AMD Vega 10, Vega 11, Vega 12 and Vega 20 Rumors and Discussion

Really? Check the Witcher benchmark. Framerate of R9 390X drops to 64 % (48,0 -> 30,8 FPS) when you enable tessellation (HairWorks). RX 480 drops to 77 % (43,4 -> 33,4 FPS). Feel free to show us, how would you compensate this by clock normalization.
 
Really? Check the Witcher benchmark. Framerate of R9 390X drops to 64 % (48,0 -> 30,8 FPS) when you enable tessellation (HairWorks). RX 480 drops to 77 % (43,4 -> 33,4 FPS). Feel free to show us, how would you compensate this by clock normalization.

Wow, Hairworks ,.,,, ..
 
Last edited:
AMD RX 500 series Launch April 4th with three rebrands
Starting April 4 you sould see the Radeon RX 570 and RX 580, followed by Radeon RX 550 en RX 560 on April 11th.

The Radeon RX 580 would be a faster version of the RX 480 with a bump in clock frequency towards 1340 MHz. The 570 would be a 38 MHz faster model based on the RX 470. Both models will again be offered in a 4GB and 8 GB version.

The RX 560 is a 460 with 1024 stream processors abd a clock frequency of 1287 MHz. This would make the 560 substantially faster compared to the 460. The RX 550 would be Polaris 12 based, a new low-end SKU. Details on that one are missing.
http://www.guru3d.com/news-story/amd-rx-500-series-launch-april-4th-with-three-rebrands.html
 
So that 580 result was due to a difference game version, otherwise just 40-80Mhz increaes won't net those large gains.

Maybe they also added some of that mysterious 8Gb 9Gbps GDDR5 that is going into the 1060?

1060-overclock.jpg


Also, anyone want to bet that AMD added an 8-pin (or 2x6-pin) power setup to the reference 580?
 
Maybe they also added some of that mysterious 8Gb 9Gbps GDDR5 that is going into the 1060?

1060-overclock.jpg


Also, anyone want to bet that AMD added an 8-pin (or 2x6-pin) power setup to the reference 580?


I would not bet against that, that is pretty much a given, and a new power layout on their board too.
 
Really? Check the Witcher benchmark. Framerate of R9 390X drops to 64 % (48,0 -> 30,8 FPS) when you enable tessellation (HairWorks). RX 480 drops to 77 % (43,4 -> 33,4 FPS). Feel free to show us, how would you compensate this by clock normalization.

That is not how you deal with bottlenecks. 390 can have a higher drop because without tesselation it can put its wider parts into use. 480 leading by 10 % when having 20 % higher clock is not a proof of architectural improvement.
 
The same benchmark proves you wrong:
1. R9 380X is definetely narrower than RX 480 and its tessellation performance drop is higher than RX 480's.
2. RX 480 with tessellation enabled is faster than R9 390X.
Your logic doesn't explain it, in fact it contradicts these results.
 
The same benchmark proves you wrong:
1. R9 380X is definetely narrower than RX 480 and its tessellation performance drop is higher than RX 480's.
2. RX 480 with tessellation enabled is faster than R9 390X.
Your logic doesn't explain it, in fact it contradicts these results.

1. I agree, but for different reasons - it drops more than 390X, per clock by 3 %. I am still hesitating to draw conclusions from such a small difference in one game.
2. Already explained above. Not logically, because logic deals with necessary conclusions.
 
A little ot but are amd cards still unable to run the nvidia geforce 2 lightning demo
http://www.nvidia.co.uk/object/demo_lightning.html (3mb)

If not anyone know why ?
That's an old OpenGL demo using NV register combiner extensions. Obviously that can't run on non NV hardware (as the required extensions are not supported). Does not mean AMD could not run equivalent code with standard OpenGL functionality.
 
Thanks, I didn't know anything about register combiner extensions
I assume they are not part of the opengl spec and the amd equivalent is
GL_ATI_fragment_shader
at least nvidia havnt dropped support for them
 
Last edited:
Thanks, I didn't know anything about register combiner extensions
I assume they are not part of the opengl spec and the amd equivalent is
GL_ATI_fragment_shader
at least nvidia havnt dropped support for them
ATI_fs was for r200 generation, so it's quite a bit more powerful than NV_register_combiners. Corresponds more to NV_texture_shader. NV_register_combiners can't do dependent texturing for instance.
For r100 class hardware, ATI had ATI_texture_env_combine3 plus ATI_envmap_bumpmap IIRC. Most of the functionality the hardware could do was exposed by (new by then) standard GL functionality such as ARB_texture_env_dot3, ARB_texture_env_crossbar. register combiners may be a bit more powerful though, not sure (though the GeForce 2 doesn't support NV_register_combiners2, just the first version).
Anyway, certainly everybody could implement this stuff (nowadays drivers will just straight convert that to shaders just like they do with "standard" GL fixed function), but IHVs typically don't (or might not even be allowed to) implement proprietary extensions from other vendors. But vendors usually don't drop support for old crappy extensions neither due to backward compatibility. (FWIW with a mesa based driver you get support for ATI_fragment_shader for all cards but not NV_register_combiners.)
 
That is not how you deal with bottlenecks. 390 can have a higher drop because without tesselation it can put its wider parts into use. 480 leading by 10 % when having 20 % higher clock is not a proof of architectural improvement.

Try this:
http://www.pcgameshardware.de/AMD-R...97/Tests/RX-480-Test-1199839/galerie/2598540/

At 64× tessellation the RX 480 is 34.7 % faster than the Fury X even, despite a clock advantage of only 20.6 %. That's ~12% more performance per clock.
 
Try this:
http://www.pcgameshardware.de/AMD-R...97/Tests/RX-480-Test-1199839/galerie/2598540/

At 64× tessellation the RX 480 is 34.7 % faster than the Fury X even, despite a clock advantage of only 20.6 %. That's ~12% more performance per clock.


Now I am sold.
However, these gains are still underwhelming compared to what Polaris promised. Would it be possible to run the test with various MSAA modes to determine whether it comes from primitive discard or other improvements?
 
Now I am sold.
However, these gains are still underwhelming compared to what Polaris promised. Would it be possible to run the test with various MSAA modes to determine whether it comes from primitive discard or other improvements?
Did you notice that this performance chart is logarithmic. Polaris fps is around 40% better than Fury X at maximum tessellation factor. That's a pretty good improvement already.

Polaris is also likely bottlenecked by other areas. It's after all way behind Fury X in bandwidth and compute performance. Could be also rasterizer bound as highly tessellated geometry increases quad overdraw factor. We need to wait for Vega to see the real impact of these geometry pipeline improvements. Vega is the first high end product after Polaris geometry pipeline improvements and Vega also has much improved rasterizer.
 
Last edited:
Back
Top