AMD High Overhead Drivers on DX11

[ObscureAngelPT]

Maybe in 23 february or something i can get a 750TI lended by another friend.
To benchmark this.

Nvidia, AMD, the game developers, the OS are all done by different people. As hard as some may try to get even frames on all systems with varying hardware and software and OS it just is never going to happen. Your always going to see different results. In a case such as this, I attribute it the following stated and a clear hardware difference.

That isn't to say AMD might not be suffering from some serious issues as to what those are exactly I can't say, but no one really can I think.

I am not being clarifying well.
AMD might not have any issues at all.
This difference that i claim here (Forget about the fact that the GTX 770 is more powerful just for a second, and that GAMEGPU and DF benchmarks are legit), is that nvidia can produce more frames even when i am bottlenecked by something we don't know.
This difference appeared after one driver that nvidia released that created something that really left the CPUs of the people much more confortable, or by your words, may it be another thing..

But i discover this from tom's hardware, it might be a good Reading.
Take a look

http://www.tomshardware.com/news/nvidia-geforce-337.50-driver-benchmarks,26473.html

It was after this drivers, that people with low end CPU's like me started to get a big boost on performance comparing to AMD.
And thats why Star swarm escalated pretty much than other since its a CPU Bound benchmark

 

[Ferrum Master]

So it is a driver problem?

No it is pure anemy of your aged system actually... I have something similar at work... it only moves pulling on 4.00Ghz... and 880 chipset... it lacks first of all SIMD extensions... that actually games uses.

Okay... then... how you can comment this one then? What is your vision?

 

[ObscureAngelPT]

No it is pure anemy of your aged system actually... I have something similar at work... it only moves pulling on 4.00Ghz... and 880 chipset... it lacks first of all SIMD extensions... that actually games uses.

Okay... then... how you can comment this one then? What is your vision?

My vision is, you are missing the point..
They are testing Ryse, which is a fully GPU Bound game with a i7 extreme at 4.5GHZ.
Its obvious that its not going to bottleneck shit with that CPU.

Now ask them to do the same with an i3 for exemple and you will be seeing the results go anormaly down to amd.
BTW AMD Gpus have great performance advantage on GPU processing in Ryse, Far cry 4 and Shadow of mordor.

 

[Ferrum Master]

My vision is, you are missing the point..
They are testing Ryse, which is a fully GPU Bound game with a i7 extreme at 4.5GHZ.
Its obvious that its not going to bottleneck shit with that CPU.

Now ask them to do the same with an i3 for example and you will be seeing the results go anomaly down to amd.

No you missed this one... didn't look at the dates... and the mathematical algorithm changing resolution and effects is linear in this game. As I said... nvidia fixed multi threading in their department, they recompiled the driver using latest tool chains... I hope you have ever tried Gentoo or compiled an Android kernel for yourself? AMD made that do actually... thank them. And vice versa it goes on.

So... one has to under clock/disable cores on his own CPU and bench on the same screwed testbed using same frequencies otherwise it is POS. Each system clicks on their own and has to be fine tuned. Game reviewers mostly are short on time and use automated scripts without much thinking. Like the i3 in that racing game screen shot does not have Turbo and the i5 has actually on not too shabby 3.9GHz? Do you accept that AMD driver is single threaded and nVidia is Multi?

 

[ObscureAngelPT]

Then its amds fault

 

[Ferrum Master]

Then its amds fault

It is childish to bash it is someones fault... calm down, those are apples and oranges. We are still talking about specific scenarios. And no bullet proof benchmark is shown... all examples are rejected due to lack of truthfulness...

Driver dates/optimizations
System config
Crippled tester hands/beer
Fanboyism/Fanatism
EOL/Archieved systems should not be taken into account due to lack of latest SIMD instructions.

 

[R-T-B]

I'm not sure the amd driver is not multithreaded. That kinda screams [citation needed] to me. Certainly if it's single threaded and NVIDIA's is not, this would be well documented.

Not saying it isn't true, just want to see supporting evidence.

 

[Ferrum Master]

I'm not sure the amd driver is not multithreaded. That kinda screams [citation needed] to me. Certainly if it's single threaded and NVIDIA's is not, this would be well documented.

Not saying it isn't true, just want to see supporting evidence.

