AMD Driver overhead tested (PCLAB.pl)

[Majestic]

So many people have been bringing this argument up, but evidence has been mostly correllation. There aren't many R9-390 vs GTX 970 with multiple CPU's to pass around in order to emperically prove this point.

I've been asking hardware.info for this test, but their initial testing ruled out no big differences. Mainly due to the fact they weren't testing with different CPU's rather checking CPU load. However, on a 8C/16T CPU that can be very deceptive.

 

Apparently PCLAB.pl recently did a test which shows just exactly what is meant when people say AMD driver overhead (CPU).

 

Full test here; http://pclab.pl/art60000-21.html

 

The pages of interest are;

GTX 960 vs 380X; https://translate.googleusercontent.com/translate_c?act=url&depth=1&hl=nl&ie=UTF8&prev=_t&rurl=translate.google.nl&sl=auto&tl=en&u=http://pclab.pl/art60000-20.html&usg=ALkJrhg06hJlLQLpavuzqv7fVoF1pYaluw

GTX 970 vs 390; https://translate.googleusercontent.com/translate_c?act=url&depth=1&hl=nl&ie=UTF8&prev=_t&rurl=translate.google.nl&sl=auto&tl=en&u=http://pclab.pl/art60000-21.html&usg=ALkJrhiCiWM_xct_ojMuSXiHwHgD0l1DMQ

GTX 980TI vs Fury X: https://translate.googleusercontent.com/translate_c?act=url&depth=1&hl=nl&ie=UTF8&prev=_t&rurl=translate.google.nl&sl=auto&tl=en&u=http://pclab.pl/art60000-22.html&usg=ALkJrhiWkWMCsnphhFbjlt_6l88wTqwzkw

 

Test is in polish, i've added google-translated links. I don't want to steal traffic away from them by copying all of their results. So visit those pages for reference. 

 

This places into new light the current recommendations of in particular the 380 over the 960 (though bearing in mind, they're testing 380X vs 960 here), and the 390 over the 970. It also highlights that the more powerfull the GPU is, the more CPU is an issue. 980TI vs. Fury X results are pretty telling.

 

Some examples;

http://pclab.pl/zdjecia/artykuly/chaostheory/2016/01/gfx/cpu_960/c3w_1920vh.png

http://pclab.pl/zdjecia/artykuly/chaostheory/2016/01/gfx/cpu_960/f4b_1920u.png

 

These results indicate the issue with the way GPU tests are done. Usually tested with very fast CPU's (4,7ghz 6700K's, 4,6ghz 5960X's) to eliminate CPU bottlenecking. Which is ofcourse the correct way to do it. But it can remove some of the nuance. Yes, on a fast CPU the 380X is no doubt faster than the GTX 960. I've always said AMD's hardware is faster. But usually when recommending those cards, OP has something like an i3 or an i5 locked. And in those scenario's you can see the GTX 960 just edge out.

 

Except for extremely GPU bound locations;

http://pclab.pl/zdjecia/artykuly/chaostheory/2016/01/gfx/cpu_960/w3f_1920u.png

 

Same goes for the GTX 970 vs. the 390. Though it's much less in favor of the 390 in the CPU-bound games like GTA, Fallout 4 etc.

 

http://pclab.pl/zdjecia/artykuly/chaostheory/2016/01/gfx/cpu_970/f4b_1920u.png

http://pclab.pl/zdjecia/artykuly/chaostheory/2016/01/gfx/cpu_970/c3w_1920vh.png

 

Anyway, thought it was an interesting test. Check it out. And take this into consideration when entering another topic labelled "Which GTX 960" and answering it with "R9-380". Please check which scenario, and which CPU OP has.

If it's some FX-6300 or i3...don't.

 

[Morgan MLGman]

Next time when you need a translation from PClab.pl, just pm me dude ^^

[Majestic]

Thems lots of words

 

It also highlights that pairing AMD CPU's with AMD GPU's is the worst possible combination. AMD CPU's have double the framerate on Nvidia hardware. Likewise, AMD GPU's have much better performance on Intel CPU's.

[Monarch]

Thank you for posting this. And props to pclab for doing the tests. It was thanks to them that the first topics about AMD's driver overhead emerged, which after months of suffering poor performance in certain games and CPU-bound situations with my 290X finally answered my question as to why that had been happening. 

