During AMD’s Capsacian event we were all rather eager to see a plethora of demos, perhaps several slides explaining what AMD’s plans for their upcoming range of Radeon 400 GPU’s, or best of all some concrete specs. Unfortunately this wasn’t to be, and we were instead given a pretty short Hitman 2016 demo (running DirectX 12), but there has been some information which has slipped out.
On top of what AMD themselves mentioned during Capsacian regarding their goals of improving the “performance per dollar” of graphics cards, numerous journalists saw the PCB of the Polaris 10 and Polaris 11.
Hitman 2016 was running on a Polaris 10 engineering sample GPU, and according to AMD’s Raja Koduri ran at the highest settings in DirectX 12 mode, at a resolution of 1440P and maintained a constant 60FPS throughout. This impressive level of performance doesn’t necessarily mean the GPU was tapped out either; it’s pretty clear the game was set to not go above this frame rate, so just how much more was left in the Polaris 10 tank remains a mystery. This means that the GPU is at least as capable (if not more so) than the Radeon Fury X.
The connections of the Polaris 10 were a pretty standard affair, 3 DisplayPorts meeting the 1.3 standard, 1 HDMI 2.0 and a DVI Dual-Link – so five connections in total. The interesting part about the GPU was the size; reports have it that the GPU was tiny – about comparable to the PCB of the Fury Nano.
AMD’s Polaris 11 is the lower end chip (somewhat confusingly… as you’d assume 11 -> 10) and will likely find itself firmly at home in a lot of mobile gaming solutions. Astoundingly the Polaris 11 can run 4K VR content passively, with no fans whirring away to help cool down the graphics card.
What does all of this mean? Well, just how much of a leap in performance the Radeon 400 series will give over the current crop of graphics cards (either AMD’s or Nvidia’s) remains a bit of a mystery – and some folks are already touting the GPU’s could be more of a Fury replacement than anything. We know that AMD are only planning to release 2 Polaris cards in 2016 and next year will see the release of AMD’s Vega lineup of graphics cards; supposedly the first of their cards to use HBM2.
With AMD likely pushing HBM2 back until next year (thanks to larger 4GB HBM2 DRAM chips being available in Q3, and 8GB in Q4 2016) this means the GPU’s will be sticking with current technology of a GDDR5 of HBM1 variant. Interestingly enough; reports have popped up that Polaris 11 does indeed use GDDR5 (or possibly GDDR5X).
The current limiting factor with HBM1 is the 4GB limit (which is present on the Fury lineup of graphics cards), and not necessarily a bandwidth dilemma – and despite reports of Dual-Interposers from SK Hynix last year, nothing concrete has been revealed to know if this technology is viable to produce in either mass quantities (due to the complexity of the interposer) or too expensive.
With AMD placing so much focus on Performance Per Dollar, we could see the company shifting to a more value focus for this generation – but this doesn’t necessarily mean the GPU’s won’t be competitive to much of (if not all of) Nvidia’s Pascal range of cards.
Image credit: VR World