While graphics cards with more than 8GB of memory might seem like overkill to gamers, those in the creative industries like VFX and 3D modelling can't get enough of the stuff. After all, VFX studios like MPC often create scenes that require upwards of 64GB per frame to render. The trouble is, even the most capacious graphics card—AMD's FirePro S9170 server GPU—tops out at 32GB GDDR5, and there are steep cost and design issues with adding more.
AMD has come up with another solution. Instead of adding more expensive graphics memory, why not let users add their own in the form of M.2 solid state storage? That's the pitch behind the all new Radeon Pro SSG (solid state graphics), which was revealed at the Siggraph computer graphics conference on Monday.
The Radeon Pro SSG features two PCIe 3.0 M.2 slots for adding up to 1TB of NAND flash, massively increasing the available frame buffer for high-end rendering work. The SSG will cost you, though: beta developer kits go on sale immediately for a cool $9999 (probably £8000+).
Details on the rest of the card are thin, but AMD has said the SSG is based on some form of Fiji chip, which is currently being used in its Fury range of graphics cards. AMD claims that the SSG has a dramatic effect on the rendering speed of high-end workloads. To demonstrate, the company showed a high-end system without SSG scrubbing through a RAW 8K video file at 17FPS. On the same system with the SSG it scrubbed through at a steady 30FPS. It even claims that a smooth 90FPS is possible.
Those using massive datasets such as in medical imaging and oil and gas exploration also stand to benefit from the huge frame buffer. While current graphics cards can offload to main system memory, this has a huge effect on performance as the GPU waits for the CPU to address requests followed by relatively slow data flows back to the GPU over the PCIe bus. Local storage, even in SSD form, should be much faster, although overall bandwidth is unlikely to be as good as native GDDR5.
In addition to the Radeon Pro SSG, AMD also unveiled a new line of workstation graphics cards based on Polaris dubbed the Radeon Pro WX Series. Aside from being decked out in a fetching blue, they're essentially rebrands of the company's RX 480, RX 470, and RX 460 consumer graphics cards, albeit with professional driver support. The top end Radeon Pro WX7100 features 32 compute units (CUs) delivering over five teraflops of single precision floating point performance. It includes 8GB of GPU memory on a 256-bit bus and will cost "less than $1000." (~£800).
The WX5100 includes 28 CUs for four teraflops of single precision performance, and will also feature 8GB of memory on a 256-bit bus. Last but not least, the WX4100 is a half-height GPU with 16 CUs for two teraflops of single precision floating point performance, and 4GB of memory on 128-bit bus.
The Radeon Pro SSG will only be available in a reference design, which the company says is a move to control quality—a wise move given the included 10-year warranty.
The last few months have been an incredibly busy time for graphics cards. AMD released its affordable RX 480 earlier this month to rave reviews, while Nvidia has released the GTX 1080, GTX 1070, and GTX 1060 to similarly high praise. Just last week, Nvidia unveiled yet another graphics card, a new version of the Titan X. It features the company's latest Pascal architecture, sports 12GB of GDDR5X memory, and is claimed to be around 24 percent faster than the GTX 1080, and 60 percent faster than the old Titan X. It goes on sale August 2 for $1200 (or probably about £1,100).