Go Back   nV News Forums > Hardware Forums > Benchmarking And Overclocking

Newegg Daily Deals

Reply
 
Thread Tools
Old 05-21-03, 01:42 AM   #325
StealthHawk
Guest
 
Posts: n/a
Default

Quote:
Originally posted by zakelwe
Hello Stealthawk,

I did not pick that one specifically when I started my testing, but it was one of the ones I chose because it was always the fastest due to issues with at least GT1.

Here is a link to my total runs, it's an excel spreadsheet and no macros but please virus check it if you download just to make sure.

http://www.aocb77.dsl.pipex.com/runs.xls

As you can see I started off with 40.72's which were always pretty fast, my 40.52's had corrupted. I did lots of drivers 4 times each and averaged them out. I also ran 3dmark2001 Nature DX8 test as a sort of baseline.

As you can probably guess it took me a while to do and was extremely boring. At least the chick with the sword was cute to look at for the first 15 times ( joke )

I'm happy to answer any questions and some people might read my results differently, in which case if they seem more logical i will change my viewpoint

Regards

Andy
Zakelwe,

I think that you should take note that 42.68 was one of the first drivers that was "optimized" for 3dmark03. It did this by lowering precision in GT4.

Discussion
Discussion

NV30 was scoring around 3000 points, and this driver boosted scores up to around 5000. The argument many people are presenting is that nvidia has solved the precision problems, and retained the high score in newer drivers 43.51+, but they have done this by substituting cheats, by adding in the clipping planes instead of lowering the precision.

In other words, you are comparing a cheat driver with another cheat driver, which is hardly the way to find out whether or not nvidia is cheating in the DetFXs. Every single one of nvidia's high performing 3dmark03 drivers has something questionable about it(either lowered precision or clipping planes).

If you want to compare drivers that don't cheat, I suggest you ask for a comparison with drivers earlier than 42.68, or driver 43.03. 43.03 fixed the precision problem, but it lowered scores back to the pre-cheat drivers.
  Reply With Quote
Old 05-21-03, 02:34 AM   #326
ChrisRay
Registered User
 
ChrisRay's Avatar
 
Join Date: Mar 2003
Location: Tulsa
Posts: 5,101
Default

Quote:
Originally posted by StealthHawk
Zakelwe,

I think that you should take note that 42.68 was one of the first drivers that was "optimized" for 3dmark03. It did this by lowering precision in GT4.

Discussion
Discussion

NV30 was scoring around 3000 points, and this driver boosted scores up to around 5000. The argument many people are presenting is that nvidia has solved the precision problems, and retained the high score in newer drivers 43.51+, but they have done this by substituting cheats, by adding in the clipping planes instead of lowering the precision.

In other words, you are comparing a cheat driver with another cheat driver, which is hardly the way to find out whether or not nvidia is cheating in the DetFXs. Every single one of nvidia's high performing 3dmark03 drivers has something questionable about it(either lowered precision or clipping planes).

If you want to compare drivers that don't cheat, I suggest you ask for a comparison with drivers earlier than 42.68, or driver 43.03. 43.03 fixed the precision problem, but it lowered scores back to the pre-cheat drivers.

This is only with Game Test 4, They doubled the performance in Game Test 2 and 3, as well. I think he's mentioning that as well
__________________
|CPU: Intel I7 Lynnfield @ 3.0 Ghz|Mobo:Asus P7P55 WS Supercomputer |Memory:8 Gigs DDR3 1333|Video:Geforce GTX 295 Quad SLI|Monitor:Samsung Syncmaster 1680x1080 3D Vision\/Olevia 27 Inch Widescreen HDTV 1920x1080

|CPU: AMD Phenom 9600 Black Edition @ 2.5 Ghz|Mobo:Asus M3n HT Deluxe Nforce 780A|Memory: 4 gigs DDR2 800| Video: Geforce GTX 280x2 SLI

Nzone
SLI Forum Administrator

NVIDIA User Group Members receive free software and/or hardware from NVIDIA from time to time to facilitate the evaluation of NVIDIA products. However, the opinions expressed are solely those of the members
ChrisRay is offline   Reply With Quote
Old 05-21-03, 02:45 AM   #327
Sazar
Sayonara !!!
 
