Low 2D shapes performance with NVidia cards

[SilverWS602]

SLI only works with a small number of 3D applications (the big name games in general). There is a larger number of 3D apps for which SLI does nothing. Having said that we are putting in some work for PerformanceTest V7 to try and get SLI and CrossFire to accelerate the 3D tests. But nvidia and AMD (ATI) aren't really interested in helping because we aren't selling a 100 million dollar game.

Why, oh why does that not suprise me? Also because your not 3dmark -- which is the only benchmark nvidia seems to care about. They release drivers just to make the score go up in that! (175.70 was released to increase vantage scores).

Personally I just use PT to compare my system vs my friends and to give me a baseline idea as to how much an improvement and upgrade made. Does it stink a little that it isn't the greatest at 3d tests and can't measure sli/crossfire? Sure it does. Does it REALLY REALLY matter? No, not really. We have standalone tests to do that.

[David (PassMark)]

Someone I missed the final few posts a few weeks ago. But I thought I should just add a reply to MasterX.

SLI never effects 2D performance (at least not in a positive fashion). So comparing 2D performance with SLI on or off, and expecting to see a difference doesn't make any sense.

You guys tell me....... When SLI is engages the 3D performance should gain 50-90% performance!

We never claimed this. I don't think even nvidia would be silly enough to make this blanket claim.

What MasterX is claiming also doesn't make much sense. To go from 70FPS to 150FPS, 114% improvement, with SLI, doesn't match with other results I have seen reported on the web. And doesn't make logical sense.

SLI only works with a small number of 3D applications (the big name games in general). There is a larger number of 3D apps for which SLI does nothing. Having said that we are putting in some work for PerformanceTest V7 to try and get SLI and CrossFire to accelerate the 3D tests. But nvidia and AMD (ATI) aren't really interested in helping because we aren't selling a 100 million dollar game.

[Windsurf48]

I do not see your point. SLI does nothing for 2d ...

I like your low 2d shapes test, a TNT2 would smoke it. Nice broken driver.

As far as I've been able to tell, all 2D benchmarks give increasing poor results using increasingly powerful 3D video cards, at least under the 32-bit and 64-bit verions of Windows XP SP2 and 32-bit XP with the latest SP3 release candidate. However, 32-bit and 54-bit Vista give impressive 2D scores with nVidia Geforce 8800 GT on the same benchmarks, with all the hardware the same and just the operating system changing. 32-bit Vista SP1 doesn't seem to show an improvement over plain 32-bit Vista, but still gets 2D scores which seem appropriate relative to the 3D scores in Performance Test.

2D scores are also better in PCMark05 and both PCPitstop benchmarks using Vista; there are comments on this in the PCPitstop forums Discuss Your Score section, which was the first place where I found references to the improvements in 2D scores under Vista as compared to those under XP.

Previously, there seemed to be a lot of messages indicating that the later nVidia cards had poor 2D results in one benchmark or another because the hardware was oriented towards 3D performance, but the fact that there are huge jumps in 2D performance going from XP to Vista on the same machine seems to discredit that.

I haven't come across any explanation for the difference so far.

[wonderwrench]

Do not use PT to test SLI performance, its that simple. SLI is more smoke and mirrors than anything else. If it worked as it should it would increase performance in all full screen 3d apps and give nearly double the performance over a single card. This of course would be dependent on having a CPU(s) with enough power to not cause a bottle neck.

Performance test is a general synthetic benchmark program to allow the user to compare overall performance between systems or gauge changes made to a given system. The 3d bench capability of PT is its only real weak spot. This could be overcome to some degree but I think Passmark realizes there is no truly accurate synthetic 3d benchmark as performance is game dependent.

OK, i realized a new version was released for PT, so i downloaded it and it scores a tad worse then 1008 did!

I ran all these tests just now so there fresh and from the same system load!

OMG does this say a lot for the validity of PT

I guess when you dont know why something doesnt quite work right blame it on the other guy....
the latest nvidia drivers 174.53 for windows xp actually gave me 5-8 FPS gain in games!

There getting there shit wired in my opinion, still would be nice to have a non-flickering control panel app tho!