 

Some people claimed recently that AMD fixed their drivers, but this proves that their overhead is still much higher than Nvidia's.

[Majestic]

19 hours ago, Monarch said:

Thank you for posting this. And props to pclab for doing the tests. It was thanks to them that the first topics about AMD's driver overhead emerged, which after months of suffering poor performance in certain games and CPU-bound situations with my 290X finally answered my question as to why that had been happening. 

 

Some people claimed recently that AMD fixed their drivers, but this proves that their overhead is still much higher than Nvidia's.

Yep, PCLAB.pl has the most consistent results. It also explains way more than other outlets *looks at techspot*.

 

Haven't seen this "fix" displayed in the latetst tests. Like this one.

[Agosto]

Wait, what? No other test showed that much difference between a locked haswell i5 and an OCed i7...

There should be a minor difference between a 4570 and a 6700k even on CPU intensive games, while in this test there's a much bigger impact even on nvidia cards

[Majestic]

2 minutes ago, Agost said:

Wait, what? No other test showed that much difference between a locked haswell i5 and an OCed i7...

There should be a minor difference between a 4570 and a 6700k even on CPU intensive games, while in this test there's a much bigger impact even on nvidia cards

Sorry to be so blunt...but

https://yourlogicalfallacyis.com/bandwagon

 

Unless you elaborate..

[i_build_nanosuits]

22 hours ago, Majestic said:

So many people have been bringing this argument up, but evidence has been mostly correllation. There aren't many R9-390 vs GTX 970 with multiple CPU's to pass around in order to emperically prove this point.

 

are you kidding me? digital foundry has been doing that for a few months now with various AMD and nvidia GPU's as well as with various intel and AMD CPU's...it's been proved times and times again.

 

https://www.youtube.com/user/DigitalFoundry

 

[Majestic]

Just now, i_build_nanosuits said:

are you kidding me? digital foundry has been doing that for a few months now with various AMD and nvidia GPU's as well as with various intel and AMD CPU's...it's been proved times and times again.

 

https://www.youtube.com/user/DigitalFoundry

 

They hinted at it, yes. But there was never an elaborate test for it. They only mentioned it in the games where it occured to a significant degree.

[Majestic]

 

Relevant timestamp; https://youtu.be/frNjT5R5XI4?t=7m15s

 

Another example of the issue, this time not from PCLAB.pl but from DigitalFoundry.

[App4that]

Using the hell out of this topic, keep up the good work.

[Majestic]

It shouldn't be as big an issue as it is. No overclockable Haswell+ i5? Get a 970. Not quite sure why it's as polarising as it is, or what the invested interests of some are to will/wish these results away.

[Shoob]

The Division Beta managed to hit 100% CPU load and was stuttering from time to time. Can this be caused by the AMD driver overhead? I have an i5-4590 and a R9 280X and I wasn't running any CPU demanding software in the background.

Thanks for sharing those tests. Seems like I'll need to visit pclab more often.

[App4that]

1 minute ago, Pandalf said:

The Division Beta managed to hit 100% CPU load and was stuttering from time to time. Can this be caused by the AMD driver overhead? I have an i5-4590 and a R9 280X and I wasn't running any CPU demanding software in the background.

Thanks for sharing those tests. Seems like I'll need to visit pclab more often.

That's it, go for any settings involving shadows first, this will help out a lot.

[Majestic]

3 minutes ago, Pandalf said:

The Division Beta managed to hit 100% CPU load and was stuttering from time to time. Can this be caused by the AMD driver overhead? I have an i5-4590 and a R9 280X and I wasn't running any CPU demanding software in the background.

Thanks for sharing those tests. Seems like I'll need to visit pclab more often.

Hard to say without a test. Guess we'll have to wait for PCLAB.pl or DigitalFoundry to test it. Though their testing between 390 vs 970 OC showed the 970 winning significantly in heavy scenario's. 

 

 

 

[Prysin]

On 18/02/2016 at 6:20 PM, i_build_nanosuits said:

are you kidding me? digital foundry has been doing that for a few months now with various AMD and nvidia GPU's as well as with various intel and AMD CPU's...it's been proved times and times again.

 

https://www.youtube.com/user/DigitalFoundry

 

just a fair warning to you. Unless it says PClabs.pl/hammers AMD/shows despite being worse Nvidia has good points, Majestic will call it fake and/or un-trustworthy. Just wait until i'm done testing my FX 8320. He'll call my results fake too.

