AMD High Overhead Drivers on DX11

[ObscureAngelPT]

Hi there Techpowerup community.
This is my first post but a very very important one.

Dispite the fact that i'm new here, i'm not new on many other fórum, i do love HW and gaming at the same time.

I have a Phenom II X4 945@ 3.7 and a HD 7850.
The CPU is old so its normal that i get bottlenecked in many titles.

Altough, lately some weird reviews had came out by GameGPU and Digital Foundry which are sites that i don't trully trust for hardware benchmarking, but they are the ones that benchmarks their games in low-end, mid end and high end CPUs and were the first to recognized that AMD have really a problem with DX11.

So with that i started my own investigation, a friend of mine lend me its GTX 770 which is obviously much better than my HD 7850, but i decided to benchmark games either Nvidia TWMTBP and AMD Gaming evolved where i usually have bottleneck judging by MSI AB GPU Usage.

And the fact is, in bottleneck situation i can archive an increase of 10-20fps comparing to using my AMD on a bottleneck situation, which was what those reviewers had stated.

Anandtech released a new analysis too about DX12 on star swarm and there was completly noticebla that AMD only archive 8FPS even with a R9 290X and a nvidia 750Ti could archive 20's fps.
Star Swarm is a very demanding benchmark to the CPU and a good overhead or a good API really helps there, and it was a benchmark released by AMD and Oxide games in order to demonstrate AMD Mantle.

So in case you are confused i created a vídeo that proofs everything with all my own investigation with benchmarks, games and other reviewers.

My objective is not earning views on youtube, i dont care about that, but it is to force AMD to see that they have a problem there, since for 2 months they never replied my Report issue.
So i started to spread the world through the best hardware forums in order to make some noise about it.

Here is the vídeo:

 

[xfia]

if there is problems then AMD and Microsoft both know about it.

no doubt the best optimized games are always gpu bound but there is some pretty fun cpu bound games too.

if you want good performance to matter what game you play then you have to do what everyone else does and buy a i5.

 

[ObscureAngelPT]

But i am aware of that.
But the fact os that in cpu bound Games using dx11. Nvidia gpus can archive better framerate.
its a difference of 10-20fps due to better optimized.

I dont deny that an i5 would fix this, but if i got a 660 for example, i would get better performance vecause of less overhead on those Games. Not sure if you are understanding.

 

[erocker]

I doubt AMD is watching.

Are you basically saying that with a lower end CPU, Nvidia's cards perform better than AMD's cards in DX11? Where are the results you're getting for lower end CPU tests?

 

[ObscureAngelPT]

i am not only saying but confirming it on the video.
I benchmarked on the video 2 nvidia the way to be played Games (dying Light and project cars). I testes too star swarm and saints row 4 which are both amd gaming evolved.
star swarn was benchmarked in extreme 1080p on rts scenario.

The resulta are basically on the video o made in order to prove it.
i started talking about this on guru3d forums where more people were complaining about it, many people wasnt convinced until the video.

gamegpu and digital foundry notice that already. Borh are websites that i dobt trust. But they are the only ones that benchmark with low end cpus or outdated, so they notice it first the the big websites like yours or guru3d.

I benchmarked more Games but didnt made a video. Sorry for some errors but i am writing through smphone. I am not home right now.

EDIT: I am back now in the desktop.
Resuming our talk, i start to discover this through some benchmarks at GameGPU.Ru where games like Alien isolation, dead rising 3 Dragon Age inquisition seems to deliver more frames on Nvidia when CPU Bound.
Altough its a russian website and unknown to many, i didnt trust it at all, but i was curious to check if that was true.

Later i digital foundry (another site that performs framerate console benchmarking and some PC Benchmarks too), stated the same thing, even now on their review of the GTX 960 they talked about that, dispite the fact that the review of the GPU is inconsistence, they proved that AMD GPUs bottleneck low end CPU's.
So its 2 websites saying the same.

