d3d12 on apus.

[monstercameron]

source - http://www.anandtech.com/show/8998/directx-12-star-swarm-intel-igpu-performance-preview

The end result though is that one way or another Intel ends up shifting from being CPU limited to GPU limited under DX12. And with a weaker GPU than similar AMD parts, performance tops out much sooner. That said, it’s worth pointing out that we are looking at desktop parts here, where Intel goes heavy on the CPU and light on the GPU; in mobile parts where Intel’s CPU and GPU configurations are less lopsided, it’s likely that Intel would benefit more than they do on the desktop, though again probably not as much as AMD has.

interesting.

 

[myocardia]

I fail to see how moving from unplayable framerates to unplayable framerates makes any difference whatsoever. Add a $100 dedicated GPU, and you'll have 6-700% higher framerates, at the same resolution and settings.

 

[ShintaiDK]

Since Intel was never CPU limited besides in the submission time. There is no change as expected. The IGP is the bottleneck, not the CPU.

It seems Intel is struggling with the same multithreading issues as AMD in its driver.

 

[frozentundra123456]

Very strange results for intel. The batch submission times go down, but the framerate does not go up. You could say they are gpu limited, but if so, one would think the GT3e would be faster, but it really isnt.

 

[BSim500]

I fail to see how moving from unplayable framerates to unplayable framerates makes any difference whatsoever. Add a $100 dedicated GPU, and you'll have 6-700% higher framerates, at the same resolution and settings.

Precisely. And on "low" quality at that. Star Swarm is also not very representative of a real game either. Same "game" was used last February to show a +100-200% performance in Mantle vs DX11. Back in the real world, BF4 (like most other Mantle games) showed a more realistic 10-30%. DX12 will probably show the same range in most actual games (give or take a few percent).

 

[monstercameron]

Precisely. And on "low" quality at that. Star Swarm is also not very representative of a real game either. Same "game" was used last February to show a +100-200% performance in Mantle vs DX11. Back in the real world, BF4 (like most other Mantle games) showed a more realistic 10-30%. DX12 will probably show the same range in most actual games (give or take a few percent).

frostbite was originally a d3d11 engine, so there still stands the possibility that a d3d12 engine built from scratch could potentially see much larger gains.

 

[frozentundra123456]

Since Intel was never CPU limited besides in the submission time. There is no change as expected. The IGP is the bottleneck, not the CPU.

It seems Intel is struggling with the same multithreading issues as AMD in its driver.

Even then, as I said in my other post, the results dont make sense. Even the author's own conclusion that intel is gpu limited does not address the thing that I dont understand--- if the framerate is gpu limited under DX12, then should it not be faster with GT3e than with GT2? There seems to be some sort of scaling problem with intel such that neither decreasing the cpu load nor increasing the gpu power has any effect.

 

[sm625]

I fail to see how moving from unplayable framerates to unplayable framerates makes any difference whatsoever. Add a $100 dedicated GPU, and you'll have 6-700% higher framerates, at the same resolution and settings.

What $100 video card can deliver a 6-700% fps increase vs an A10-7800??

 

[Denithor]

I see DX12 as basically offloading some functions that were performed on CPU in DX11 onto the GPU which does a much more efficient job of handling these particular functions. This allows for overall much higher throughput as the CPU is now freed up from a bottleneck to push the GPU harder (and, therefore, why we see larger improvements at the top end of the GPU range - these are held back more by the CPU bottleneck).

What I want to see now, which has been completely ignored by AT, is a comparison of current AMD & Intel CPUs on these top GPUs. If the bottleneck is removed from the AMD lineup, they may become MUCH more competitive for gaming applications and could actually become viable again for high performance systems.

 

[PG]

Even then, as I said in my other post, the results dont make sense. Even the author's own conclusion that intel is gpu limited does not address the thing that I dont understand--- if the framerate is gpu limited under DX12, then should it not be faster with GT3e than with GT2? There seems to be some sort of scaling problem with intel such that neither decreasing the cpu load nor increasing the gpu power has any effect.

The article explains this in multiple places:

"The explanation, we believe, lies in the one part of an Intel GPU that doesn’t get duplicated in GT3e, which is the front-end, or as Intel calls it the Global Assets. Regardless of which GPU configuration we’re looking at – GT1, GT2, or GT3e – all Gen 7.5 configurations share what’s essentially the same front-end, which means front-end performance doesn’t scale up with the larger GPUs beyond any minor differences in GPU clockspeed."

"However in the case of Star Swarm the batch counts are so high that it appears GT2 and GT3e are bottlenecked by their GPU front-ends, and as a result the gains from enabling DX12 at very limited. "

 

[PG]

This is what I'm interested in: "...and with any luck we’ll find the final details on feature level 12_0 and whether any current GPUs are 12_0 compliant. "

 

[Bryf50]

I fail to see how moving from unplayable framerates to unplayable framerates makes any difference whatsoever. Add a $100 dedicated GPU, and you'll have 6-700% higher framerates, at the same resolution and settings.

Star Swarm isn't a game so there are no playable framerates.

 

[AnandThenMan]

I fail to see how moving from unplayable framerates to unplayable framerates makes any difference whatsoever.

You missed the point completely, look at the % improvement that could easily be the difference between unplayable and very playable :thumbsup:

 

[frozentundra123456]

The article explains this in multiple places:

"The explanation, we believe, lies in the one part of an Intel GPU that doesn’t get duplicated in GT3e, which is the front-end, or as Intel calls it the Global Assets. Regardless of which GPU configuration we’re looking at – GT1, GT2, or GT3e – all Gen 7.5 configurations share what’s essentially the same front-end, which means front-end performance doesn’t scale up with the larger GPUs beyond any minor differences in GPU clockspeed."

"However in the case of Star Swarm the batch counts are so high that it appears GT2 and GT3e are bottlenecked by their GPU front-ends, and as a result the gains from enabling DX12 at very limited. "

OK, thought I read the whole thing, somehow i missed that. Thanks.

 

[frozentundra123456]

You missed the point completely, look at the % improvement that could easily be the difference between unplayable and very playable :thumbsup:

Thats true, but otoh, this is a specially made demo to showcase Mantle/DX12, so the percentage increase may not carry over into a real game.