- Joined
- Mar 3, 2011
- Messages
- 125
- Motherboard
- ASUS H270 Plus
- CPU
- i5-7600K
- Graphics
- AMD RX 570 4 GB
Yep that is a real mystery - main gpu boost is only shown in GB OpenCL+Metal bench. Not shown in all other OpenCL+Metal benche / apps which have an bench points result, FPS counter etc.
My RX 460 runs about 30% faster!!! - but only in GB bench. Even other benches may get lower boost by little other gpu usage by their code, i would aspect to see at least 10% boost in Luxmark, N-Body Metal Bench,.... - but its 0%?!
An 30% Gpu boost by modded VBIOS (gpu,VRAM clk) or by power play table would be impossible, OC in this way its limited to 10-15% stable OC.
But in this OC case - higher gpu clk / better VRAM timings - will boost all gpu task (gpu compute & games & apps).
So i think the GB boost has to do with some GB code in their benchmarks - maybe they check something to optimise their code (OpenCL + Metal) by checking some gpu features or similar.
Like:
"If gpu.feature == something then
use code1 (optimised code)
else if
use code2 (not optimised code)
end if
So the gpu itself isn't really faster, the code which GB uses for the bench is by an other/optimised OpenCL/Metal program code part!?
My RX 460 runs about 30% faster!!! - but only in GB bench. Even other benches may get lower boost by little other gpu usage by their code, i would aspect to see at least 10% boost in Luxmark, N-Body Metal Bench,.... - but its 0%?!
An 30% Gpu boost by modded VBIOS (gpu,VRAM clk) or by power play table would be impossible, OC in this way its limited to 10-15% stable OC.
But in this OC case - higher gpu clk / better VRAM timings - will boost all gpu task (gpu compute & games & apps).
So i think the GB boost has to do with some GB code in their benchmarks - maybe they check something to optimise their code (OpenCL + Metal) by checking some gpu features or similar.
Like:
"If gpu.feature == something then
use code1 (optimised code)
else if
use code2 (not optimised code)
end if
So the gpu itself isn't really faster, the code which GB uses for the bench is by an other/optimised OpenCL/Metal program code part!?
Last edited: