Intel shows 3DMark API overhead test between DX12/DX11

[Faa]

So just looking at the start of the DX12 run 264 fps and 64 fps for DX11. That's quite a huge difference. Test was done with a broadwell i5.

3DMark API overhead is a cross-platform benchmark mainly tells how many drawcalls your CPU is capable of, let's say we are limited to 20K per frame with DirectX11, DX12 (going to a much lower-level & better multithreaded support) would allow to push a lot more draw calls out per frame -> more CPU performance. This test is only testing the CPU performance, so it's not testing how much GPU (read: GPU) performance DirectX12 adds.

Pcper will release soon an article about this, so stay tuned.

http://www.pcper.com/news/Graphics-Cards/GDC-15-Intel-shows-3DMark-API-Overhead-Test-Work


 

[cesrai]

Nice post.

Kappa

---

More information please.

---

Never mind checked the article it has like no information.

[Faa]

Nice post.

Kappa

---

More information please.

I can't provide you more information than the source provides. I assume you guys know this is basically testing CPU performance between two API's.

[cesrai]

I can't provide you more information than the source provides. I assume you guys know this is basically testing CPU performance between two API's.

Thanks, checked the article it has like no information.

[qwertywarrior]

a bench that focuses on draw calls ?

[cesrai]

a bench that focuses on draw calls ?

No information whatsoever.

[Thony]

Another test where results for DX12 multiply DX11 performance...

I will believe it when I PLAY on it (or when Linus does) !

[FakezZ]

Best case scenario for DX12, worst case scenario for DX11...

[Victorious Secret]

Another test where results for DX12 multiply DX11 performance...

I will believe it when I PLAY on it (or when Linus does) !

Well it makes sense since the tests are designed to saturate what DX11 can do and show you how much more headroom you have under DX12

[Sakkura]

Well, we've already seen benchmarks like this with Star Swarm, but another demonstration of why we need DirectX 12 / Vulkan / Mantle is fine by me. The Star Swarm results are equally extreme - an R9 290X jumping from 8.3 to 42.9 FPS stands out.

[Thony]

Well it makes sense since the tests are designed to saturate what DX11 can do and show you how much more headroom you have under DX12

I like te see the impressive performance leap but we need to see and test it ourselves in games

So far it looks like 120/144hz screens will be not just for privileged (top end gaming rigs) but needed for even casual gamers I want one myself so need to keep them frames above 90fps.

[ivan134]

I like te see the impressive performance leap but we need to see and test it ourselves in games

So far it looks like 120/144hz screens will be not just for privileged (top end gaming rigs) but needed for even casual gamers I want one myself so need to keep them frames above 90fps.

Does anyone still make 120Hz screens?

[Sakkura]

I like te see the impressive performance leap but we need to see and test it ourselves in games

So far it looks like 120/144hz screens will be not just for privileged (top end gaming rigs) but needed for even casual gamers I want one myself so need to keep them frames above 90fps.

 

Well, check the results with Mantle. DirectX 12 should yield very similar results.

 

And then imagine if AssCreedUnity had supported one of these APIs for all those draw calls...

[Humbug]

Another test where results for DX12 multiply DX11 performance...

I will believe it when I PLAY on it (or when Linus does) !

They are not claiming that gains will be this big in games.

These tests are designed to isolate one aspect of graphics performance- CPU limited draw call performance.

[Faa]

Well, check the results with Mantle. DirectX 12 should yield very similar results.

 

And then imagine if AssCreedUnity had supported one of these APIs for all those draw calls...

DX12 is ~ 50% faster than Mantle.

[Sakkura]

DX12 is ~ 50% faster than Mantle.

 

Anandtech shows Mantle being slightly faster than DirectX 12 in Star Swarm.

[Thony]

Does anyone still make 120Hz screens?

Don't know, theres very few. I would love to know where these refresh rates come from actually... Like why is it 144 and not 140 ?!

[Thony]

Well, check the results with Mantle. DirectX 12 should yield very similar results.