Sazar's Avatar
 
Join Date: Jan 2003
Location: Austin, Texas
Posts: 9,297
Default

Quote:
Originally posted by ChrisRay
This is only with Game Test 4, They doubled the performance in Game Test 2 and 3, as well. I think he's mentioning that as well
yes.. quite...

the main problems existing prior to the clipping issue was precision in GT4... I think the general consensus was it was probably int12 or fp16... whereas the dx9 spec specifies a minimum precision of fp32...

the argument presented was since not ALL the parts of nature test were rendered using dx9 spec shaders et al... that nvidia could well have changed its code to use different precision in different parts of the test (I think thats what the argument was that was raised... I am just using common sense here)

the problem remained that the sky was being rendered with lower precision.. till det 44.03... and here the problem is the whole cliping issue...
Sazar is offline   Reply With Quote
Old 05-21-03, 02:52 AM   #328
ChrisRay
Registered User
 
ChrisRay's Avatar
 
Join Date: Mar 2003
Location: Tulsa
Posts: 5,101
Default

Quote:
Originally posted by Sazar
yes.. quite...

the main problems existing prior to the clipping issue was precision in GT4... I think the general consensus was it was probably int12 or fp16... whereas the dx9 spec specifies a minimum precision of fp32...

the argument presented was since not ALL the parts of nature test were rendered using dx9 spec shaders et al... that nvidia could well have changed its code to use different precision in different parts of the test (I think thats what the argument was that was raised... I am just using common sense here)

the problem remained that the sky was being rendered with lower precision.. till det 44.03... and here the problem is the whole cliping issue...

I'm aware of the issues with Game Test 4, I actually believe Nvidia cut corners on Game Test 2 and 3 as well. But as I have said many times. I don't care what ATI or Nvidia does with this benchmark
__________________
|CPU: Intel I7 Lynnfield @ 3.0 Ghz|Mobo:Asus P7P55 WS Supercomputer |Memory:8 Gigs DDR3 1333|Video:Geforce GTX 295 Quad SLI|Monitor:Samsung Syncmaster 1680x1080 3D Vision\/Olevia 27 Inch Widescreen HDTV 1920x1080

|CPU: AMD Phenom 9600 Black Edition @ 2.5 Ghz|Mobo:Asus M3n HT Deluxe Nforce 780A|Memory: 4 gigs DDR2 800| Video: Geforce GTX 280x2 SLI

Nzone
SLI Forum Administrator

NVIDIA User Group Members receive free software and/or hardware from NVIDIA from time to time to facilitate the evaluation of NVIDIA products. However, the opinions expressed are solely those of the members
ChrisRay is offline   Reply With Quote
Old 05-21-03, 02:53 AM   #329
zakelwe
Registered User
 
Join Date: Dec 2002
Posts: 768
Default

Quote:
Originally posted by StealthHawk
Zakelwe,

I think that you should take note that 42.68 was one of the first drivers that was "optimized" for 3dmark03. It did this by lowering precision in GT4.

Discussion
Discussion

NV30 was scoring around 3000 points, and this driver boosted scores up to around 5000. The argument many people are presenting is that nvidia has solved the precision problems, and retained the high score in newer drivers 43.51+, but they have done this by substituting cheats, by adding in the clipping planes instead of lowering the precision.

In other words, you are comparing a cheat driver with another cheat driver, which is hardly the way to find out whether or not nvidia is cheating in the DetFXs. Every single one of nvidia's high performing 3dmark03 drivers has something questionable about it(either lowered precision or clipping planes).

If you want to compare drivers that don't cheat, I suggest you ask for a comparison with drivers earlier than 42.68, or driver 43.03. 43.03 fixed the precision problem, but it lowered scores back to the pre-cheat drivers.
Hello Stealthawk,

I know they were the first to be optimised, Futuremark took all my scores down for a while . Your evolving cheat theory could be true, it had passed my mind.

However pleasse note that the increase for my card for 42.68 is not due to precision quality in GT4 as the GF4 card can not play this game.

For my card the increase is seen in 42.68, due to a lot of the action missing from GT1, ie bullets and explosions not there . So I thought that was to blame, but from my results it also shows that GT2 got a boost as well. That is why I asked Dave Baumann to say whether 42.68 also had this problem with GT2, ie clearing buffers properly and therefore 2 optimisations. If it does then it looks like two things working in conjunction, if it doesn't then the later cheat does not help, at least for the GF4 series.


Note that 42.68 is now accepted by Futuremark and so from their point of view 42.68 is OK .. which is strange in itself if it does not draw everything.

I guess more testing needs to be done as you say, including Gt1 scores to see how the various "problems" evolve. Shame I don't have a FX card

This is what the Extremetech article should have done to back up their pictures. Rather than being a Sunday Times investigation it turned out to be the equivalent of a Daily Star Scoop ...
rolleyes:

I shall report back, unless I go mad first.

