The 51%/98% GPU usage dilemma

[sm625]

I mean how else do you explain the success of the GTX 960 without accepting that it lived in the 980, Titan X and 980 ti's halo? The 280x was faster, the 380x is faster. I can buy the power savings argument for the 750 ti that doesn't need a 6 pin, but the 960 was a clear example of Nvidia's marketing winning when their technology lost.

I think its very simple. Never underestimate consumer stupidity. I can almost guarantee that 10% of the people browsing graphics cards, who know nothing about the brands, look at a name like Radeon and they think: "Cancer? Cancer! My grandma died of cancer.... dont look at this brand.... GTX? Now that sounds fast. And yet not dangerous. I'm going to buy that." I bet this is how people think. It doesnt even matter if they spent years with a radeon-powered game console in their home.

 

[BlitzWulf]

uh oh =(

https://www.reddit.com/r/Amd/comments/4m692q/concerning_the_aots_image_quality_controversy/d3sw31g

//EDIT: To clarify this, the scaling from 1->2 GPUs in the dual RX 480 test we assembled is 1.83x. The OP was looking only at the lowest draw call rates when asking about the 51%. The single batch GPU utilization is 51% (CPU-bound), medium is 71.9% utilization (less CPU-bound) and heavy batch utilization is 92.3% (not CPU-bound). All together for the entire test, there is 1.83X the performance of a single GPU in what users saw on YouTube. The mGPU subsystem of AOTS is very robust

 

[littleg]

Puts a single card at around 34fps or about the same ballpark as 390/390X.

 

[antihelten]

uh oh =(

https://www.reddit.com/r/Amd/comments/4m692q/concerning_the_aots_image_quality_controversy/d3sw31g

So that would put a single RX 480 at 34.2 FPS and thus 58.7% the performance of a 1080 (58.7 FPS) and by extension about 73% of a 1070 (a 1070 being about 80% the performance of a 1080).

 

[exar333]

uh oh =(

https://www.reddit.com/r/Amd/comments/4m692q/concerning_the_aots_image_quality_controversy/d3sw31g

So if you use the 62.6fps (2x GX480) compared to the 58.7fps (1080) and use the efficiency factor given of 1.83, that means the single GX480 is ~58.7% as fast as the 1080. If you reduced the 62.5 by 25% (to reflect the 1070) the comparison would be 77% of a 1070.

That is still pretty solid, and hopefully is representative rather than the 'best case'

Hopefully my math is right here...

 

[Noctifer616]

So that would put a single RX 480 at 34.2 FPS and thus 58.7% the performance of a 1080 (58.7 FPS) and by extension about 73% of a 1070 (a 1070 being about 80% the performance of a 1080).

Keep in mind that the 480 was CPU bottlenecked with single and medium batches so the game probably doesn't have any Polaris optimizations as was mentioned in the AMD event.

 

[antihelten]

Keep in mind that the 480 was CPU bottlenecked with single and medium batches so the game probably doesn't have any Polaris optimizations as was mentioned in the AMD event.

There's also the fact that the 1080 may have had a small advantage from not rendering some of the shaders properly.

So put all that together and a 480 is properly closer to 65% of a 1080 and 80% of a 1070, rather than 59% and 73% respectively.

 

[tential]

So if you use the 62.6fps (2x GX480) compared to the 58.7fps (1080) and use the efficiency factor given of 1.83, that means the single GX480 is ~58.7% as fast as the 1080. If you reduced the 62.5 by 25% (to reflect the 1070) the comparison would be 77% of a 1070.

That is still pretty solid, and hopefully is representative rather than the 'best case'

Hopefully my math is right here...

What about factoring in GPU utilization?
He says all together for the entire test they achieve a 1.83x scaling, but all together, what is the GPU utilization? At best case, it was 92%.

 

[Bacon1]

This is far beyond impossible.

AMD gets good scaling with Crossfire. Their slide that shows two 480s as slightly faster than a 1080 means one is a lot slower than a 1080, probably more like a 970/290.

I don't think even AMD would claim Polaris is 1070 level or close to it. They're pricing it at $199, would be at least $300 if it could trade blows with 1070s.

The cards you're talking about will be launched later this year.

If its CPU bound 100% scaling goes out the window. If the CPU can't push the cards to 100% usage then scaling drops fast.

They were only performing at 151% not closer to 200%.

That means that each card is pretty close to 1080 performance in DX12, about 75% each, meaning they are an amazing price/performance compared to the 1080.

 

[crisium]

GX480

For the first 10 times I thought it was a typo, but you are being deliberate.

What is a GX480?

 

[Hitman928]

If its CPU bound 100% scaling goes out the window. If the CPU can't push the cards to 100% usage then scaling drops fast.

They were only performing at 151% not closer to 200%.

That means that each card is pretty close to 1080 performance in DX12, about 75% each, meaning they are an amazing price/performance compared to the 1080.

It was clarified that the average scaling across the different batches (which was what the reported fps comes from) was 183%. The 151% was for the low draw call test only. So roughly a single Polaris card would be ~58% of the 1080. This would put it about on par with a 980 for this test. Again, just rough estimates, a lot of variables in play.