High DX11 CPU overhead, very low performance.

hi expert i have a little problem in gta5 but it more a technical issues
a lot of stuttering with 3 go of ram that the main reason of the problem
but the stutter occur in some places only and when open task manager reason stutter is hdd write is working not read
so maybe its a memory flush or cash or something so any fix until i upgrade my ram
cpu i7 920 r7 260x win 8.1 64bit catalyst 14.4 15.018 15.223
btw i know my ram in too low minimal requirement is 4g
so plz give me a fix for troubleshooting only
thank you

 

the same problem in battlefield 4
youtube.com/watch?v=x6O9ZJHYoRk

 

Are you sure that both cards have the exact same settings? I didn't see the settings for the AMD card on the right.

Also, do you play on Mantle or DX11? Have you tried switching between them?

 

Yes

You can see at 2:03.

Performance when used Mantle is almost the same both at NV + DX11 plus 3-4fps.

 

One last request, could you make the video clearer somehow? I can't seem to read the upper part of it. The case you present seems pretty solid actually, and since we're talking about multiplayer it has a lot more implications.

Are you absolutely sure that your power profiles/cpu parking are the same between the tests?

 

Try to disable msi\riva monitor and use only fraps for fps, from TechnoKitchen channel on youtube, i know that it can do noticeable slow down to r9 280x on some machines, maybe it's your problem who knows.(they test GTA 5)
1)test with monitor LINK
2)test without monitor LINK

 

No he does not have same settings. And even if he has same settings in menus, back-end is totally different. That is why nV shows at bottom DX 11.0 and AMD shows DX 11.1.

And He should use Mantle anyway unless it gives some problems which it does not in BF4.

 

I installed Win 10 to test out difference between win 8 driver and win 10 this what I got. Win 10 pre did run wonky as hell thou CCC.exe crashed after 3rd boot and pc was slow to boot, crossfire only worked in mantle..(could be ~10y old hdd fault too ).

 

This is the comparison of which driver to which driver exactly?

 

Driver version
14.502.1014.0 win 8.1
15.200.1023.0 win 10

 

Hi all,
Great thread that dissapoints , but at the same time gives hope as we clearly see there is a lot of space for improvement on our gpus.
Hopefully AMD will do their best to be quick about it, as some might turn green with envy
Unfortunately i cant post images because of my low post count, but here are my tests with 14.502 and the 15.200.
Done them on Win 8.1 @ 1080p i5 4440S CPU:

R9 290 1000/1300Mhz
14.502.1014.0
DX 11 ST - 635,020
DX 11 MT - 602,135
Mantle - 11,453,173

15.200.1018.1
DX 11 ST - 808,875
DX 11 MT - 821,482
Mantle - 10,398,516


R9 280x 1020/1500Mhz (only the modded driver).
15.200.1018.1
DX 11 ST - 852,200
DX 11 MT - 710,474
Mantle - 10,538,673

As you see Mantle score goes down with 15.200 on the 290. Its also kind of funny that 280x would get better result in ST and Mantle than 290.
What gives, is it in the nature of the problem?
I have retested the R9 290 3 times and that's the best results out of three.

P.S.
Can anyone tell me where can i get the .1023 drivers, and do they give even better results?

 

Maybe the ones that are getting better results with the 280x have better CPU's (I haven't scrolled back the pages and checked, but that's probably why).

As for the 1023 Drivers, yes you should try them yourself and let us know what you think (test with games you usually play, not just with benchmarks) I've been using 1023 for 2 days now.

http://forums.guru3d.com/showthread.php?t=399099

 

I dont know about the others, but those results are from my machine. Same cpu, same 15.200 drivers, and 280x gets better score in DX ST and MANTLE than the 290?!? Thats what strikes me as strange...

 

Oh sorry, I thought you were comparing it to other people that own a 280x... Yes, that is strange. :3eyes:

 

Definitely needs some looking into.
I own 7950, 280x and 290 so i can do some testing on same CPU.
Game benchmarks i wouldnt have the time to do, unless there is some software suite that you guys would prefer me do the tests on.
In the next few days when i have some more time i will do the 3dMark feature tests on the same CPU, comparing 7950/280x/290.

 

Maybe they have to nerf mantle to make dx11 working. This just shows that draw calls has nothing to do with performance.

 

I don't know if it has been posted before. But look at this

https://www.youtube.com/watch?v=CSCSYVEA0qc

 

You don't see frametimes. There are two measures in performance. Frame rate, and frame delivery. If you have a higher frame rate but uneven frame delivery, you get a card that benches well but plays horribly. On the other hand, you can't neglect frame rate (for obvious reasons).

The draw calls increase is one of the ways of increasing both frame rate under stress, and frame delivery. Sometimes it is only frame delivery that gets better.

I'm completely happy with the 1023 drivers. I played some GTA V in the weekend, they seem to include the profile from the 15.4.1. When AMD releases the Witcher driver I sincerely hope they release a similar Windows 10 driver.

I will benchmark at least 4 drivers once the Witcher is out, it would be interesting to see.

EDIT: I just saw the video you posted. I wonder how things would be if he used the 1023 drivers. Also a big surprise seeing the R9 280x beating the GTX 770 in Project CARS. Wtf. It seems like NVIDIA has dropped the ball completely on Fermi era cards. Thank God I didn't get that 580 back then.

 

So you say 1023 is better than 1018? I don't want to change only to find out it's bad and change back to 1018. I know every system can give different results but if I at least know from a few people it works better than 1018 then I will change and find out.

I agree would be nice to see the frametimes between the 4 drivers.

oh also remember they using 5930k @ 4.7GHz so maybe it's fast enough to make full use of the r9 280x so overhead is less of an issue than even a 4790K @ 4.4GHz. We all know newer games tend to use more threads. And if you use 5-6 threads a 5930k will win cause it has 6 cores 12 threads os instead of using 4 cores and 2 HT cores on 4790k it will use 6 real cores and that's already faster.

 

1023 is basically 1018 with the 15.4 profiles.

Overhead is probably not an issue in the way they test, that CPU is a beast.
The game is out today, I'll try to have some numbers and framedumps by tomorrow afternoon.