Regards

Andy
zakelwe is offline   Reply With Quote
Old 05-21-03, 03:22 AM   #330
zakelwe
Registered User
 
Join Date: Dec 2002
Posts: 768
Default

Quote:
Originally posted by Sazar
yes.. quite...

the main problems existing prior to the clipping issue was precision in GT4... I think the general consensus was it was probably int12 or fp16... whereas the dx9 spec specifies a minimum precision of fp32...

the argument presented was since not ALL the parts of nature test were rendered using dx9 spec shaders et al... that nvidia could well have changed its code to use different precision in different parts of the test (I think thats what the argument was that was raised... I am just using common sense here)

the problem remained that the sky was being rendered with lower precision.. till det 44.03... and here the problem is the whole cliping issue...
So the total number of suggested optimisations made up to know boils down to :-

42.68

GT1 Missing events
GT2 uncleared buffers???
GT3 OK
GT4 Lower precision

44.03

GT1 OK ?
GT2 uncleared buffers
GT3 OK
GT4 clipping planes


I think I'll have a look at how Gt1 changes between the two tonight, just for my own interest. I get the feeling that 44.03 will be OK and therefore give a lower score.

Regards

Andy
zakelwe is offline   Reply With Quote
Old 05-21-03, 03:26 AM   #331
zakelwe
Registered User
 
Join Date: Dec 2002
Posts: 768
Default

Please note that 42.01 shows the same value for GT3 as 42.68/ 43.51/43.51/44.03, it is only GT2 that shows a big increase and that is the game that is being questioned as well as GT4.

So it looks as if GT3, at least, has not been optimised for GF4 since 42.01.

Regards

Andy
zakelwe is offline   Reply With Quote
Old 05-21-03, 05:56 AM   #332
StealthHawk
Guest
 
Posts: n/a
Default

Quote:
Originally posted by zakelwe
Hello Stealthawk,

I know they were the first to be optimised, Futuremark took all my scores down for a while . Your evolving cheat theory could be true, it had passed my mind.

However pleasse note that the increase for my card for 42.68 is not due to precision quality in GT4 as the GF4 card can not play this game.

For my card the increase is seen in 42.68, due to a lot of the action missing from GT1, ie bullets and explosions not there . So I thought that was to blame, but from my results it also shows that GT2 got a boost as well. That is why I asked Dave Baumann to say whether 42.68 also had this problem with GT2, ie clearing buffers properly and therefore 2 optimisations. If it does then it looks like two things working in conjunction, if it doesn't then the later cheat does not help, at least for the GF4 series.
The gfFX cards and the DX8 generation cards are very different. Who knows what is going on at the driver level. Asking for benchmarks from a gfFX user doesn't really tell you anything about gf3/4, and vice versa.

I'm sure nvidia has some legitmate optimization, but who knows how much...that is the big question.

Quote:
Note that 42.68 is now accepted by Futuremark and so from their point of view 42.68 is OK .. which is strange in itself if it does not draw everything.
I thought only WHQL drivers were accepted. You can run non-WHQL drivers and compare your scores with others I think, but they are not officially entered into the Orb database, correct?

Quote:
I guess more testing needs to be done as you say, including Gt1 scores to see how the various "problems" evolve. Shame I don't have a FX card
As I understand it, the effects were not missing on an FX card, only on a gf3/4 card.
  Reply With Quote

Old 05-21-03, 05:56 AM   #333
StealthHawk
Guest
 
Posts: n/a
Default

Quote:
Originally posted by zakelwe
So the total number of suggested optimisations made up to know boils down to :-

42.68

GT1 Missing events
GT2 uncleared buffers???
GT3 OK
GT4 Lower precision

44.03

GT1 OK ?
GT2 uncleared buffers
GT3 OK
GT4 clipping planes


I think I'll have a look at how Gt1 changes between the two tonight, just for my own interest. I get the feeling that 44.03 will be OK and therefore give a lower score.

Regards

Andy
With the 44.03 drivers, I thought GT2 has clipping planes too. AFAIK there has been no cheating in GT1 or GT3 for FX cards.
  Reply With Quote
Old 05-21-03, 06:13 AM   #334
zakelwe
Registered User
 
Join Date: Dec 2002
Posts: 768
Default

Originally posted by StealthHawk [/i]
The gfFX cards and the DX8 generation cards are very different. Who knows what is going on at the driver level. Asking for benchmarks from a gfFX user doesn't really tell you anything about gf3/4, and vice versa.

True, but the claim for GT2 is that workload is being saved , so my assumption is that it should be saved for whatever card, whether it be GF4 or GF FX. GT2 does increase more than GT3 for the GF4 relatively for 42.68, 43.45, and 43.51. By fair means or foul ?

I'm sure nvidia has some legitmate optimization, but who knows how much...that is the big question.

I think so too.


I thought only WHQL drivers were accepted. You can run non-WHQL drivers and compare your scores with others I think, but they are not officially entered into the Orb database, correct?