And As of today i have THE FASTEST AMD CPU Retailing and the FASTEST AM2 NForce 590 motherboard
Being:
AMD Phenom 9600+ Black Edition with unlocked multipliers
ASUS M2N32 SLI Deluxe with 2 TRUE FULL 16X pci-e slots for 32X SLI

[wonderwrench]

I do not see your point. SLI does nothing for 2d and only effects full screen 3d and only when the application is designed to use SLI. Performance test was not made to test SLI performance. Why, because SLI can increase, decrease or make no difference in performance depending on the application AKA game or bench test used. It's best left out of the mix. A user of a SLI setup needs to gauge performance on a game by game basis.

I like your low 2d shapes test, a TNT2 would smoke it. Nice broken driver.

Dont go there yet, performance test has a quirk somewhere because heres a few scores that make no damn sense!

graphicstest-splitframe is when i have SLI turned on!
mx-quadcore-phenom is when i have SLI Turned off, single card rendering!

Theres almost no difference, which means performance test is not working correctly!!!!!!!!



My Specs:
AMD PHENOM QUAD CORE 9600+ BLACK EDITION (2.64Ghz)
ASUS M2N32 SLI Deluxe with 32X PCI-E lanes - Fastest AM2 Board!
4 GB Patriot DDR2 800 low latency (4-4-4-12)
2 X BFG Tech GeForce 7900 GT-OC (with bridge, not the cheapo bridgeless)
2 x Seagate baracuda 160GB in raid 0
Everything watercooled!!!!

Desktop Resolution: 1680x1050x60hz
Graphics drivers: 174.53
Perf Test version: 6.1 (build 1008 )
Windows XP SP2 x32
All unneeded processes trimmed!
Defragged last wednesday

Performance Test Settings:
Disk Drive under test: C:
CD Drive under test: F:
Test Durration: 3
Computer name: mx-quadcore-phenom
supress warning: 0
number of processes: 4 (1 cpu x 4 cores - quad core)
3d graphics: max refresh rate

You guys tell me.......
When SLI is engages the 3D performance should gain 50-90% performance!

Call of Duty 4 single card i get 20-40 FPS @ 1440x900x75hz
Call of Duty 4 SLI enabled I get 40-70 FPS @ 1440x900x75hz
Battlefield 2 single card i get 60-80 FPS @ 1280x900x80hz
Battlefield 2 SLI enabled I get over 150 FPS @ 1280x900x75hz

Whats the deal?

[MasterX]

OK, i realized a new version was released for PT, so i downloaded it and it scores a tad worse then 1008 did!

I ran all these tests just now so there fresh and from the same system load!

OMG does this say a lot for the validity of PT

I guess when you dont know why something doesnt quite work right blame it on the other guy....
the latest nvidia drivers 174.53 for windows xp actually gave me 5-8 FPS gain in games!

There getting there shit wired in my opinion, still would be nice to have a non-flickering control panel app tho!

And As of today i have THE FASTEST AMD CPU Retailing and the FASTEST AM2 NForce 590 motherboard
Being:
AMD Phenom 9600+ Black Edition with unlocked multipliers
ASUS M2N32 SLI Deluxe with 2 TRUE FULL 16X pci-e slots for 32X SLI

[MasterX]

Dont go there yet, performance test has a quirk somewhere because heres a few scores that make no damn sense!

graphicstest-splitframe is when i have SLI turned on!
mx-quadcore-phenom is when i have SLI Turned off, single card rendering!

Theres almost no difference, which means performance test is not working correctly!!!!!!!!



My Specs:
AMD PHENOM QUAD CORE 9600+ BLACK EDITION (2.64Ghz)
ASUS M2N32 SLI Deluxe with 32X PCI-E lanes - Fastest AM2 Board!
4 GB Patriot DDR2 800 low latency (4-4-4-12)
2 X BFG Tech GeForce 7900 GT-OC (with bridge, not the cheapo bridgeless)
2 x Seagate baracuda 160GB in raid 0
Everything watercooled!!!!

Desktop Resolution: 1680x1050x60hz
Graphics drivers: 174.53
Perf Test version: 6.1 (build 1008 )
Windows XP SP2 x32
All unneeded processes trimmed!
Defragged last wednesday

Performance Test Settings:
Disk Drive under test: C:
CD Drive under test: F:
Test Durration: 3
Computer name: mx-quadcore-phenom
supress warning: 0
number of processes: 4 (1 cpu x 4 cores - quad core)
3d graphics: max refresh rate

You guys tell me.......
When SLI is engages the 3D performance should gain 50-90% performance!