I can say not completely... multi threading is very complicated thing when doing drivers, as the CPU scheduler(data prefetch logic that actually differs with each CPU architecture and is smarter(debut Pentium 3 Tualatin)) must be aware of the flowing data... AMD driver simply has less chunks that can be ran through in parallel, it simply does not do that. There for the difference occurs... I hope I made my self clear enough... just remember P4 why their HT sucked because their pipes were too long and vice versa from the data point of view. Why GPU parallel tasks are not for every task. it might be connected GCN and their kernel as such. We can only speculate, and it is way too disperse to make judgments on different CPU's.

 

[ObscureAngelPT]

Guys, one question, if i benchmark the games (TWMTBP and AMD Gaming evolved) with a GTX 650 vs a HD 7850.
I gonna change the papers and test with a worse graphic card than mine?

What do you think about it?
Problem is, i need to test that, after day 18 of this month, and its only possible of the good will of the guy that is going to lend me, the GPU.

I can say no completely... multi threading is very complicated thing when doing drivers, as the CPU scheduler(data prefetch logic that actually differs with each CPU architecture and is smarter(debut Pentium 3 Tualatin)) must be aware of the flowing data... AMD driver simply has less chunks that can be ran through in parallel, it simply does not do that. There for the difference occurs... I hope I made my self clear enough... just remember P4 why their HT sucked because their pipes were too long and vice versa from the data point of view. Why GPU parallel tasks are not for every task. it might be connected GCN and their kernel as such. We can only speculate, and it is way too disperse to make judgments on different CPU's.

I guess we are trying to say the same thing, when i sayed its amd's fault, its not about putting the final claim on them.

But that nvdia MAY have a better multi core support which cause a better performance improvement or less overhead or whatever, but in the end gave me more fps in a situation that was not possible to happen. Again speculation.

Imagine that when i put the GTX 650 of that friend with the same settings, i would get more fps on a bottleneck situation in an AMD Gaming evolved game.
That would point out that something is wrong in something from AMD.

 

[Ferrum Master]

Guys, one question, if i benchmark the games (TWMTBP and AMD Gaming evolved) with a GTX 650 vs a HD 7850..

I am afraid... those both cards will be bottlenecked so the tests will show haywire results. Especially in CPU intensive games... you will not have enough budget time... so the test results are not really worthy. As the FPS won't be stellar, each 5 FPS will give a drastic damage. It won't be a benchmark, more like a lottery.

 

[Toothless]

If someone shipped me an AMD cpu and board, I could test with my 660..

 

[ObscureAngelPT]

If you had an AMD GPU similar to the 660, you were able to disable clocks and underclock your CPU to simulate a low end CPU

 

[erocker]

If you had an AMD GPU similar to the 660, you were able to disable clocks and underclock your CPU to simulate a low end CPU

Nope. Not unless you're demonstrating CPUs in the same family.

 

[Ferrum Master]

If someone shipped me an AMD cpu and board, I could test with my 660..

I have some working Sempron 754 combo, it even has SATA, if you are in kind of masochistic mood

 

[ObscureAngelPT]

Well i have a Pentium G haswell in my home too

 

[Ferrum Master]

Well i have a Pentium G haswell in my home too

Unlocked one?

 

[Toothless]

I have some working Sempron 754 combo, it even has SATA, if you are in kind of masochistic mood

I'll take that and hate myself while feeling good about it.

 

[ObscureAngelPT]

Unlocked one?

No..
It is the G3220 (3.0GHZ) in fact its not haswell, my bad.
Paired up with a chip H81 board and integrated GPU only.

I build that desktop for my parents, they like to go browse on web, and my father loves to play pinball FX2 and Solitarie ^^
And some times judging by the history that are some mean websites he search for

Thats why this is a so low end computer

I still have an AMD Sempron 3000+ Palermo too, but i think the MB might got damaged, it is always restarting, its not the rams i tested it before.

Served me well for some years

 

[Ferrum Master]

And some times judging by the history that are some mean websites he search for

Don't poke your nose where you shouldn't. Both systems are too weak...

Get and unlocked Pentium if that H81 supports them unlocked and swap the PC's... Cheapest Upgrade you ever will have.

