AMD High Overhead Drivers on DX11

[LightningJR]

Your method is still flawed. You would be better off disabling all but 1 core of the CPU and then benching both GPUs. The benchmark should be held back by only having 1 core to use so the framerate will be the deciding factor. Maybe even underclocking the single core. You need the CPU to be 100% usage for the total of the run to get accurate results.

 

[R-T-B]

My main concern here is in establishing driver overhead though, not the fps impact per say... Someone else can figure that out, my brother only let me have his precious computer for a few minutes. *shrugs*

Will edit with results in a few minutes. Bear in mind this is a very primitive bench.

 

[ObscureAngelPT]

Heck good luck getting anywhere with nvidia too that cpu needs a good overclock period.

I am aware. I just say that with NVIDIA I can have an extra 10-20fps in CPU bottleneck situation.. It's the difference between having playing dying light at 18fps of a minimum and playing >30.
its a difference between having minimums of 40fps on saints row IV or 25fps.

after all we all want dx12 to take better advantage of the CPU. I feel it's not bad to ask the same on dx11

 

[Toothless]

I wish OneMoar didn't run off like usual.. I like listening to his insulting arguments.

This is a pretty interesting thread I do have to say. Makes me want to get my Athlon 620 back and test between my 5870 and 660.

 

[eidairaman1]

I am aware. I just say that with NVIDIA I can have an extra 10-20fps in CPU bottleneck situation.. It's the difference between having playing dying light at 18fps of a minimum and playing >30.
its a difference between having minimums of 40fps on saints row IV or 25fps.

after all we all want dx12 to take better advantage of the CPU. I feel it's not bad to ask the same on dx11

Cpu bottleneck is a cpu bottleneck youre not going to miraculously get more frames when a cpu is slow. It be like putting a gtx780 on an Athlon X4. Stop wasting our time and either get a Haswell i5 rig or overclock your cpu or grab a 1100T otherwise you're SOL.

 

[BiggieShady]

I don't see how VLIW has anything to do with CPU overhead.

I'm not saying it does, when I saw the thread I immediately remembered that in the days of 6970, AMD drivers were leaner for the cpu than nvidia's. I do think that investing in drivers optimizations has something to do with cpu overhead.

 

[LightningJR]

My main concern here is in establishing driver overhead though, not the fps impact per say... Someone else can figure that out, my brother only let me have his precious computer for a few minutes. *shrugs*

Will edit with results in a few minutes. Bear in mind this is a very primitive bench.

My method would give you your answer about driver overhead. The more FPS you achieve means there's less driver overhead. This is because the GPU will have more CPU cycles to use toward giving you better performance. The way you guys are doing it creates many more variables which will make you result less accurate.


edit: Ahh well if you have no time for it then that's all can be done I guess..

 

[MxPhenom 216]

and once again @OneMoar comes into a thread with his blatant arrogance.

I dont know why you bother posting anymore, nothing you post is helpful in anyway. This guy is just trying to share some things he has found.

 

[R-T-B]

Actually, my brothers PC is different in more ways than I remember. For example, different mobos too. Same ram but he has half the amount. Yeah it's not perfect but he's being a butt and screaming when I try to steal his GTX 780, lol. So this is again, to be taken with a grain of salt...

Complete specs will be provided with this bench regardless.

 

[ObscureAngelPT]

Cpu bottleneck is a cpu bottleneck youre not going to miraculously get more frames when a cpu is slow. It be like putting a gtx780 on an Athlon X4. Stop wasting our time and either get a Haswell i5 rig or overclock your cpu or grab a 1100T otherwise you're SOL.

so gaining 10-20fps in changing brand for me is gaining miraculousy hehehe.

and ita not wasting time its protesting for better optimization that would benefit anybody with and

 

[Ferrum Master]

Mantle just don't do it...

IMHO there are command sets that execute better or worse on each specific hardware, those benchmarks just poked on a slow AMD extension that needs few more cycles to execute, the pipe is too short, that's it. It is a hardware miss, not a driver. I would like to wote to do benchmark on OpenGL, and then see what's actually happening.

 

[Toothless]

Actually, my brothers PC is different in more ways than I remember. For example, different mobos too. Same ram but he has half the amount. Yeah it's not perfect but he's being a butt and screaming when I try to steal his GTX 780, lol. So this is again, to be taken with a grain of salt...

Complete specs will be provided with this bench regardless.

I love how you describe your brother. I'm the oldest of 7 so I can relate.

 

[R-T-B]

Conclusion: AMD is... more efficient?

This is a slightly flawed benchmark, for the reasons listed below for those interested. But it's still interesting.

Spoiler

Systems are not identical. NVIDIA one has 12GBs of ram and AMD one has 24GBs. AMD one has motherboard in my profile, NVIDIA one has a ASUS Sabertooth X58. Otherwise identical.

NVIDIA one appears to have more background process noise as well, though only slightly. this may put it at a slight disadvantage. There is also a slight clock skew due to how the boards handle "stock clocks"

Regardless of all that, here ya go:

Heaven Benchmark with task manager running in background. At the end, task manager screenshotted:

AMD R9 290X:

NVIDIA GTX 780:

Discuss.

 

[ObscureAngelPT]

Mantle just don't do it...

IMHO there are command sets that execute better or worse on each specific hardware, those benchmarks just poked on a slow AMD extension that needs few more cycles to execute, the pipe is too short, that's it. It is a hardware miss, not a driver. I would like to wote to do benchmark mantle helped me to get stable 60fps on bf4.on OpenGL, and then see what's actually happening.

Mantle just don't do it...