[Majestic]

7 minutes ago, Prysin said:

just a fair warning to you. Unless it says PClabs.pl/hammers AMD/shows despite being worse Nvidia has good points, Majestic will call it fake and/or un-trustworthy. Just wait until i'm done testing my FX 8320. He'll call my results fake too.

Oh look, more character assassinations. And just how does that give you any right to lecture me. 

 

And i'm sure you always being in favor of the FX series will in no way affect the way you conduct the tests, choose what to publish or interpret. Not at all.

You're completely impartial, as is demonstrated by how you never use an increasing amount of highlights and bold text when someone doesn't buy your story.

 

See, i can also partake in character assassinations. 

[don_svetlio]

8 minutes ago, Prysin said:

just a fair warning to you. Unless it says PClabs.pl/hammers AMD/shows despite being worse Nvidia has good points, Majestic will call it fake and/or un-trustworthy. Just wait until i'm done testing my FX 8320. He'll call my results fake too.

How many times have we debunked PClabs.pl now? 11? 12? He just ignores it and continues to post those deceptive benches....

[Majestic]

Just now, don_svetlio said:

How many times have we debunked PClabs.pl now? 11? 12? He just ignores it and continues to post those deceptive benches....

You haven't debunked shit, just convinced yourselves that you did, and repeated to say you did.

 

But never really do when pressed.

[Prysin]

14 minutes ago, don_svetlio said:

How many times have we debunked PClabs.pl now? 11? 12? He just ignores it and continues to post those deceptive benches....

Dont bring me into your own head-count

 

Ive not debunked anything, just stated multiple times that @Majestic should get a proper translation going, other then broken ass google translate, so we can clearly get a good, accurate description on how they conduct their tests. Iv'e also called him out on the audacity of calling other benchmarks then his own fake by using his own arguments against him. Whether he is willing to admit he can be wrong or not, is more of the issue.

 

And i repeat, i have yet to dismiss PClabs.pl, i simply wont swallow every benchmark Majestic posts until iv'e gotten a proper technically and grammatically correct translation so i know what their articles actually say, not what google translate THINKS they say.

[don_svetlio]

Just now, Prysin said:

Dont bring me into your own head-count

 

Ive not debunked anything, just stated multiple times that @Majestic should get a proper translation going, other then broken ass google translate, so we can clearly get a good, accurate description on how they conduct their tests. Iv'e also called him out on the audacity of calling other benchmarks then his own fake by using his own arguments against him. Whether he is willing to admit he can be wrong or not, is more of the issue.

 

And i repeat, i have yet to dismiss PClabs.pl, i simply wont swallow every benchmark Majestic posts until iv'e gotten a proper technically and grammatically correct translation so i know what their articles actually say, not what google translate THINKS they say.

Fair, sorry for being so general.

[Majestic]

22 minutes ago, Prysin said:

Ive not debunked anything, just stated multiple times that @Majestic should get a proper translation going, other then broken ass google translate, so we can clearly get a good, accurate description on how they conduct their tests. Iv'e also called him out on the audacity of calling other benchmarks then his own fake by using his own arguments against him. Whether he is willing to admit he can be wrong or not, is more of the issue.

 

And i repeat, i have yet to dismiss PClabs.pl, i simply wont swallow every benchmark Majestic posts until iv'e gotten a proper technically and grammatically correct translation so i know what their articles actually say, not what google translate THINKS they say.

Ask @Morgan MLGman

 

And the reason I call out some of the larger benchmark websites is the fact they show extremely agnostic results when they make no sense. Like DigitalFoundry mentioning that Fallout 4 suffers from AMD driver overhead. And that signals that the game is CPU heavy. Which PCLAB also shows with their CPU test

 

http://pclab.pl/zdjecia/artykuly/chaostheory/2015/11/game_f4/charts4/f4_cpu_nv.png

 

And subsequent GPU test;

http://pclab.pl/zdjecia/artykuly/chaostheory/2015/11/game_f4/charts5/f41_1920u.png

 

But then techspot comes along with a benchmark which has these CPU results;

http://www.techspot.com/articles-info/1089/bench/CPU_01.png

 

But then these GPU results;

http://www.techspot.com/articles-info/1089/bench/1080p.png

 

When they don't coincide with DigitalFoundry. PCLAb.pl, Hardware.info, or any sense of logic judging by their own CPU results. That high IPC scaling should extremely adversly affect AMD, and yet you see AMD take the lead (both min. framerate and average). That means they were running the test in some distant unpopulated zone which had no CPU load and no zone transitioning, or the results are just grabbed out of thin air. Now you can ascribe that to incompetence, but after so many years of experience they should've known better and rather ascribe it to 'on purpose'. Another possibility is that they tweaked the tessellation for the game, seeing as the 780 is on the 970's heels.