So i asked for a friend of mine to lend me its GTX 770 in order to benchmark, it was not the best decision since the GTX 770 is far more powerful than my HD 7850, but with MSI Afterburner it was easy to see when my GPU is getting bottlenecked, so thats why i choose saints row IV, project cars and dying light and Star swarm, since they some of the games that my GPU usage are always far away from 99% which means i am bottlenecked by something.

So for exemple like in the vídeo at the same settings, i am severly bottlenecked on my Phenom II X4 having just 33FPS looking through a Windows at médium settings.
With the GTX 770 obviouly i am still bottlenecked, but i got 50FPS.
Project cars, same thing, crossing through a group of houses makes my GPU usage go down on AMD and framerate dropping from 50-60 to 30's.
On nvidia at the same settings it drops from 90 to 60/70.

I though could be gameworks.
But testing out Saints row IV which is a AMD Gaming evolved game where i am severly bottlenecked too, since my GPU is always at 50-80% of usage and getting 30-40FPS while roaming through the city, on the nivida side i get >40FPS.

I benchmarked too NFS Rivals where exists places where i get bottlenecked and my HD 7850 drops to 24FPS, unlocking the framerate to 60 i got only 30/40 bottlenecked too judging by the GPU Usage, with the nvidia, i am still bottlenecked, but i can archive 60FPS and dropping to 46/48FPS on the location where in my AMD i dropped to 24FPS.

And now we have the Star swarm Benchmark from anandtech which shows in the RTS mode and in extreme R9 290X getting only 7/8FPS full of bottleneck, and a simple 750TI in the same CPU running too at the same settings and in the same CPU archiving 22FPS.

This is a very serious thing.
Dispite the fact that DX12 is out there and mantle too.
Mantle is not compatible to all the games, and DX12 is going to be released, but i bet many games will only use DX11, even serious Sam III only uses DX9 after so many years after DX11 was released.
So AMD in the future even a dev team releases a CPU bound game in DX11 it could mean trouble for AMD if they dont fix this.

 

[erocker]

I'll take your word for it. Good luck getting anywhere with AMD.

 

[ObscureAngelPT]

Hahaha.. I guess your words says it all.
Well since more people are getting aware of this, i am expecting that could happen exactly how it happened last year (or 2 years) when someone discovered that the framepacing on the Crossfire was all ****ed up ^^
So one of good websites not sure if it was this, guru3d or Tom's hardware, benchmarked their cards with FCat and discovered that the crossfire was full of stuttering.
You maybe remember that

Well you should remember too that they were working 2/3 months fixing the frametimes of the Crossfire after that thing was discovered..
I want to try to do the same, spread the word to cause chaos.
Problem is.. DX12 is close to release and they may use that as an excuse to not optimize their DX11.

Anyway, if anybody want to discuss this, they have now a place where to do it.

 

[ISI300]

I agree with the topic starter. I also noticed that my GTX 580 was getting somewhat higher frame rates than my friend's 7970. I did log the numbers down, I'll post them if I can find them.
One thing that I do remember is that in FC4 as soon as I made it to the village the 7970 dropped to 24 fps while the 580 was pushing 40. The CPU was an x4 955 at stock.

 

[BiggieShady]

So one of good websites not sure if it was this, guru3d or Tom's hardware, benchmarked their cards with FCat and discovered that the crossfire was full of stuttering.
You maybe remember that

I believe it was pc perspective ... I also remember in the days of VLIW architecture amd drivers were less cpu heavy than nvidias on fermis back then. Many things changed since amd cut off money to the r&d

 

[R-T-B]

As much as I don't really want to confirm this, I'm pretty sure he is right.

Since switching from my GTX 580 to an R9 290X on my profile system, I must admit it is a kick in the pants. But when the CPU (an old nehalam thing) gets bogged down badly, like when loading a system in Elite Dangerous or something, I noticed signifigantly more FPS drop than the GTX 580 suffered. Because of the raw power the R9 290X has it's not a big deal and I still average far higher FPS, but if my CPU were weaker (say an old Phenom) it probably would be.

I think this issue needs attention as well. Mantle isn't a fix for this at all, nor is DirectX 12. I find it kinda ironic that the company squealing the most about API overhead have the worst performing DirectX 11 stack... tend to your own overhead issues before starting a low-level scramble that only makes devs life hell, AMD.

