AMD: Speculation, Rumors, and Discussion (Archive)

Status
Not open for further replies.
Which is going to be ironic when the percentage of 480s actually ever hooked up to a VR headset will turn out to be the exact inverse of that 33/40 ratio.
I only counted 28 sentences. And one of them said "virtual reality", not "VR". I don't know if that qualifies.
 
  • Primitive Discard Accelerator ( conservative rasterization. )
  • Hardware Scheduler
  • Instruction Pre-Fetch
  • Improved Shader Efficiency
  • Memory Compression
What's the source of this list?

I am a bit disappointed if the "Primitive Discard Accelerator" turned out to be a marketing name for conservative rasterization. This naming doesn't make much sense. I was hoping that AMD takes a big leap forward in their primitive processing performance. Nvidia is much faster rejecting invisible triangles.

Hopefully "Memory Compression" means that writes from compute shaders also support delta compression. It would be awesome. I already have some ideas how to exploit that (in sparse data structures).

I actually had a conversation regarding to instruction pre-fetch just a few weeks ago in Twitter. I didn't find any public documents describing GCN (1.0-1.2) instruction pre-fetch. This is important to know if you want to build a "jump table" style shader system. Sort all shaders by GPR count and bucket them to GCN occupancy classes (http://amd-dev.wpengine.netdna-cdn.com/wordpress/media/2014/05/gcn_vgpr_table.png). This results in 10 different shaders in total, allowing you to execute any amount of different shaders in just 10 compute dispatches. Nice trick for tiled/clustered lighting for example (especially when combined with deferred texturing). Simple pre-fetch (load first N instructions of each shader) is not a good idea for shaders like this. Pre-fetch after the jump would of course work just fine.
 
Yeah that CR thing is new in regards to that list AFAIK. Also on that diagram with what's new in Polaris rasterization was not included as being updated.
 
...I was hoping that AMD takes a big leap forward in their primitive processing performance. Nvidia is much faster rejecting invisible triangles.
...
I actually had a conversation regarding to instruction pre-fetch just a few weeks ago in Twitter. I didn't find any public documents describing GCN (1.0-1.2) instruction pre-fetch. This is important to know if you want to build a "jump table" style shader system. Sort all shaders by GPR count and bucket them to GCN occupancy classes (http://amd-dev.wpengine.netdna-cdn.com/wordpress/media/2014/05/gcn_vgpr_table.png). This results in 10 different shaders in total, allowing you to execute any amount of different shaders in just 10 compute dispatches. Nice trick for tiled/clustered lighting for example (especially when combined with deferred texturing). Simple pre-fetch (load first N instructions of each shader) is not a good idea for shaders like this. Pre-fetch after the jump would of course work just fine.
http://people.engr.ncsu.edu/hzhou/ipdps14.pdf
If you haven't seen that it would be worth a read. It would seem to lay out a lot of possibilities that coincide with the features AMD listed. Shader Efficiency, Prefetch, and possibly the primitive discard. Primitive discard could be that flexible scalar re-grouping thread blocks or processing primitives directly. No idea if this is what they did, but they were at least looking into it and it does make sense.
 
£200ish for the 8Gb version sounds good. I tend to buy cards around that point every couple of years and that serves me nicely. Lower power use would be nice compared to nVidia's efforts, but it's still 1x 6 pin at least.

I'll be waiting to see what the 1060ti offers in comparison but looking forward to running everything I have now @1080 with silly settings.

(A HMD would be nice too, but not until they're much cheaper)
 
I get the feeling that the RX480 is the "380 Tonga" all over again?
Is the 480 really the full Polaris10? Or is it like the 380, a cut down version?

What do you guys think?
Maybe they keep the 480x to counter a 1060 Ti when it arrives?
That would make sense since Nvidia doesn't have anything to compete with in the mainstream market when RX480 launch. So why show your hand unless you have to?

A $299: RX480X, 8Gb 2560 sp, Gddr5x would make sense me thinks.
 
I only counted 28 sentences. And one of them said "virtual reality", not "VR". I don't know if that qualifies.
STRG+F finds 52 instances of VR being mentioned.

