Driver Optimization Reality.

[xkm1948]

Alright this time we talk about old AMD GPUs and how much driver optimization they got over the course of their life span.

So, take a deep breath and remain calm. No name calling, no cursing and no your own predictions.

This time we use only data and we only discuss data.


So I will lead with a review done by HardOCP specifically on how driver optimization helped FuryX and RX480.


link:
https://www.hardocp.com/article/2017/01/30/amd_video_card_driver_performance_review_fine_wine/1


Summary, check the source for yourself as well.

FuryX:
Crysis3: 51.3--->52.7 2.7%
FarCry4: 58.9--->62 5.26%
Witcher3: 48.8--->51.4 5.3%
Fallout4: 42.7--->55.4 29.74%
ROTR: 48.1--->51 6%
Hitman: 66.4--->70.5 6.17%
DOOM: 75.6--->92.4 22.2%
DE:MD: 46.1--->46.7 1.3%
GoW4: 62.1--->67.5 8.6%
WD2: 32.2--->34.2 6.2%


So over the course of 2 yrs of driver optimization, Fury X gained an average of 9.347%


Adding to that is my test from launch driver to latest driver. Fire-strike Extreme.

Release driver:
http://www.3dmark.com/fs/5488050

Current driver:
http://www.3dmark.com/fs/12815443

4.9% increase.

This is just to give a glimpse of what driver optimization AMD has conjured up for FuryX over the course of 2 yrs. This result can serve as a guide line to show how much we should expect out of driver optimizations for other AMD flagship GPUs.

Of course RX480/580 or 290/390 and etc are welcome as well. As long as it is real data, not speculation.

 

[MrGenius]

Well...it depends on how bad the launch driver is optimization wise. And/or if it has any critical performance features missing/malfunctioning/disabled/not turned on. Meaning it's not always just about "optimization". Drivers can be "broken" too.

Case in point.

AMD recently disabled async compute on GCN 1.0 with drivers 16.4.2 through 17.1.2. Which made a significant difference for GCN 1.0 cards in DX12. I have evidence of that to show you.

Here's what my 280X did in Time Spy with driver 16.11.5.

Here's what my 280X did with driver 17.3.1.

That's a 4.05% increase in DX12 performance(based on graphics scores) despite a 13MHz decrease in core clock. Just from using a driver with async compute enabled instead of a driver with async compute disabled. I guess I can't say for sure it's "just from async compute being enabled", but that's likely a contributing factor.

 

[xkm1948]

Well...it depends on how bad the launch driver is optimization wise. And/or if it has any critical performance features missing/malfunctioning/disabled/not turned on. Meaning it's not always just about "optimization". Drivers can be "broken" too.

Case in point.

AMD recently disabled async compute on GCN 1.0 with drivers 16.9.2 through 17.1.2. Which made a significant difference for GCN 1.0 cards in DX12. I have evidence of that to show you.

Here's what my 280X did in Time Spy with driver 16.11.5.

Here's what my 280X did with driver 17.3.1.

That's a 3.57% increase in DX12 performance despite a 13MHz decrease in core clock. Just from using a driver with async compute enabled instead of a driver with async compute disabled.

No matter the jump in performance regarding features being turned on or off. The end result is you gained performance in that specific application over the progression of driver optimization. So yeah I would still consider that as improvement.


Basically for all that we care is old initial driver versus most recent driver.

Do you have any FireStrike Extreme, or Unigine benchmark results? I would love to see how how driver optimization has worked for DX11 based games, which still hold the majority of games.

 

[Tatty_Two]

Does this not also apply to NVidia, would be nice if there was a comparison, many think/assume/believe/know that AMD start at an inferior point on initial GPU release but gain huge ground, be interesting to see how their improvements scale with the other sides........ Ohhhhh and try not to keep double/triple posting, I like to keep my cupboards clean!

 

[EntropyZ]

I would love to see how how driver optimization has worked for DX11 based games, which still hold the majority of games.