I believe it was pc perspective ... I also remember in the days of VLIW architecture amd drivers were less cpu heavy than nvidias on fermis back then. Many things changed since amd cut off money to the r&d

I don't see how VLIW has anything to do with CPU overhead. If you ask me the R9 290X is an excellent card, it's the drivers that need work. Not the engineering.

 

[ObscureAngelPT]

I had a Vliw, it was my HD 5770, it decided to die after 2 years and 1 month, the brand didnt gave me warranty, so i waited some months with my integrated GPU, and buyed the HD 7850, i was about to consider a GTX 660 but nvidia was not talkative about the release date of it
So i buyed the HD 7850 for 250€ and 3 weeks later NVidia released their GTX 660 and HD 7850 dropped to 200€, i would have bought the GTX 660 if i had waited in that time ^^.

But its very sad, what is happening, i like a lot of AMD not because of features, but the openess of their technologies, i think they create something that all the market can use, nvidia have too their PhysX3.dll libraries and gameworks, but nvidia is much more propreatary, and i think AMD have more humble atittudes which benefit the market.
But looking to their CPU performance (lets hope AMD ZEN is good), and looking now to their GPUs and now the drivers with high overhead, i have affraid that AMD could go more low than already is.
If nvidia start to monopolize the PC, where are doomed, the prices will start to raise as an arrow, if AMD do something, then well.. Nvidia take an arrow on the knee..

Mantle isn't a fix for this at all, nor is DirectX 12.

Well i disagree.
DX12 fixes this problem.
This only started to happen after AMD released Mantle.
3 months later nvidia released a driver for star swarm that improved CPU performance a lot, people benchmarked star swarm and really improved, somehow nobody benchmarked the other games LOL.
It was after that driver that nvidia started to have advantage over AMD DX11 codepath.

I disagree with you because before that driver, i think nvidia and AMD had more and less the same performance, so looking to the anandtech benchmark of DX12 on star swarm it seems that on DX12 AMD seems to be as fine as Nvidia.
So the DX12 can really fix this.

The problem is, only 1 game is confirmed with DX12, and a lot of games will come out before and after the DX12 release.
Year after year we see an increase number of detail distance and draw calls on games, if devs continue to use DX11 for developing games with a yearly rising detail distance/draw calls, then AMD can really pack the bag and go home, unless they start working.

So like i sayed, DX12 is the solution, but while DX12 is not mainstream AMD should be working on DX11, which is obviously that they aren't and they seem to not care at all, since all the problems even difefferent problems i report to their "Submit bugs" of their drivers, i never ever recieve a reply.

For a company with a lower marketshare, i would expect to have a better support or a more talkative company to create loyalty users, but.. no.....
I'm very sad with them.

 

[OneMoar]

you are running a phenom II with a relatively weak 7850 the problem is your system man not AMD's admittedly crappy drivers
a 770's is orders of magnitude faster then a 7850 any gains you are seeing are related to the gpu being more powerful

 

[R-T-B]

you are running a phenom II good luck ...

What he's running doesn't change the fact CPU utilization is clearly higher on a Radeon GCN product.

I love AMD. Heck I wish they'd make a competitive CPU too. I'm getting worried about them though and while I can't say I'm unhappy with my last card purchase from them, I think cracks are starting to show in the company that best be patched up quickly less the whole bloody dam gives way.

 

[R00kie]

I can confirm this as well. When I was running my FX8320, jumping from a R9 290 to a GTX970 improved a lot of games that were CPU dependent. These games are: Call of Duty Advanced Warfare (this one in particular) , Assassin's Creed Unity, Far Cry 4, All of the Saints Row games (literally every single one) and some other games. Yes, I do realise most of those games are Nvidia optimised, but that doesn't change the fact that the CPU utilisation almost halved.

 

[OneMoar]

What he's running doesn't change the fact CPU utilization is clearly higher on a Radeon GCN product.