---
On the topic at hand:
5,5 TFLOPs (via leak, seems legit now) ist right between R9 290 and R9 390X, the former being the official VR starting point. Those reach 50 and 60 points respectively in our 2160p-index (the 970 is at 49 out of a 100 pts). Per theoretical TFLOPS, both cards get around 10 Index Points per TFLOPS. If we'd apply the same metric to RX 480, we'd see performance a in the middle between 970 and 980. If we apply Nvidia-level efficiency in terms of TFLOPs per actual performance (which for 980 and 1070 is between 13,5 and 14), we'd get to around (edit: actually a little above) 980-Ti-Level of performance. If they get only around 10% more gaming-performance out of each TFLOPS, that'd put RX480 right at 980-level.

Not helpful at all, since we're still guessing in the dark. :)
 
Last edited:
Exactly, we don't have any information about architectural developments, apart from that slide that showed which functional blocks had been updated without indicating how.
This gives us pretty much a lower bound, but no upper.
 
Good thing, though, is: They go down on the number of CUs and up on the clocks. And in the past, Radeon architecture scaled better with clocks than with CU count. So that's very good, especially since they stay below 150 Watt despite higher clocks. :)
 
Just got the press release, which has been posted before.
One thing struck me though - maybe someone from AMD monitoring this can clarify:
„The $199 starting SEP for select Radeon RX Series GPUs is an integral part of AMD’s strategy to dramatically accelerate VR adoption and unleash the VR software ecosystem.“
199 US-D for select RX series GPUs, no 480 explicitly mentioned. So, 199 still for RX 480? Or for lower models?

edit: nevermind, a little above that, the 4-GB-480 is explicitly mentioned:
"The Radeon RX 480 delivers premium VR capability at a stunning price of starting at just $199 for the 4GB edition." Sorry.
 
Last edited:
Anyone know when this is going to be available? Looking to replace my 660 soon...

June 29th according to AMD. I wonder if AMD will allow for reviews before the 29th.

I'm more interested in the Rx 460, but if the Rx 480 is good enough I might grab one just because it's so cheap.

Regards,
SB
 
Just got the press release, which has been posted before.
One thing struck me though - maybe someone from AMD monitoring this can clarify:
„The $199 starting SEP for select Radeon RX Series GPUs is an integral part of AMD’s strategy to dramatically accelerate VR adoption and unleash the VR software ecosystem.“
199 US-D for select RX series GPUs, no 480 explicitly mentioned. So, 199 still for RX 480? Or for lower models?

edit: nevermind, a little above that, the 4-GB-480 is explicitly mentioned:
"The Radeon RX 480 delivers premium VR capability at a stunning price of starting at just $199 for the 4GB edition." Sorry.

Yeah, as far as I can tell it's 199 USD for the 4 GB version and 229 USD for the 8 GB version.

Regards,
SB
 
We need to see how things pan in future dx 12 titles. The fury x is within 10% of 1080 gtx in some dx12 benchmarks and that's a last gen 28nm card.

The 1080 also does not sustain anywhere near 2Ghz in the founder's edition without tweaking hardware settings. Perhaps in other vendors versions.
Depends upon the reviewer and what tools if any used to analyse.
Also depends which Chapter of Hitman a publication uses, Chapter 2 works massively better for Nvidia and can put it a fair bit ahead; PCGameshardware are one of the few publications to benchmark both Chapter 1 and Chapter 2.

Back to DX12 in general and difference between FuryX and 1080, here is HardwareCanucks using Intel's PresentMon:
GTX-1080-REVIEWS-74.jpg


GTX-1080-REVIEWS-76.jpg


They do show Hitman being close between FuryX and 1080, but then as I mentioned one really needs to use Chapter 2 as 1 is really unoptimised for Nvidia.
Note that PresentMon will not give the same fps as the internal benchmark for AoTS.
Cheers
 
June 29th according to AMD. I wonder if AMD will allow for reviews before the 29th.

I'm more interested in the Rx 460, but if the Rx 480 is good enough I might grab one just because it's so cheap.

Regards,
SB

I'm expecting reviewers to get cards after E3... so about a week or so after that.
 
Status
Not open for further replies.
Back
Top