CPU DirectX 12 draw calls TestUpdate AnandTech Test

[csbin]

http://www.computerbase.de/2015-03/...amd-vor-nvidia/#diagramm-3dmark-amd-a10-7850k

AnandTech Test

http://www.anandtech.com/show/9112/exploring-dx12-3dmark-api-overhead-feature-test/4

 

[Erenhardt]

http://www.computerbase.de/2015-03/...amd-vor-nvidia/#diagramm-3dmark-amd-a10-7850k

Interesting. AMD cpus seem to give better results vs their intel counterparts in dx12 compared to dx11

 

[NTMBK]

Interesting. AMD cpus seem to give better results vs their intel counterparts in dx12 compared to dx11

It's only measuring draw call throughput, not total game engine performance- there are still plenty of other ways a game can be bottlenecked on single thread performance.

 

[TheELF]

Interesting. AMD cpus seem to give better results vs their intel counterparts in dx12 compared to dx11

Yep,amds also have better scores in cinebench...
You got to keep in mind that games can't use a 100% of a cpu only for draw calls ,they have to run the game so whats going to be left for draw calls will be no where near that actually.

 

[Erenhardt]

Im not saying amd will run better in games.

What I noticed is intel counterparts have higher Draw Calls in DX11 than AMD counterparts, but in DX12 the situation is opposite, and amd have more drawcalls than intel counterparts.

If it will transfer to gameplay fps is whole other matter and depends on whole game optimization, and not only rendering optimizations. But there should be some benefit.

 

[ShintaiDK]

Im not saying amd will run better in games.

What I noticed is intel counterparts have higher Draw Calls in DX11 than AMD counterparts, but in DX12 the situation is opposite, and amd have more drawcalls than intel counterparts.

If it will transfer to gameplay fps is whole other matter and depends on whole game optimization, and not only rendering optimizations. But there should be some benefit.

Look at the GTX980 with the APU and i3. Its quite clear that draw calls (by this test) doesnt matter in the big picture. Because the i3 runs in circles around the 7850K in any game with a GTX980. Despite that the 7850K got ~10% more draw calls in this test in DX11.

Not any different for the 8370 vs 4770. Night and day difference gaming wise, here pretty much even.

 

[frozentundra123456]

Where are you seeing that ? I only see a different effect on the performance of the IGP. AMDs igp gets a huge boost from mantle or DX12, while intel's gets a minimal one. We saw this earlier in another thread. Apparently there is some other limitation in Intel's igp that makes it unable to handle huge numbers of draw calls no matter what.

But with a discrete gpu, the relative increase is very close for AMD and Intel.

 

[ShintaiDK]

This thread should be merged here:
http://forums.anandtech.com/showthread.php?t=2425969

Instead of just constantly create more threads on the same topic.

 

[RoarTiger]

Odd to see the i3 integrated outperforming the 4770 in DX12.

 

[Yuriman]

Im not saying amd will run better in games.

What I noticed is intel counterparts have higher Draw Calls in DX11 than AMD counterparts, but in DX12 the situation is opposite, and amd have more drawcalls than intel counterparts.

If it will transfer to gameplay fps is whole other matter and depends on whole game optimization, and not only rendering optimizations. But there should be some benefit.

I'm seeing the FX-8 well ahead of an i3, but still well behind an i7. I suppose it's priced closer to an i3.

EDIT: Interestingly, with a GTX980, the i7 is behind the FX-8 in single-threaded draw calls in DX11 but pulls way ahead in DX12.

 

[ShintaiDK]

Odd to see the i3 integrated outperforming the 4770 in DX12.

Are you refering to that sub 0.1% variance when using IGP?

 

[RoarTiger]

Are you refering to that sub 0.1% variance when using IGP?

Yes, considering the cache differences. Please dont make up stuff when replying to my posts :"Sub .01% variance" . Turned a 10% lead in DX11 into a deficit. What else would you think I would be referring to?

 

[ShintaiDK]

Yes, considering the cache differences. Please dont make up stuff when replying to my posts :"Sub .01% variance" . Turned a 10% lead in DX11 into a deficit. What else would you think I would be referring to?

You considered IGP limitation and that there is a benchmark statistic variance?

 

[RoarTiger]

You considered IGP limitation and that there is a benchmark statistic variance?

Have you considered that is the only outlier in the entire result and questioned why? Because that was all I was doing. If you do not consider it siginifcant just say so. Quoting me while making up statistics isnt helping answer why the result looks out of place.

 

[frozentundra123456]

I'm seeing the FX-8 well ahead of an i3, but still well behind an i7. I suppose it's priced closer to an i3.

EDIT: Interestingly, with a GTX980, the i7 is behind the FX-8 in single-threaded draw calls in DX11 but pulls way ahead in DX12.

The i7 goes from about 10% slower in DX ST to 10% faster under DX12. Not sure how significant either of those are; margin of error of measurement could easily be 5% or more.

Again, I dont think this test will relate very much to real world gaming performance, but it does not lend any support to the theory of AMD fans that DX12 is suddenly going to make the FX8xxx the go to gaming processor. It does show, though, that unless Broadwell and Skylake have overcome the bottleneck in draw calls, whatever it is, Intel may have some serious modifications to do in their IGP, since DX 12 gives basically no boost over DX11.

