By Dariush Derakhshani
My first video card review was on the ATI FireGL 8800 more than 14 years ago. It was one of the first video cards that could support two monitors with only one card, which to me was a revolution. Up until then I had to jam two 3DLabs Oxygen VX1 cards in my system (one AGP and the other PCI) and wrestle them to handle OpenGL with Maya 4.0 running on two screens. It was either that or sit in envy as my friends taunted me with their two screen setups, like waving a cupcake in front of a fat kid (me).
Needless to say, two cards were not ideal, and the 128MB ATI FireGL8800 was a huge shift in how I built my own systems from then on. Fourteen years later, I’m fatter, balder and have two 27-inch HP screens sitting on my desk (one at 4K) that are always hungry for new video cards. I run mulitple applications at once, and I demand to push around a lot of geometry as fast as possible. And now, I’m even rendering a fair amount on the GPU, so my video card is ever more the centerpiece of my home-built rigs.
So when I stopped by AMD’s booth at SIGGRAPH 2016 in Anaheim recently. I was quite interested in what AMD’s John Swinimer had to say about the announcements the company was making at the show. AMD acquired ATI in 2006.
First, I’m just going to jump right into what got me the most wide-eyed, and that is the announcement of the AMD Radeon Pro SSG. This professional card mates a 1TB SSD to the frame buffer of the video card, giving you a huge boost in how much the GPU system can load into memory. Keep in mind that professional card frame buffers range from about 4GB in entry level cards up to 24-32GB in super high-end cards, so 1TB is a huge number to be sure.
One of the things that slows down GPU rendering the most is having to flush and reload textures from its frame buffer, so the idea of having a 1TB frame buffer is intriguing, to say the least (i.e. a lot of drooling). In their press release, AMD mentions that “8K raw video timeline scrubbing was accelerated from 17 frames per second to a stunning 90+ frames per second” in the first demonstration of the Radeon Pro SSG.
Details are still forthcoming, but two PCIe 3.0 m.2 slots on the SSG card can get us up to 1TB of frame buffer. But the question is, how fast will it be? In traditional SSD drives, m.2 enjoys a large bandwidth advantage over regular SATA drives as long as it can access the PICe bus directly. Things are different if the SSG card is an island in and of itself, with the storage bandwidth contained on the card itself, so it’s unclear how the m.2 bus on the SSG card will do in communicating with the GPU directly. I tend to doubt we’ll see the same performance in bandwidth between GDDR5 memory and an on-board m.2 card, but only real-world testing will be able to suss that out.
But, I believe we’ll immediately see great speed improvements in GPU rendering of huge datasets since the SSG will circumvent the offloading and reloading times between the GPU and CPU memories, as well as potentially boosting multi-frame GPU rendering of CG scenes. But in cases where the graphics sub-system doesn’t need to load more than a dozen or so GBs of data, on board GDDR5 memory will certainly still have an edge in communication speed with the GPU.
So, needless to say, but I’m going to say it anyway, I am very much looking forward to slapping one of these into my rig to see GPU render times, as well as operability using large datasets in Maya and 3ds Max. And as long as the Radeon Pro SSG can avoid hitting up the CPU and main system memory, GPU render gains should be quite large on the whole.
Wait, There’s More
On to other AMD announcements at the show: The affordable Radeon Pro WX line-up (due in the fourth quarter of 2016), refreshing the FirePro branded line. The Radeon Pro WX cards are based on AMD’s RX consumer cards (like the RX 480), but with a higher-level professional driver support and certification with professional apps. The end-goal of professional work is stability as well as performance, and AMD promises a great dedicated support system around their Radeon Pro line to give us professionals the warm and fuzzies we always need over consumer level cards.
The top-of-the line Radeon Pro WX7100 features 256-bit 8GB memory and workstation class performance, but at less than $1,000, which I believe replaces the FirePro W8100. This puts the four-simultaneous-display-capable WX7100 in line to compete with the Nvidia Quadro M4000 card in pricing at least, if not in specs as well. But it’s hard to say where the WX7100 will sit with performance. I do hope it’s somewhere in-between the Quadro M4000 and the $1,800 M5000 card. It’s difficult to answer that based on paper specs, as the number of (OpenCL) Compute Units vs. the number of CUDA cores are hard to compare.
The 8GB Radeon Pro WX5100 and 4GB WX4100 round out the new announcements from SIGGRAPH 2016, putting them in line to compete somewhere between the 8GB Quadro M4000 and 4GB M2000 and K1200 cards in performance. Seems though that AMD’s top-of-the-line will still be the $3,400+ FirePro W9100 with 16GB of memory, though a 32GB version is also available.
I have always thought AMD brought a really good price-to-performance ratio, and it seems like the Radeon Pro WX line will continue that tradition, and I look forward to benchmarking these cards in real world CG use.
Dariush Derakhshani is a professor and VFX supervisor in the Los Angeles area and author of Maya and 3ds Max books and videos. He is bald and has flat feet.