Will AMD ever fix their driver overhead?

[KingFatty]

I'm still confused here. I read the first post, and it sets up the problem as being, if you turn your visual settings down to minimum, something bad happens with AMD cards, and something good happens with NVidia cards?

What? Can someone explain a bit more in simplistic terms, what is the situation? Do you have to use settings that make sense, or is this a way to try to make NVidia look good by finding a bizarre combination of settings that nobody would ever use? Isn't the point to make the game look good? I think I'm missing the main point, because this seems like it doesn't make any sense? Halp

 

[eddman]

Yes, the performance improvement from these drivers was huge. /s

Actually, on slower CPUs where driver overhead has the biggest effect, it was. I had a core 2 duo back then and AC IV went from unplayable sub 20 FPS to playable 20-30 FPS. The difference, to me, was massive.

I don't know enough about AMD's drivers to comment though.

 

[IEC]

Overwatch doesn't need more than a Radeon 290 to satisfy 120/144 Hz junkies, which I will remind you is a three year old card at this point. A RX480 is also capable of high refresh rate ultra settings in OW.

I'll refer you to my post in a different thread where I benchmarked a 6v6 Temple of Anubis match @ Ultra settings on a R9 290:

With data to back up my claims of 1080p Ultra @ 144Hz is doable with two settings changes:

Improve FPS considerably with little to no impact on visuals
Set Dynamic Reflections to Medium or Low
Set Local Fog Detail to Low

This is using a Radeon 290 @ 1050MHz with a stock i7-6700K. Resolution was set to 1920x1080 @ 120Hz with Vsync off, using the Ultra preset with Dynamic Reflection and Local Fog set to low. Run was with a full Temple of Anubis match, live, 6v6 humans.

Results:
2016-06-27 10:01:31 - Overwatch
Frames: 53298 - Time: 372172ms - Avg: 143.208 - Min: 107 - Max: 210

Summary file:
https://www.dropbox.com/s/c358c7r8kcdgrzt/FRAPSLOG.TXT?dl=0

FPS CSV:
https://www.dropbox.com/s/fw7dfu3p8k5djto/Overwatch 2016-06-27 10-01-31-76 fps.csv?dl=0

Frametimes CSV
https://www.dropbox.com/s/wpbpu89ezb65rzl/Overwatch 2016-06-27 10-01-31-76 frametimes.csv?dl=0

Min/Max/Avg CSV
https://www.dropbox.com/s/stbvjof1ic84sd5/Overwatch 2016-06-27 10-01-31-76 minmaxavg.csv?dl=0

http://forums.anandtech.com/showpost.php?p=38320315&postcount=23

 

[3DVagabond]

Actually, on slower CPUs where driver overhead has the biggest effect, it was. I had a core 2 duo back then and AC IV went from unplayable sub 20 FPS to playable 20-30 FPS. The difference, to me, was massive.

I don't know enough about AMD's drivers to comment though.

Well, you know what they say, there's an exception to every rule.

If that's the type of case that these drivers helped in then nVidia should have conveyed that. Instead I've seen people use this to perpetuate the bad driver myth for AMD. Which is all this thread is. Every review I saw of it on reputable sites showed it to be a fairly typical performance driver, as far as it's improvements. So, nothing personal but anecdotal evidence like "It worked for me" doesn't trump all of the reviews like the one I quoted above.

Now years later, in the hope that people forgot the truth about the minimal improvement these drivers brought we have people posting in the AMD section, of all places, about this. If you look at reviews AMD performance only suffers in a few situations. Gameworks titles, of course. Where nVidia loads certain areas of the rendering where they have an advantage. Games they didn't get access to pre release to optimize for. Other than that AMD performance in DX11 hasn't been worse than nVidia. Even though their hardware is designed for modern low level API's and has been for years they have done fine.

Every GCN from GCN 1.0 has proven to be quite competitive for extended periods of time. I think this speaks to the success of AMD's driver team since it's not the hardware changing that's improved it's performance. Their drivers have continued to improve performance on hardware while nVidia's have not. As I pointed out, look at the 770 vs 280X. When Kepler came out the 680 was ~10% faster than the 7970 on the same site (TPU). It's now ~15% slower. This is on a site with a suite that always favors nVidia. Now I doubt either card got slower. It's more likely that AMD's drivers have outpaced nVidia's by ~25%. Yet here we have another thread accusing AMD of broken drivers. And it's using as evidence an nVidia marketing slide that every review showed was another nVidia lie in a long list of them.