[Prysin]

12 minutes ago, Majestic said:

Ask @Morgan MLGman

 

And the reason I call out some of the larger benchmark websites is the fact they show extremely agnostic results when they make no sense. Like DigitalFoundry mentioning that Fallout 4 suffers from AMD driver overhead. And that signals that the game is CPU heavy. Which PCLAB also shows with their CPU test

 

http://pclab.pl/zdjecia/artykuly/chaostheory/2015/11/game_f4/charts4/f4_cpu_nv.png

 

But then techspot comes along with a benchmark which has these CPU results;

http://www.techspot.com/articles-info/1089/bench/CPU_01.png

 

But then these GPU results;

http://www.techspot.com/articles-info/1089/bench/1080p.png

 

When they don't coincide with DigitalFoundry. PCLAb.pl, Hardware.info, or any sense of logic judging by their own CPU results. That high IPC scaling should extremely adversly affect AMD, and yet you see AMD take the lead (both min. framerate and average). That means they were running the test in some distant unpopulated zone which had no CPU load and no zone transitioning, or the results are just grabbed out of thin air. Now you can ascribe that to incompetence, but after so many years of experience they should've known better and rather ascribe it to 'on purpose'.

well, some of it can come down to other hardware.

Fallout4 has been shown to enjoy faster memory, it helps noticeably on the minimum FPS. So faster DDR4 would explain the Skylake results for PClabsl.pl .
BUT, did you at any point bother to see which settings they ran with?

Fallout 4 has many crucial parameters which can drastically change the outcome of a test. A little bit lower draw distance, different type of AA, how about GPU clocks (the one thing that is certainly not verified in any of the tets)....

to call a benchmark bad, you must KNOW and PROVE that the prequisite for calling BS is there. If Techspot got those results with the same settings as PClabs.pl, then i would be extremely suspicious of either. But why?

Because neither would make sense. One scoring way higher then the other, makes no sense. One scoring way lower makes no sense either.

 

Again, semantics is the core issue. The devil is in the details here. And i will not let you call any benchmark bullshit before you can PROVE they used the same settings either by replicating visuals (for proving/disproving DF videos) or by replicating FPS results (playing with settings to match FPS results of say Techspot)...

 

Also, how much does settings play into testing?

This video i think it was explains why the AMD slides show how the Fury X beat the crap outta the 980Ti/Titan X
 

 

[App4that]

4 minutes ago, Prysin said:

well, some of it can come down to other hardware.

Fallout4 has been shown to enjoy faster memory, it helps noticeably on the minimum FPS. So faster DDR4 would explain the Skylake results for PClabsl.pl .
BUT, did you at any point bother to see which settings they ran with?

Fallout 4 has many crucial parameters which can drastically change the outcome of a test. A little bit lower draw distance, different type of AA, how about GPU clocks (the one thing that is certainly not verified in any of the tets)....

to call a benchmark bad, you must KNOW and PROVE that the prequisite for calling BS is there. If Techspot got those results with the same settings as PClabs.pl, then i would be extremely suspicious of either. But why?

Because neither would make sense. One scoring way higher then the other, makes no sense. One scoring way lower makes no sense either.

 

Again, semantics is the core issue. The devil is in the details here. And i will not let you call any benchmark bullshit before you can PROVE they used the same settings either by replicating visuals (for proving/disproving DF videos) or by replicating FPS results (playing with settings to match FPS results of say Techspot)...

 

Also, how much does settings play into testing?

This video i think it was explains why the AMD slides show how the Fury X beat the crap outta the 980Ti/Titan X
 

 

I'd put my Hybrid against anyone's Fury X. Name the time and place.

[Prysin]

10 minutes ago, App4that said:

I'd put my Hybrid against anyone's Fury X. Name the time and place.

learn to read, then come back here.