far as we aware it isn't GPU accelerated unless you are using the new 'Advanced' mode, which we weren't.
The Foundry do seem to do a better job of supporting NVIDIA than AMD on Windows compared to the mac which does seem a bit bizarre, but even still, it is hard to recommend AMD for MODO users on Windows.
We ran Specviewperf 12 on both cards as well, again with the NVIDIA card outperforming the AMD.
The only issue we have with the NVIDIA is the price. However, with the extra level of performance you are getting, along with the better compatibility, lower power use and after a couple of months of use of each card, there is no getting around the fact that NVIDIA card has been by far the most stable of the two. Potentially the AMD card could be seen as a false economy.
The NVIDIA Quadro M5000 offers an excellent return on investment and more than holds it’s own for digital content creation. The only competition for the M5000 doesn't yet exist. With the announced new platforms for both AMD and NVIDIA, but both of those are at least 6-9 months away from the market in any form, and while developer’s promise OpenCL for many digital content applications, everything seems to work better if your prefer green!