Driver improvements: A quick look back to Sept 09 benchmarks for the 5850 and 5870

[dug777]

It's hard to believe the 5870 and 5850 are almost a year old

Anyway for sh1ts and giggles I decided to look back at the 5850 launch results and compare them to the most recent AT results. Just a quick and dirty comparison, of interest to me since I own a 5850

5850 launch Crysis: Warhead

http://www.anandtech.com/show/2848/4

AT's recent MSI 470 launch

http://www.anandtech.com/show/3836/msis-geforce-n470gtx-gtx-470-sli/4

Comparing the 5870, the single card improvement at 25x16 is 24.9fps to 25.9fps, a 4% increase. Nothing to see there, move along please

Now 5870 CF, 34.8fps to 40.8fps, a 17% increase. Not too shabby at all.

Comparing the 5850, the single card improvement at 25x16 is 21.2fps to 22.8fps, a 7.5% increase. Respectable I guess.

Now 5850 CF, 30.7ps to 38.1fps, a 24% increase. That's pretty good going :awe: It's faster than the 5870 CF at launch

Battleforge DX11 25x16:

EDIT: looks like those are not comparable

HAWX:

5870 13%
5870 CF 15%

5850 18%
5850 CF 14%

Some modest to reasonable gains there in places, but CF scaling doesn't look to have improved much, if at all...

 

[Will Robinson]

Interesting info there Duggy,with all the stick CrossfireX has been taking here recently you would've thought the opposite was true.:thumbsup:

 

[dug777]

Interesting info there Duggy,with all the stick CrossfireX has been taking here recently you would've thought the opposite was true.:thumbsup:

The problem I think is that CF scaling doesn't look like it has improved, apologies if you caught the post before I edited some epic mistakes

 

[Bill Brasky]

Definitely some respectable gains. I'm surprised a tech site hasn't done an actual driver comparison. Did you get your 5850 at launch, Dug? I still remember f5ing newegg like crazy trying to get an xfx. Good times.

 

[evolucion8]

Seems that the HD 4x00 series scaling is still better than the HD 5x00 series scaling. Could it be blamed for the optimizations that were removed from the die to keep the chip's size in check? Hopefully they will be back on S.I.

 

[Idontcare]

Definitely some respectable gains. I'm surprised a tech site hasn't done an actual driver comparison. Did you get your 5850 at launch, Dug? I still remember f5ing newegg like crazy trying to get an xfx. Good times.

That would definitely be a value-add review. For both gpu families.

It seems to me that it is the very fact that driver updates do have a non-zero impact on performance that it makes performance reviews and price/performance comparisons a perpetually moving target.

We don't really see this in the cpu world, bios releases for mobos and new cpu steppings tend to focus on improved compatibility/stability and at best some power-savings but performance itself changes very little over the lifetime of the product.

A Q6600 review from 3yrs ago is just about as relevant today as it was then, not so with the 5870/5850 reviews done at launch less than a year ago. (just my opinion)

 

[evolucion8]

That would definitely be a value-add review. For both gpu families.

It seems to me that it is the very fact that driver updates do have a non-zero impact on performance that it makes performance reviews and price/performance comparisons a perpetually moving target.

We don't really see this in the cpu world, bios releases for mobos and new cpu steppings tend to focus on improved compatibility/stability and at best some power-savings but performance itself changes very little over the lifetime of the product.

A Q6600 review from 3yrs ago is just about as relevant today as it was then, not so with the 5870/5850 reviews done at launch less than a year ago. (just my opinion)

IDC? Moderator? Congratulations!!

 

[Idontcare]

Thanks!

 

[GaiaHunter]

Thanks!

You sure he isn't saying "may you live in interest times"?

 

[KingstonU]

Woah! Good going IDC. Though when I think about it, people have been nominating you for a while. :thumbs up:

 

[Voo]

You sure he isn't saying "may you live in interest times"?

In the light of what happened with the last one and the whole atmosphere around here I think that goes without saying.. honestly who'd want to moderate this forum

Umn, I mean congratulations!

b2t: Would be an interesting review - especially if we could get some kind of timeframe comparisons.. i.e. what where the improvements for the 48xx series after 6 months compared to 58xx.
We already have some indicators that R800 has some performance penalties comapred to R700, so how does it look on the optimization side of things?