I love AMD. Heck I wish they'd make a competitive CPU too. I'm getting worried about them though and while I can't say I'm unhappy with my last card purchase from them, I think cracks are starting to show in the company that best be patched up quickly less the whole bloody dam gives way.

just because the utilization is higher doesn't mean there is a problem the fact that he thinks that means he doesn't know what hes talking about yes AMD's drivers are buggy and prone to weird performance bugs but this is just a case of a old system being pushed far beyond its limits
there is no free performance fix to be had here, the op needs to build a new system with a modern, preferably Intel Base Platform
the higher utilization is from driver optimizations made to extract the most from a already outdated cpu architecture
things run smoother on nvidia because there hardware and software is miles ahead of anything AMD has this has been true for years and its nothing at all new,that being said nvidia pays a lot of people a lot of money to insure that games run better on their platform

 

[R-T-B]

I don't know man. I feel it on my Nehalam system too. I'm pretty sure CPU load is higher too during gaming, anyone have both brand cards and wanna test that? I don't have my 580 anymore.

EDIT: WAIT! I do have my brothers PC which has a GTX 780. More or less equal to a R9 290X yes? Roughly anyways. I'll be back soon with CPU figures and settle this, gotta finish work first though.

 

[OneMoar]

I don't know man. I feel it on my Nehalam system too. I'm pretty sure CPU load is higher too during gaming, anyone have both brand cards and wanna test that? I don't have my 580 anymore.

all the games I see in hist video either use nvidia physix or gameworks or have the "the way its ment to be played" logo
do the math
btw if you wanna be taken seriously at least learn how to encode a video and run a proper benchmark and not rip stuff off Google in a attempt to get something for nothing

 

[R-T-B]

all the games I see in his crappy quality video either use nvidia physix or gameworks or have the "the way its ment to be played" logo
do the math

Oh I know his methodoloy is flawed. But i feel it in other games too. It could be placebo but I don't think it is in this instance... especially when I went from a far weaker card to a far more powerful one.

Hence me wanting to do my own bench above, will be back with some real figures for you. Any advice to keep it fair? I'm thinking of running a free benchmark such as the Heaven benchmark people were using here, and comparing CPU load throughout on identical hardware sans GPUs

 

[ObscureAngelPT]

you are running a phenom II with a relatively weak 7850 the problem is your system man not AMD's admittedly crappy drivers
a 770's is orders of magnitude faster then a 7850 any gains you are seeing are related to the gpu being more powerful

I wish it could be like that, but in fact its not.

All the games i benchmarked i have locations or even all the time bottleneck due to my weak Phenom II, its true..
So my GPU usage is most of time lower than 90%, which means even if i switch my GPU to a R9 290X i will not see an increase on the framerate in the same settings, and my GPU usage on a R9 290X will surely be something like 40% XD

Altough pairing up a nvidia, it makes me getting more frames where were impossible to have.

For exemple on the video on the first post, on project cars.
I have 98/99% of GPU usage on the start of the race, its obvious that its my GPU limiting the performance, having a Phenom II or having an overclocked I7 right there its equal.
Altough crossing the town, my GPU usage drops to 70% and the framerate dips down to 30's FPS, well there i have bottleneck and yes i should have a better CPU, if i pair up a R9 290X i still gonna have those 30's fps.

But if i put up the GTX 770 for exemple, i can logically get more frames on the start, since i was obviously getting no more performance due to my HD 7850.
But crossing the town where my CPU start to limite my HD 7850 to 30fps, the GTX 770 is able to get >60FPS even having my Phenom II X4 bottlenecking obviously the GTX 770.

Project cars its not the best exemple since its a nvidia title and its yet in developing, but the same happens to many other games even AMD Gaming evolved games.
Its a 20 minutes video that clearly explains everything with all the prooffs if you want to check it out.

 

[OneMoar]

Oh I know his methodoloy is flawed. But i feel it in other games too. It could be placebo but I don't think it is in this instance... especially when I went from a far weaker card to a far more powerful one.