 

[SolMiester]

Well, you know what they say, there's an exception to every rule.

If that's the type of case that these drivers helped in then nVidia should have conveyed that. Instead I've seen people use this to perpetuate the bad driver myth for AMD. Which is all this thread is. Every review I saw of it on reputable sites showed it to be a fairly typical performance driver, as far as it's improvements. So, nothing personal but anecdotal evidence like "It worked for me" doesn't trump all of the reviews like the one I quoted above.

Now years later, in the hope that people forgot the truth about the minimal improvement these drivers brought we have people posting in the AMD section, of all places, about this. If you look at reviews AMD performance only suffers in a few situations. Gameworks titles, of course. Where nVidia loads certain areas of the rendering where they have an advantage. Games they didn't get access to pre release to optimize for. Other than that AMD performance in DX11 hasn't been worse than nVidia. Even though their hardware is designed for modern low level API's and has been for years they have done fine.

Every GCN from GCN 1.0 has proven to be quite competitive for extended periods of time. I think this speaks to the success of AMD's driver team since it's not the hardware changing that's improved it's performance. Their drivers have continued to improve performance on hardware while nVidia's have not. As I pointed out, look at the 770 vs 280X. When Kepler came out the 680 was ~10% faster than the 7970 on the same site (TPU). It's now ~15% slower. This is on a site with a suite that always favors nVidia. Now I doubt either card got slower. It's more likely that AMD's drivers have outpaced nVidia's by ~25%. Yet here we have another thread accusing AMD of broken drivers. And it's using as evidence an nVidia marketing slide that every review showed was another nVidia lie in a long list of them.

The reason GCN driver gots better is AMD keeps rehashing the same old architecture, nothing new but some faster memory and optimised drivers for games.
You will have to show us proof of all these lies, cause these forums are full of it.
Oh, and I didn't see broken driver claims, just heavy overhead demand.

 

[eddman]

Umm... That wall of text wasn't needed. You don't need to convince me. I'm sure both companies are doing their best to write the best drivers they can.

I just wanted to point out that reducing driver overhead mostly results in tangible gains on slower CPUs, which is expected.

Well, you know what they say, there's an exception to every rule.

If that's the type of case that these drivers helped in then nVidia should have conveyed that. Instead I've seen people use this to perpetuate the bad driver myth for AMD. Which is all this thread is. Every review I saw of it on reputable sites showed it to be a fairly typical performance driver, as far as it's improvements. So, nothing personal but anecdotal evidence like "It worked for me" doesn't trump all of the reviews like the one I quoted above.

That's not an exception. That's the rule. Reducing driver overhead does very little for fast CPUs and a lot for slow ones.

I do agree that they didn't convey it properly. You should never put much faith in marketing/advertisement departments.
If anyone uses that driver to show that AMD has bad drivers, then they shouldn't be taken seriously.

Yes, reviews didn't show much because they used latest, fast CPUs for their tests. They did it wrong, basically.

There is nothing anecdotal about what I wrote. It's very much verifiable. It's not just a "me" case.

Still, as you mentioned, it did pretty much jack for fast processors. That part is very much true. The OP is wrong to use it as "evidence".

 

[3DVagabond]

The reason GCN driver gots better is AMD keeps rehashing the same old architecture, nothing new but some faster memory and optimised drivers for games.
You will have to show us proof of all these lies, cause these forums are full of it.

So, AMD hasn't released any new arch's since Tahiti? Just because they stick with the GCN name and don't just slap some random physicist's names on every new card doesn't mean they still have the same old architecture.

My post of AT's benches for those drivers prove the lie of the marketing slide. They gave about the same performance improvement of typical performance drivers. Nothing magical like 60%+ except where they released a new SLI profile.

Oh, and I didn't see broken driver claims, just heavy overhead demand.

Title of the thread; "Will AMD ever fix their driver overhead". You only need to fix something if it's broken.