And then imagine if AssCreedUnity had supported one of these APIs for all those draw calls...

So in game performance will not be as significant as we see in these synthetic tests.

I cant imagine anything as I don't know how draw calls are related to performance and what is the benefit of more draw calls per second in simple English

Knowing Nvidias marketing, they will still put one of the graphs (edited) on the box

[Faa]

Anandtech shows Mantle being slightly faster than DirectX 12 in Star Swarm.

Nvidia's DX12 performs much better than AMD's DX12, AMD's DX11 is nowhere near close to what Nvidia's DX11 is. You're comparing AMD's DX12 with Mantle. 

Nvidia who literally took the most out of DX12, performs 50% better than Mantle.

[samcool55]

-snip-

Don't forget DX12 isn't finished yet. And AMD knows how low-level api works. They made mantle after all and that does work very good when you can use it. 

They probably learned a lot from it, so i don't think these results are amd's final answer already. Everyone has still a way to go before it's as good as they want.

[vm'N]

Nvidia's DX12 performs much better than AMD's DX12, AMD's DX11 is nowhere near close to what Nvidia's DX11 is. You're comparing AMD's DX12 with Mantle. 

Nvidia who literally took the most out of DX12, performs 50% better than Mantle.

Is this a joke?

That is not how you compare APIs.

You would obviously have to consider the same hardware configuration.

[Sakkura]

Nvidia's DX12 performs much better than AMD's DX12, AMD's DX11 is nowhere near close to what Nvidia's DX11 is. You're comparing AMD's DX12 with Mantle. 

Nvidia who literally took the most out of DX12, performs 50% better than Mantle.

 

You can't compare across GPU architectures like that. You're seeing the difference between GCN and Maxwell more than the difference between DX12 and Mantle. The whole point of testing different APIs in the same benchmark and on the same cards is to check how the APIs stack up in a fair, direct comparison. And it's proof that when it comes to draw calls, DX12 is (currently) essentially equivalent to Mantle.

[Blackie Sheen]

Does anyone still make 120Hz screens?

 

I dont think so...But there might be still some for retail.

[Master Disaster]

Impressive results so far, I'm in the need of a new GPU but will wait for the R9 300s to drop before doing anything.

I wanna see how Fuji & Fuji XT standup to Maxwell in DX12 & the new OpenGL standard (is it called Vulcan?).

Few questions though, why is Intel showing a benchmark for a new graphics API? What GPU was used in the test? (Was it an Intel HD?)

[Faa]

Is this a joke?

That is not how you compare APIs.

You would obviously have to consider the same hardware configuration.

What's your point? That we should compare a broken DX11 (which they kept on purpose broken) with an API AMD that's been purely all about boosting their GPU/CPU sales? If the 750 Ti is outperforming the 290x in DX11 by nearly 150% or the 980 outperforming the 290x by 50% (while the gpu performance difference is only around 5-10% more) then it would make zero sense to pull a conclusion out of a failure like DX12 is only a tiny bit faster than Mantle.

RTS mode is GPU bound for around 60 secs out of the 360 secs so a 980 would have a slight performance advantage over a 290x, for the rest of the run the 290x should perform the same as the 980 if we're cpu limited which is far from what's happening; so AMD failed. If you haven't noticed by now, Anandtech used the same CPU for each GPU.

 

 

You can't compare across GPU architectures like that. You're seeing the difference between GCN and Maxwell more than the difference between DX12 and Mantle. The whole point of testing different APIs in the same benchmark and on the same cards is to check how the APIs stack up in a fair, direct comparison. And it's proof that when it comes to draw calls, DX12 is (currently) essentially equivalent to Mantle.

You can't compare a broken DX11/DX12 with a proper working DX11/DX12.

It's CPU bound where GPU performance doesn't matter a thing as both GPU's especially in SW aren't even running on their 3D clocks with usages around 10%. Both cards should perform the same when the CPU is the bottleneck. Google up what kinda CPU overhead optimizations Nvidia has been doing for the last few years for their drivers, that's the reason AMD is getting outperformed.