Haha, AMD was never big on DX11 optimizations, they refused to add that extra performance for some reason though, and it's not clear what it was. The only titles that ran well enough we're the ones having "Gaming Evolved" stuff.

Fallout 4 HAS to be the worst example at it's launch, the RX 300 series cards stuttered like hell on Ultra settings (which are unrealistic unless tweaked anyway). Even older GeForce cards were running that particular game better.

It's a fail IMO it was one of the most anticipated game at the time and AMD was like, meh. Sure, gameworks happened, but I remember playing modded on high settings w/ R9 380 (you can argue this card was nowhere meant to run AAA games @ very high settings and also mods popped on top), it was not pretty, I can tell you that. They optimized the performance a lot later, but the damage was already done.

Maybe AMD was betting everything on DX12 and the future, nobody except them know. Even though DX11 isn't going anywhere as it was said by Microsoft that DX12 isn't replacing DX11, they work alongside.

Balls, rambing again, I need sleep.

 

[xkm1948]

Does this not also apply to NVidia, would be nice if there was a comparison, many think/assume/believe/know that AMD start at an inferior point on initial GPU release but gain huge ground, be interesting to see how their improvements scale with the other sides........ Ohhhhh and try not to keep double/triple posting, I like to keep my cupboards clean!

Well it is posted in an ATi sub forum. I would not mind having nVidia results as well. Especially 980Ti and 1080, since they are 2yrs and 1 yr old now respectively.

Yeah we will keep it as clean as possible. Hopefully we can have some civil discussion here.

Haha, AMD was never big on DX11 optimizations, they refused to add that extra performance for some reason though, and it's not clear what it was. The only titles that ran well enough we're the ones having "Gaming Evolved" stuff.

Fallout 4 HAS to be the worst example at it's launch, the RX 300 series cards stuttered like hell on Ultra settings (which are unrealistic unless tweaked anyway). Even older GeForce cards were running that particular game better.

It's a fail IMO it was one of the most anticipated game at the time and AMD was like, meh. Sure, gameworks happened, but I remember playing modded on high settings w/ R9 380 (you can argue this card was nowhere meant to run AAA games @ very high settings and also mods popped on top), it was not pretty, I can tell you that. They optimized the performance a lot later, but the damage was already done.

Maybe AMD was betting everything on DX12 and the future, nobody except them know. Even though DX11 isn't going anywhere as it was said by Microsoft that DX12 isn't replacing DX11, they work alongside.

Balls, rambing again, I need sleep.

Emm, so you have any results to share?

 

[MrGenius]

I don't have any launch driver benchmark results to compare to later driver results. I don't even know what version the launch driver was for my 280X. I do have two interesting DX11 results from Heaven and Valley. Going from 15.3 to 15.7 I got a 3.49% increase in Valley. And a 0.76% increase in Heaven even with a decrease of 9MHz core clock.

Here's a post with my Valley screenshots for proof.
https://www.techpowerup.com/forums/...-benchmark-scores.183712/page-37#post-3314942

Here's a post with my Heaven screenshots for proof.
https://www.techpowerup.com/forums/...-benchmark-scores.198888/page-43#post-3314854

So yeah. Even pretty late in the game my 280X was still seeing driver optimizations. It doesn't seem to be the case as of late. In fact my scores in Superposition and Time spy have gone down slightly with the latest drivers. 17.3.1 seems to be the best driver for Time Spy. 17.4.1 seems to be the best driver for Superposition. I'll edit in some screenshots for both to compare to the latest drivers. I want to run them both again to verify.

Superposition -0.82% from version 17.4.1 to 17.6.2

Time Spy -1.01% graphics score from version 17.3.1 to 17.6.2

What that's telling me so far is driver optimization has already peaked for my 280X and the latest drivers are now hindering performance. Probably resulting in peak DX11 performance with 17.4.1 and peak DX12 performance with 17.3.1. I'd need to do more research to confirm that. But preliminary results clearly show that driver optimization has ended for my 280X. It appears to be all down hill from here on out. But who knows. As long as they keep releasing new drivers that are compatible with it I'll keep trying them.