IMHO there are command sets that execute better or worse on each specific hardware, those benchmarks just poked on a slow AMD extension that needs few more cycles to execute, the pipe is too short, that's it. It is a hardware miss, not a driver. I would like to wote to do benchmark on OpenGL, and then see what's actually happening.

No

Battlefield 4 with mantle really improves my performance.
At the high settings in a 64 player map, i get bottleneck in DX11 with serious drops to 40fps depending on the sequence, and the GPU usage goes down to 60/70%.

With mantle i get 60FPS all the time and my GPU Usage is locked to 99-100%.
So the hardware is weak, but Mantle and DX12 will improve it, the same way that the DX11 of nvidia improved it.

 

[ObscureAngelPT]

Conclusion: AMD is... more efficient?

This is a slightly flawed benchmark, for the reasons listed below for those interested. But it's still interesting.

Spoiler

Systems are not identical. NVIDIA one has 12GBs of ram and AMD one has 24GBs. AMD one has motherboard in my profile, NVIDIA one has a ASUS Sabertooth X58. Otherwise identical.

NVIDIA one appears to have more background process noise as well, though only slightly. this may put it at a slight disadvantage. There is also a slight clock skew due to how the boards handle "stock clocks"

Regardless of all that, here ya go:

Heaven Benchmark with task manager running in background. At the end, task manager screenshotted:

AMD R9 290X:

NVIDIA GTX 780:

Discuss.

Can you try Star Swarm? and leave here the log files on Extreme and RTS mode?

 

[OneMoar]

and once again @OneMoar comes into a thread with his blatant arrogance.

I dont know why you bother posting anymore, nothing you post is helpful in anyway. This guy is just trying to share some things he has found.

once again you come in here and try and start something knowing full well I don't give a krud
bottom line is I am right and the op doesn't know what hes talking about
the 780 is showing higher cpu load because 290x is more powerful then the 780 and is shifting the load around while the cpu waits on the gpu
jesus people think with that giant hunk of gray squiggly junk inside your head
and post the frame-rates you got

 

[AphexDreamer]

I too fail to see how this is adequate proof to an issue?

The cards are no where near the same spec wise.
http://www.hwcompare.com/14638/geforce-gtx-770-vs-radeon-hd-7850/
The Memory Bandwidth is nearly two times faster on the 770.
143% increase in Texel rate.
22% increase in pixle rate.
Way more power draw.

Why are you surprised that it is handling rendering more object better than the low-mid end last gen AMD GPU?

If you want to test CPU bottle necking you should at least lower the resolutions when testing. If your frames stay the same then its CPU bottle necking if the frame rate improved your CPU isn't even bottle-necking.

 

[R-T-B]

I should have, and may later, but I am presently banned from his computer. If I touch it he says he'll feed me the contents of the cat box, so if I can pull it off it'll have to be later.

Also, it could be that the flaws in my benchmark are artificially inflating nvidias CPU usage, dunno. Someone else is welcome to follow up.

Oh, and OneMoar, just FYI, the GTX 780 has LESS fps and a worse score.

 

[Ferrum Master]

I guess I may bet that nvidia driver is better multi threaded not more overhead, that's the catch... on a powerfull clocked CPU the difference should be gone... on AMD... yes... they are kind of stuck...

Mantle is a different scenario don't touch it... if nvidia used it gained the same boost... So no proper judging.

 

[MxPhenom 216]

once again you come in here and try and start something knowing full well I don't give a krud
bottom line is I am right and the op doesn't know what hes talking about
the 780 is showing higher cpu load because 290x is more powerful then the 780 and is shifting the load around while the cpu waits on the gpu
jesus people think with that giant hunk of gray squiggly junk inside your head

I dont care that you don't care, I just enjoy calling you out on your crap, and many others seem fond of it as well.

 

[OneMoar]

Can you try Star Swarm? and leave here the log files on Extreme and RTS mode?

Star Swarm is CPU bound
it uses a metric ton of AI and physics calculations it is THE MOST CPU Demanding Game BAR NONE it is a terrible terrible benchmark of DX API Performace + it uses a absurd amount of draw calls that completely brake the direct3d11 renderpath

 

[Fluffmeister]

PCLab.pl did a good article testing how different CPU's impact the performance of the 290X and the GTX 970, they include Mantle too where they can, worth a read:

http://pclab.pl/art57717.html

 

[ObscureAngelPT]

startswarm is CPU bound jesus this is why I wasn't gonna waste time here
it uses a metric ton of AI and physics calculations it is THE MOST CPU Demanding Game BAR NONE

Exactly...
Thats the point.
If nvidia gets better performance on DX11 than AMD at DX11 that proofs the theory of nvidia having better drivers leaving the CPU more confortable.
Right?
Or am i wrong again?

 

[ObscureAngelPT]

PCLab.pl did a good article testing how different CPU's impact the performance of the 290X and the GTX 970, they include Mantle too where they can, worth a read:

http://pclab.pl/art57717.html

Thanks men..

AMD Gaming Evolve Game..

 

[OneMoar]

Exactly...
Thats the point.
If nvidia gets better performance on DX11 than AMD at DX11 that proofs the theory of nvidia having better drivers leaving the CPU more confortable.
Right?
Or am i wrong again?

you don't get it it runs like crap on AMD because the AMD cpu is woefully inadequate no optimization on earth is gonna change that not even mantle
the CPU is at fault For Being Slow nothing is gonna change that assuming you could completely re-write the DX11 API to handle drawcalls better which is a Microsoft imposed limitation NOT A AMD driver issue it still would't matter
the nvidia cards run better because the code is optimized better in other places such as the actual final render stage