 

[evolucion8]

In the light of what happened with the last one and the whole atmosphere around here I think that goes without saying.. honestly who'd want to moderate this forum

Umn, I mean congratulations!

b2t: Would be an interesting review - especially if we could get some kind of timeframe comparisons.. i.e. what where the improvements for the 48xx series after 6 months compared to 58xx.
We already have some indicators that R800 has some performance penalties comapred to R700, so how does it look on the optimization side of things?

That explains why a single HD 5870 looses against the HD 4870X2, considering that Crossfire doesn't scale 100%, the fact that the HD 5770 with a 100MHz core advantage can't outperform the HD 4870 1GB, also the HD 4550 with a 50Mhz core deficit, outperforms the identical spec HD 5450. I think that some optimizations were removed in the Evergreen shaders to keep the die size in check.

 

[Idontcare]

That explains why a single HD 5870 looses against the HD 4870X2, considering that Crossfire doesn't scale 100%, the fact that the HD 5770 with a 100MHz core advantage can't outperform the HD 4870 1GB, also the HD 4550 with a 50Mhz core deficit, outperforms the identical spec HD 5450. I think that some optimizations were removed in the Evergreen shaders to keep the die size in check.

I wonder if they intended to really neuter the shaders that much or if it turned out they cut a little too much muscle out of the architecture without realizing how deep the cuts would run.

A 100MHz deficit to gain back the performance seems a little excessive to me, pretty big trade-off considering the power-consumption gain they have to make to clock it up that extra 100MHz.

Or maybe they were thinking the extra 100MHz clocking wouldn't be such a big deal because of the expectations of there being more than enough power-consumption savings coming with 40nm?

 

[evolucion8]

I wonder if they intended to really neuter the shaders that much or if it turned out they cut a little too much muscle out of the architecture without realizing how deep the cuts would run.

A 100MHz deficit to gain back the performance seems a little excessive to me, pretty big trade-off considering the power-consumption gain they have to make to clock it up that extra 100MHz.

Or maybe they were thinking the extra 100MHz clocking wouldn't be such a big deal because of the expectations of there being more than enough power-consumption savings coming with 40nm?

I think that's a balance between performance and die size, considering that the HD 5770 performs almost the same as the HD 4870 and runs much cooler with a much lower TDP, it ain't bad at all. The HD 5770 would be all the time slower than the HD 4870 without its 100Mhz core advantage, but considering that the 40nm brought great gains in that department, it doesn't matter much, a single HD 5870 consumes slightly less power than the HD 4890!!!

 

[cusideabelincoln]

I did the same analysis for the GTX480 SLI, 480, and 470 using AT's reviews:
older drivers: http://www.anandtech.com/show/2977/...x-470-6-months-late-was-it-worth-the-wait-/18
newer drivers: http://www.anandtech.com/show/3809/nvidias-geforce-gtx-460-the-200-king/15

I just used 2560x1600 resolution (saves time).

GTX 480 SLI average improvement: 7.01%
GTX 480 average improvement: 7.41%
GTX 470 average improvement: 4.48%

Game breakdown:

CRYSIS:
480 SLI improves 0.9% in averages
480 SLI improves 18.5% in mins
480 improves 12.3% in averages
480 improves 19.6% in mins
470 improves 12.7% in averages
470 improves 10.7% in mins

BATTLEFORGE DX10:
480 SLI improves 4%
480 improves 2.9%
470 improves 0%

BATTLEFORGE DX11:
480 SLI improves 2.4%
480 improves 2.75%
470 improves 0%

HAWX:
480 SLI improves 0%
480 improves 4.3%
470 improves 4.2%

LEFT 4 DEAD
480 SLI improves 0%
480 improves 5%
470 improves 5.1%

BAD COMPANY 2 CHASE:
480 SLI improves 2.2%
480 improves 13.4%
470 improves 14.1%

BAD COMPANY 2 WATERFALL:
480 SLI improves 14.9%
480 improves 14.7%
470 improves 6%

STALKER COP:
480 SLI decreases 1.7%
480 improves 4%
470 improves 2.5%

DIRT 2:
480 SLI improves 0%
480 improves 2.5%
470 improves 0%

MASS EFFECT 2:
480 SLI improves 21%
480 improves 7.5%
470 improves 6.5%

WOLFENSTEIN:
480 SLI improves 21.7%
480 improves 0%
470 improves 0%

 

[betasub]

Thanks cuside! Shows where Nvidia's efforts have been focused: improving SLI generally, and single card performance in a few headline games.