 

[parvadomus]

The i7 goes from about 10% slower in DX ST to 10% faster under DX12. Not sure how significant either of those are; margin of error of measurement could easily be 5% or more.

Again, I dont think this test will relate very much to real world gaming performance, but it does not lend any support to the theory of AMD fans that DX12 is suddenly going to make the FX8xxx the go to gaming processor. It does show, though, that unless Broadwell and Skylake have overcome the bottleneck in draw calls, whatever it is, Intel may have some serious modifications to do in their IGP, since DX 12 gives basically no boost over DX11.

This however, does not tell otherwise. This is pure integer work, only draw calls that do nothing other than showing that AMD processors can submit work as fast as intel in a thread per thread basis, at least up to 6 cores.
This will for sure put A8/A10/FX6000 and FX8000 in way better light vs pentiums/i3s/i5 (if game engine is coded properly).
Btw this also shows that Nvidia gpu "command processor" is not better than AMD's, as some NVs fans claimed.
Looking at GPU part only, AMD cards perform much better with DX12 vs Nvidia than with DX11 vs Nvidia. AMDs DX11 dirvers are certainly damaging the performance of their cards.

 

[frozentundra123456]

The data does not show that at all. No data is shown at all for an i5. FX and i7 show very similar gains going from DX11 to DX12, as does the i3 and the A10. From the information in this thread, there is absolutely no way to justify the conclusion that AMD will benefit more from DX12 than intel.

I am talking about with a discrete card, btw. Obviously there is something in the Intel igp that does not allow it to take advantage of DX12, so AMD does show a much bigger gain there.

 

[III-V]

The data does not show that at all. No data is shown at all for an i5. FX and i7 show very similar gains going from DX11 to DX12, as does the i3 and the A10. From the information in this thread, there is absolutely no way to justify the conclusion that AMD will benefit more from DX12 than intel.

I am talking about with a discrete card, btw. Obviously there is something in the Intel igp that does not allow it to take advantage of DX12, so AMD does show a much bigger gain there.

I think the data's pretty clear. For once, AMD's higher core count is paying off -- DX12 simply takes better advantage of multiple threads, and that extra threading boosts the efficiency of their CMT, to the point that it actually outpaces HT. Historically, this has not been the case for gaming... those extra threads do next to nothing.

Also, seeing as AMD is coming from behind, they stand to benefit more than Intel, especially when you consider the diminishing returns of increasing CPU performance, since the GPU will become a bigger bottleneck.

It's a win for everybody, but it's a bigger win for AMD, because they it polishes one of their most tarnished spots on their CPUs. Too bad DX12 wasn't here prior to Bulldozer's debut.

 

[frozentundra123456]

I see what you are saying. What i am saying is that DX 12 gives basically the same benefit to AMD and Intel in this test. I stand by that because i calculated the ratios of DX12 to to DX11 and they are the same for both manufacturers. Now if that increase removes the cpu bottleneck for both, then the gpu becomes limiting and the benefit to amd is more in the gaming benchmarks. So in the broader scheme you could say the benefit is more to AMD as long as the game is cpu limited under DX11 but not under DX12. I also stand by my statement that Ernhart is incorrect in stating that AMD performs better than intel under DX12. It still performs worse, but it may not matter because the cpu is no longer the limiting factor.

 

[PPB]

I would regret leaving my 8320 behind, but then again I think the platform was really outdated and at the time of my new purchase the best Vishera out there for OC, the 8370E wasnt still being sold where I live. But it paints a fair picture of what most of us have being saying about the FX, it really got better as software caught up over time.

 

[escrow4]

How will DX 12 make someone buy an ageing CPU on an obsolete chipset that has barely been updated since 2009? 990/970 is basically a rebadged and tweaked 700 series. By the time DX 12 is in actual use Skylake and beyond will be out. You want to buy an FX CPU and associated chipset when its obliterated by Intel's offering? The cost savings are a myth, you are buying in an ancient CPU and associated platform, unless you get it dirt cheap.

 

[PPB]

No one is buying a FX deriberately for DX12. Most people are just stuck with AM3+ and are deciding wether to upgrade to intel or stay with what they have got.

 

[VirtualLarry]

No one is buying a FX deriberately for DX12. Most people are just stuck with AM3+ and are deciding wether to upgrade to intel or stay with what they have got.

Some of us still have AM3+ mobos with a Thuban in them. Like me.

Given the advent of DX12 and improved multi-threading in terms of draw calls and drivers, and hopefully the games themselves, would a Thuban @ 3.5 out-perform a G3258 @ 3.6?

Edit: I realize that an OCed Haswell i5 is probably better than both, but that's not the question.

 

[Yuriman]

Some of us still have AM3+ mobos with a Thuban in them. Like me.

Given the advent of DX12 and improved multi-threading in terms of draw calls and drivers, and hopefully the games themselves, would a Thuban @ 3.5 out-perform a G3258 @ 3.6?

Edit: I realize that an OCed Haswell i5 is probably better than both, but that's not the question.

I'm sure things will improve for Thuban but there will doubtless still be plenty of cases where the Pentium is faster. What's more important is that existing AM3 chips will have their useful life extended a bit.

 

[cbn]

http://www.anandtech.com/show/9112/exploring-dx12-3dmark-api-overhead-feature-test

Anandtech article up.

Interesting that Mantle scales better with hexcore than DX12. (hopefully Vulkan does as well)