 

[erocker]

I suppose when you stick with one architecture for so long, you're bound to improve it with drivers with time. It's really made my 7970 a lot more serviceable.

 

[Sasqui]

Alright this time we talk about old AMD GPUs and how much driver optimization they got over the course of their life span.

So, take a deep breath and remain calm. No name calling, no cursing and no your own predictions.

This time we use only data and we only discuss data.


So I will lead with a review done by HardOCP specifically on how driver optimization helped FuryX and RX480.


link:
https://www.hardocp.com/article/2017/01/30/amd_video_card_driver_performance_review_fine_wine/1


Summary, check the source for yourself as well.

FuryX:
Crysis3: 51.3--->52.7 2.7%
FarCry4: 58.9--->62 5.26%
Witcher3: 48.8--->51.4 5.3%
Fallout4: 42.7--->55.4 29.74%
ROTR: 48.1--->51 6%
Hitman: 66.4--->70.5 6.17%
DOOM: 75.6--->92.4 22.2%
DE:MD: 46.1--->46.7 1.3%
GoW4: 62.1--->67.5 8.6%
WD2: 32.2--->34.2 6.2%


So over the course of 2 yrs of driver optimization, Fury X gained an average of 9.347%


Adding to that is my test from launch driver to latest driver. Fire-strike Extreme.

Release driver:
http://www.3dmark.com/fs/5488050

Current driver:
http://www.3dmark.com/fs/12815443

4.9% increase.

This is just to give a glimpse of what driver optimization AMD has conjured up for FuryX over the course of 2 yrs. This result can serve as a guide line to show how much we should expect out of driver optimizations for other AMD flagship GPUs.

Of course RX480/580 or 290/390 and etc are welcome as well. As long as it is real data, not speculation.

The fact there was a relatively small gain over 2 years speaks a lot to the team that realeased the original drivers. I'm curious how NV compares

 

[xkm1948]

I suppose when you stick with one architecture for so long, you're bound to improve it with drivers with time. It's really made my 7970 a lot more serviceable.

Yeah AMD seems to have been on GCN like, forever.

The fact there was a relatively small gain over 2 years speaks a lot to the team that realeased the original drivers. I'm curious how NV compares

Curious as well. Waiting for some nvidia users' data as input.

Really it would be super nice if that is something @W1zzard can do. He is browsing here as we type after all.

 

[Xzibit]

Does this not also apply to NVidia, would be nice if there was a comparison, many think/assume/believe/know that AMD start at an inferior point on initial GPU release but gain huge ground, be interesting to see how their improvements scale with the other sides........ Ohhhhh and try not to keep double/triple posting, I like to keep my cupboards clean!

They also have one on Nvidia

HardOCP - NVIDIA Video Card Driver Performance Review

NVIDIA Driver Performance Over Time

Simply put we did not see a lot of NVIDIA driver performance improvements over time on either video card. When we did experience driver performance gains over time the actual percentage difference was very small, generally well under 5%. There was the occasional game where a bug fix, or new game support was added that boosted performance. However, from "Game Ready" driver to "Game Ready" driver, the improvements were nil to none.

 

[dorsetknob]

this time we talk about old AMD GPUs and how much driver optimization they got over the course of their life span.

Curious as well. Waiting for some nvidia users' data as input.

Nvidia owners are probably respecting your request and to avoid Flameing/trolling accusations

happy for this answer to be here i hope

 

[kruk]

Here is the RX Vega driver optimization reality: 110% of unknown is still unknown .

 

[RejZoR]

People still going on only about framerates, entirely ignoring framerate consistency. What good is 250fps if everything is jerky because the frametimes are all over the place?

 

[Fluffmeister]

Don't worry, I still remember the outrage caused by "the 7950 is faster but the GTX 660 Ti is smoother".