Call of Duty 4 single card i get 20-40 FPS @ 1440x900x75hz
Call of Duty 4 SLI enabled I get 40-70 FPS @ 1440x900x75hz
Battlefield 2 single card i get 60-80 FPS @ 1280x900x80hz
Battlefield 2 SLI enabled I get over 150 FPS @ 1280x900x75hz

Whats the deal?

[wonderwrench]

The latest 174.74 beta drivers still kill 2d shapes. Tested in XP pro sp2 32 bit running a 8800GT. I guess Nvidia is too dumb to fix this or just does not give a crap. ATI here I come!

[David (PassMark)]

I'll send off a couple of E-mails to the contacts we have and see if we get any answer. Don't hold your breath however.

Just an update on this. I did get an acknowledgement from NVidia saying that they agreed it was strange and they would "send around a few [internal] E-mails to if anyone can help".

That was more than two months ago now. There was never any further response.

[wonderwrench]

Well thanks for trying I really appreciate it. I have known about this bug since the first 1xx.xx drivers came out. Its been so long I can't remember when that was for sure. I have worn my fingers to the bone posting about this problem everywhere I can think of. It would seen most owners of Nvidia cards do not care 2d is broken or are blind to the problem. I first thought the 1xx.xx drivers were only messed up on 6 series cards as thats all I had for testing. A few months ago I got a guy to do some testing on his 7600gt and he found the same problem. Seems the average PC enthusiast cares nothing about 2d performance. They just want a higher 3Dmark score.

BTW I have owned Performance Test and Burnin Test since version 1. They have proven to be some of the most valuable tools I have ever purchased.
I was a beta tester at one time but I must have gotten lost in the shuffle.
If you ever need a dedicated beta tester let me know.

Thanks for the help and the great tools.

[David (PassMark)]

We confirmed the problem on a P4 with a NVidia 6600GT card here,

With old drivers (Forceware Version 84.21) the performance in the unfilled shapes test was good. With new drivers (Version 169.21) the result drops dramatically. More than a 10x loss of performance on this one test. Other tests are not really effected. I can visually see the drop in performance it is so bad.

Seems to definitely be a device driver bug.



Our past attempts to report bugs to NVidia have been a waste of time, but I'll send off a couple of E-mails to the contacts we have and see if we get any answer. Don't hold your breath however.

[wonderwrench]

Update this bug effects 6, 7 and 8 series cards. I just picked up a EVGA 8800GT SSC and the 2d shapes score is 343% slower than my old 6800GT was using 94.24 drivers. I cant test 9x.xx drivers on my 8800GT because the drivers do not support the 8800GT. All other 2d tests are way faster on the 8800GT as they should be. Same resolution bit depth and refresh rate.

BTW Passmark do you have any pull with anyone at Nvidia? as they must be blind to the problem. Question could 1xx.xx drivers have this problem because device bitmap caching is disabled for some reason?

[wonderwrench]

Your not the only one with problems. I'm running a EVGA 6800GT 256 meg AGP card and any 1xx.xx driver kills Graphics 2D - Shapes.
with driver version 94.24---30
with driver version 163.71--1

The only down side is if a new game has compatibility issue's with an old driver you will be stuck with a unplayable game or weak 2d performance.

I wonder if this bug only effects 6xxx series cards?

XP pro sp2 32 bit, latest updates

[strafe]

Thank you for your reply.

What am I loosing if I stick with 93.71 drivers instead of new ones.

Any suggestions?

[David (PassMark)]

For the 2D tests, there is no attempt to 'detect' the hardware in use. The same commands are sent to all video cards. (In fact we use the Windows O/S GDI API to draw the shapes for this test)

A result of 3.7 is low. But we have seen other 64bit machines with NVidia cards also do badly. For many many years now NVidia have been struggling to produce stable drivers, they release drivers without enough testing (recently their problems have been with Vista, 64bit and the 8800 cards). So it woudn't surprise me at all if they have broken something in their recent driver releases.