I thought this too as I had lost my scores, but then I notcied some subsequent scores appearing using 42.68. So I emailed worm and he replied saying they were ok to submit. So I did. Check out the Orb for the GF4 4200 top scores in 3dmark03. Look at the driver versions and dates submitted for the top 2, Mr Icee and me.


As I understand it, the effects were not missing on an FX card, only on a gf3/4 card.

Don't know. But I want to include GT1 now just to see if, for the Gf4 card the problem is fixed and whether the scores go down or up from 42.68. If the problem is fixed and the scores do go up then this shows nVidia has been optimising , probably fairly as nonone has said there is a problem with GT1.

Regards

Andy
zakelwe is offline   Reply With Quote
Old 05-21-03, 06:14 AM   #335
ChrisRay
Registered User
 
ChrisRay's Avatar
 
Join Date: Mar 2003
Location: Tulsa
Posts: 5,101
Default

Quote:
The gfFX cards and the DX8 generation cards are very different. Who knows what is going on at the driver level. Asking for benchmarks from a gfFX user doesn't really tell you anything about gf3/4, and vice versa.

I'm sure nvidia has some legitmate optimization, but who knows how much...that is the big question.
I am quite curious what you consider legitimate? I mean ATI optimises its drivers for 3dmark quite a bit as well. Cat 3.4 didn't really improve my over all benchmark scores in anything but 3dmark, Which I won't use.

They did improve FSAA performance in 4x, Nothen changed in 2x or 6x tho,

But the overall ATI does optimise its drivers for 3dmark, What do you consider legitimate? When does optimisation defeat the purpose of the benchmark?

When is optimising for this benchmark cheating? If so, Why is optimising for a benchmark thats supposed to provide generic results for DX 9.0 standards good?

I think you see where I'm going, I don't see any optimisation for this benchmark as a good thing, How do we know these improvements affect anything but 3dmark?

9 times out 10 any improvement from either company that affects 3dmark does not help any games. I don't think anyone can list one game that optimisation for the battle proxymon has helped.

See where I'm going? The purpose of this benchmark was defeated long ago. To Measure Generic DX 9.0 performance. How can you measure this accurately when both companies optimise like crazy for this benchmark?

You can't. You simply can't. Therefore this benchmark simply cannot work as an accurate guide for measuring performance.

It's simply not even. It never will be. It simply cannot be, As long ATI and Nvidia need to look good
__________________
|CPU: Intel I7 Lynnfield @ 3.0 Ghz|Mobo:Asus P7P55 WS Supercomputer |Memory:8 Gigs DDR3 1333|Video:Geforce GTX 295 Quad SLI|Monitor:Samsung Syncmaster 1680x1080 3D Vision\/Olevia 27 Inch Widescreen HDTV 1920x1080

|CPU: AMD Phenom 9600 Black Edition @ 2.5 Ghz|Mobo:Asus M3n HT Deluxe Nforce 780A|Memory: 4 gigs DDR2 800| Video: Geforce GTX 280x2 SLI

Nzone
SLI Forum Administrator

NVIDIA User Group Members receive free software and/or hardware from NVIDIA from time to time to facilitate the evaluation of NVIDIA products. However, the opinions expressed are solely those of the members
ChrisRay is offline   Reply With Quote
Old 05-21-03, 06:19 AM   #336
zakelwe
Registered User
 
Join Date: Dec 2002
Posts: 768
Default

Quote:
Originally posted by StealthHawk
With the 44.03 drivers, I thought GT2 has clipping planes too. AFAIK there has been no cheating in GT1 or GT3 for FX cards.
I think it's non cleared buffers in GT2 but maybe that is related to clipping planes, at least that is what Extremetech stated. Agreed, nobody has mentioned GT1 or GT3 for cheating so I think they are both OK now as well.

Regards

Andy
zakelwe is offline   Reply With Quote
Reply


Thread Tools

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Forum Jump


Similar Threads
Thread Thread Starter Forum Replies Last Post
Bored, impressed, and giddy: Our final thoughts on E3 2012 (with photos) News Archived News Items 0 06-13-12 07:00 AM
Thoughts from console owners on NVIDIA's GEFORCE GRID MikeC Console World 11 05-27-12 09:43 AM
Looking for a good 21"/22" Monitor...any thoughts? Guuts General Hardware 13 09-22-02 12:04 PM
Thoughts on the command line as an interface. lunix Microsoft Windows XP And Vista 10 09-12-02 09:44 PM
GTA Thoughts? Typedef Enum Gaming Central 5 09-03-02 05:51 PM

All times are GMT -5. The time now is 07:41 PM.


Powered by vBulletin® Version 3.7.1
Copyright ©2000 - 2014, Jelsoft Enterprises Ltd.
Copyright 1998 - 2014, nV News.