 

[ObscureAngelPT]

Don't poke your nose where you shouldn't. Both systems are too weak...

Get and unlocked Pentium if that H81 supports them unlocked and swap the PC's... Cheapest Upgrade you ever will have.

Of course it dont support
Neither CPU and MB.

I have the right to poke it, its me how tell him about those websites
That system is obviously weak and its enough for their activities, they arent going to be doing anything demanding. with it.
And hey the integrated GPU plays just fine CS:GO at 50/60fps at 1366X768 at high hehehe

 

[LightningJR]

This is an interesting thread I really wish some review website would test this theory properly, I would really like to see the accurate results. hint TPU...


If you bottleneck the CPU enough it doesn't matter if you use a 260 or a 290X the FPS will be the same since the CPU can't push the data to the GPU quick enough to take advantage of the leftover power of the video cards. This is the essence of a CPU bottleneck.
If the game/benchmark you run stresses the gimped CPU to 100% for the total of the test (very important) then the FPS difference between say a 7850 and a 780 will accurately show you which drivers use less CPU.

It's also important to make sue both GPUs never reach 100% usage or 100% of VRAM or on the memory controller

The thing is, I am not 100% sure if this is rules out all other variables. This is the most accurate I can think of doing this experiment.

 

[ObscureAngelPT]

This is an interesting thread I really wish some review website would test this theory properly, I would really like to see the accurate results. hint TPU...


If you bottleneck the CPU enough it doesn't matter if you use a 260 or a 290X the FPS will be the same since the CPU can't push the data to the GPU quick enough to take advantage of the leftover power of the video cards. This is the essence of a CPU bottleneck.
If the game/benchmark you run stresses the gimped CPU to 100% for the total of the test (very important) then the FPS difference between say a 7850 and a 780 will accurately show you which drivers use less CPU.

It's also important to make sue both GPUs never reach 100% usage or 100% of VRAM or on the memory controller

The thing is, I am not 100% sure if this is rules out all other variables. This is the most accurate I can think of doing this experiment.

Thats the way i benchmark, problems is no game uses my CPU at 100%, altough GPU usage drops in the games, thats what i call my bottleneck

I really really really need a more weaker or on match GPU from nvidia to confirm all this things.
I would really really really love that TPU could pick up this rumors or our uncertain confirmations to create a good article to get finnaly the truth out of this.

We are almost with #100 posts, with a lot of crappy talk and Off topic, but its in fact an interesting thing.
I should be studing for my tomorrow network security exam, but i can't think in it because of these bottleneck talk have been quite a lot exciting
I think i gonna fail because of this -.-

 

[LightningJR]

Thats the way i benchmark, problems is no game uses my CPU at 100%, altough GPU usage drops in the games, thats what i call my bottleneck

I really really really need a more weaker or on match GPU from nvidia to confirm all this things.
I would really really really love that TPU could pick up this rumors or our uncertain confirmations to create a good article to get finnaly the truth out of this.

We are almost with #100 posts, with a lot of crappy talk and Off topic, but its in fact an interesting thing.
I should be studing for my tomorrow network security exam, but i can't think in it because of these bottleneck talk have been quite a lot exciting
I think i gonna fail because of this -.-

Well the GPU not being utilized at 100% all the time doesn't always mean you have a CPU bottleneck, remember that.

LOL, I hope you don't fail because of TPU. We're not that important Good luck.

 

[ObscureAngelPT]

Well the GPU not being utilized at 100% all the time doesn't always mean you have a CPU bottleneck, remember that.

LOL, I hope you don't fail because of TPU. We're not that important Good luck.

Yes, but having a better CPU seems to deliver more frames in those cases.

 

[MxPhenom 216]

Can we just conclude to the fact that Nvidia released new drivers last year to minimize CPU overhead issues with DX11, in efforts to respond to Mantle without creating a whole new API. I believe they have done this will relative success. What else is there to really talk about? The OP is comparing apples to oranges here, with not much success.

A relatively pointless discussion IMO. Just re-iterating already known information, and its been known for quite sometime.

The whole overhead issues has been known by AMD, Microsoft, Nvidia, and Intel. AMD took the initiative to develop an API to make overhead and negligible issue. And now in response, it got Microsoft off their asses, to make DX12 do the same thing.