Hence me wanting to do my own bench above, will be back with some real figures for you. Any advice to keep it fair?

use standard bench-marking procedures you need to make base runs on each platform and determine how to play the game so you get consistent data
you will be wasting your time tho I know nvidia will win regardless

 

[OneMoar]

I wish it could be like that, but in fact its not.

All the games i benchmarked i have locations or even all the time bottleneck due to my weak Phenom II, its true..
So my GPU usage is most of time lower than 90%, which means even if i switch my GPU to a R9 290X i will not see an increase on the framerate in the same settings, and my GPU usage on a R9 290X will surely be something like 40% XD

Altough pairing up a nvidia, it makes me getting more frames where were impossible to have.

For exemple on the video on the first post, on project cars.
I have 98/99% of GPU usage on the start of the race, its obvious that its my GPU limiting the performance, having a Phenom II or having an overclocked I7 right there its equal.
Altough crossing the town, my GPU usage drops to 70% and the framerate dips down to 30's FPS, well there i have bottleneck and yes i should have a better CPU, if i pair up a R9 290X i still gonna have those 30's fps.

But if i put up the GTX 770 for exemple, i can logically get more frames on the start, since i was obviously getting no more performance due to my HD 7850.
But crossing the town where my CPU start to limite my HD 7850 to 30fps, the GTX 770 is able to get >60FPS even having my Phenom II X4 bottlenecking obviously the GTX 770.

Project cars its not the best exemple since its a nvidia title and its yet in developing, but the same happens to many other games even AMD Gaming evolved games.
Its a 20 minutes video that clearly explains everything with all the prooffs if you want to check it out.

you don't know what you are talking about and, I have no time to educate you
sorry
./unsub good luck people you are gonna need it

 

[ObscureAngelPT]

all the games I see in hist video either use nvidia physix or gameworks or have the "the way its ment to be played" logo
do the math
btw if you wanna be taken seriously at least learn how to encode a video and run a proper benchmark and not rip stuff off Google in a attempt to get something for nothing

Well Star Swarm is made in parternership with AMD, and Saints row IV is an AMD Gaming evolved title, i test both games.
And i didnt record with neither MSI AB or any other software to capture the real frames of it, and it was recorded with a low end Smartphone.
The quality was the less thing i was interested, i was more interested in showing that this is real.
YOu can denial as much you want, i just hope if you have AMD you don't start to be affected by their lazyness.

I showed all the facts, if you want to defend them its up to you, i want them to fix this in other to me, and you and any user with AMD have better performance and better drivers.
Defending them will not gonna make the problems go away.

But don't worry, i will not reply more your posts, you are clearly a blind fanboy that can't even make an effort to see the problem. And yes i like AMD, my case is full of stickers of AMD all over the place they even sent me some stickers saying AMD Proud team member.
Although, liking it is one thing, being a blind fanboy is another.

 

[R-T-B]

use standard bench-marking procedures you need to make base runs on each platform and determine how to play the game so you get consistent data
you will be wasting your time tho I know nvidia will win regardless

You know the NVIDIA 780 will win regardless? In heaven benchmark it actually looses, but I'm more interested in CPU usage overhead from the drivers.

EDIT:

Anything wrong with this testing methodology?

Identical systems, one with a GTX 780, other with an R9 290X. Stock clocks on CPU and GPUs.

Start task manager:

Run Unigine Heaven benchmark using TPU's settings for that benchmark (there's a thread somewhere here).

Screenshot task manager CPU usage log.

Both systems are identical to the one in my profile, sans GPUs obviously.

If no one points out anything wrong with this I'll do it in about 1 hour.

 

[ObscureAngelPT]

I would love too, but i already returned the GTX 770 to the friend.
I think the CPU usage don't raise with AMD GPU..
I think both CPU usage and GPU usage go down, its like an overhead there preventing the system to get use, but i'm not sure.

Benchmark star swarm in RTS Extreme 1080P in DX11, you will get the FPS average on AMD and nvidia, you can leave the log here.
You will see that your GTX 780 will have more than double frames than the R9 290X, and this is an AMD parternership benchmarked

 

[eidairaman1]

I'll take your word for it. Good luck getting anywhere with AMD.

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