It was beautiful.

 

[Octopuss]

Fallout 4 HAS to be the worst example at it's launch, the RX 300 series cards stuttered like hell on Ultra settings (which are unrealistic unless tweaked anyway). Even older GeForce cards were running that particular game better.

It's a fail IMO it was one of the most anticipated game at the time and AMD was like, meh. Sure, gameworks happened, but I remember playing modded on high settings w/ R9 380 (you can argue this card was nowhere meant to run AAA games @ very high settings and also mods popped on top), it was not pretty, I can tell you that. They optimized the performance a lot later, but the damage was already done.

Maybe AMD was betting everything on DX12 and the future, nobody except them know. Even though DX11 isn't going anywhere as it was said by Microsoft that DX12 isn't replacing DX11, they work alongside.

What makes you believe it's 100% AMD's fault? How about a hypothetical situation where Bethesda did a crappy programming job, and/or possibly optimized the engine for Nvidia only, or something like that?
I still remember that post floating around the internet where a guy who was once working at Nvidia talked about how fucked up the code many/most even AAA devs produce.

 

[Vayra86]

Don't worry, I still remember the outrage caused by "the 7950 is faster but the GTX 660 Ti is smoother".

It was beautiful.

http://techreport.com/review/23981/radeon-hd-7950-vs-geforce-gtx-660-ti-revisited/4

Here it is, good times.

 

[bug]

So, talking number only, we have 8 titles with insignificant changes (less than 10% is not noticeable outside benchmarks) and 2 titles with significant improvements. Of these, one was and remained below 60fps and one was already into playable realm and stayed there.

I'll take any improvement, especially if it's free, yet I fail to see anything special about AMD.

 

[EntropyZ]

What makes you believe it's 100% AMD's fault? How about a hypothetical situation where Bethesda did a crappy programming job, and/or possibly optimized the engine for Nvidia only, or something like that?
I still remember that post floating around the internet where a guy who was once working at Nvidia talked about how fucked up the code many/most even AAA devs produce.

I didn't say it was only their fault, as a matter of fact the devs (Bethesda Game Studios, not to be confused with the publisher Bethesda) know nothing about optimization, that old rebranded game engine from early 2000's that can't even multi-thread properly is also the reason for the performance being so low. What I mentioned was one of few cases. I bet the only hardware for developing they have is Intel/Nvidia based, since having that combo nets you a lot higher performance. The benchmarks show that, and also I should know when I converted the AMD Phenom II X6 960T to an i5 2400.

It's too bad that it's (not) in developer's interest to keep up with the hardware updates/changes. Since optimization on their part nets greater efficiency than a driver update, they only do so much. (The Ryzen Tomb Raider patch is a great example of developers doing something good even though it's way past the launch of the game, this is less than 1% of the game industry doing that, which is a great shame. It's better for the publishers to just make everyone move on to the "next best thing", but enough about that.)

 

[Th3pwn3r]

http://techreport.com/review/23981/radeon-hd-7950-vs-geforce-gtx-660-ti-revisited/4

Here it is, good times.

The amount of toxicity there rivals these AMD/Vega threads

 

[jboydgolfer]

I suppose when you stick with one architecture for so long, you're bound to improve it with drivers with time

 

[TheoneandonlyMrK]

Does this not also apply to NVidia, would be nice if there was a comparison, many think/assume/believe/know that AMD start at an inferior point on initial GPU release but gain huge ground, be interesting to see how their improvements scale with the other sides........ Ohhhhh and try not to keep double/triple posting, I like to keep my cupboards clean!

I agree but its also a couple of games ,who knows how they were picked but i suppose its something.
Also the generational differences were small , vegas difference is so great optimisation could benefit more and show.
Like the hd 5### series got big gains as did the 7### once they sussed their tech

 

[Vayra86]

The amount of toxicity there rivals these AMD/Vega threads

Nah it shows how gamers are such great social people.

Probably the reason they game