is there a way to disable PS 1.4 support of 9700?

[DOOM III]

i want to see how 9700 performances without the help of PS 1.4

 

[Colourless]

In theory, yes there 'is' a way by creating a 3DAnalyser like Direct3D 9 interceptor dll (which would also disable PS2.0), BUT it will likely cause some other extra minor performance penalties as well due to extra function call overhead. As for some driver/registry based mothod, not as far as I know.

 

[depth_test]

I doubt it would change anything. The underlying HW is the same. The DX drivers probably compile both 1.4 and 2.0 shaders to the same underlying internal code.

 

[MDolenc]

But if DX sends three passes of ps_1_1 driver still can't produce one ps_1_4 out of it.

 

[Bambers]

There was registry option for the 8500 to set pixelshader versions. Knocked about 10fps off the advanced pixelshader score in 3dmark using 1.3 instead of 1.4. Older drivers though so I don't know the current version and if it works on r300s

 

[Ichneumon]

I haven't had a chance to test it yet, but yea there were settings in old ATI drivers to specify PS/VS support. My guess is they won't work with the 9xxx series, but i'll give it a shot and see. =)

 

[Ichneumon]

well i'll be damned... it works. (at least its reported by the details that it does)

testing performance to see if it shows a difference. back in a bit.

 

[Ichneumon]

3DMark DB is sloooow today. I'll post up detailed info in a little bit after i do one more test, and can get the ORB to lay things out as I want, but for preliminary info:

System:
AXP2600+
512 ram
ATI 9700 Pro

Pixel Shader 1.1 and Vertex Shader 1.1 forced:
3189 3DMarks

Pixel Shader 1.4 and Vertex Shader 1.1 forced:
3648 3DMarks

Will get the breakdowns shortly. I specifically forced 1.4 and 1.1 to take out the game 4 scores from the 3dmark... now if the ORB will stop takin its sweet time.

 

[Brent]

well i'll be damned... it works. (at least its reported by the details that it does)

testing performance to see if it shows a difference. back in a bit.

these settings are in the registry?

what key might i ask

 

[Ichneumon]

well i'll be damned... it works. (at least its reported by the details that it does)

testing performance to see if it shows a difference. back in a bit.

these settings are in the registry?

what key might i ask

If you know what to add...

However, on a serious note, gimme just a bit longer and i'll put a post together on all of it... still testing some things.

 

[Himself]

well i'll be damned... it works. (at least its reported by the details that it does)

testing performance to see if it shows a difference. back in a bit.

these settings are in the registry?

what key might i ask

If you are running XP it's going to be some random nonsense, so that wouldn't help you anyway. They are not in the registry as such, but you can add them and the drivers will read them.

 

[Ichneumon]

A basic comparison... Will do some more testing shortly.

Interesting results. 1.4 makes a difference but not a *huge* difference. I think scores from a 8500 would be more telling than the 9700 scores since the 9700 can absorb the additional work better. Will bench those a bit later tonight.

Going to get out the updated tweaker first.

Anyway here's the multi-compare link. It'll be slow to load (if the ORB isn't on break again... lazy bastard ).

 

[Ostsol]

Cool find, Ichneumon! 8)

 

[tb]

I'll release 3D-Analyze 2.0 tomorrow. There you can force the app to 1.1 or 1.4 shaders. You can also disable the two sided stencil test. If you use these features, you will see, how bad the overall 3dmark score computation is, because the only difference with a r9700 pro and 1.4 and 2.0 shaders is, that test 4 isn't executed and this has a huge impact on the overall scores:

AMD Athlon(tm) XP 1800+, WinXP, R9700 Pro (default clock)

forced to 1.4 shaders:

3DMark Score 3232 3DMarks

GT1 - Wings of Fury 112,9 fps
GT2 - Battle of Proxycon 30,4 fps
GT3 - Troll's Lair 27,2 fps
GT4 - Mother Nature 0,0 fps Not supported

2.0 shaders:

3DMark Score 4281 3DMarks

GT1 - Wings of Fury 113,0 fps
GT2 - Battle of Proxycon 30,5 fps
GT3 - Troll's Lair 27,3 fps
GT4 - Mother Nature 27,0 fps


Regards,
Thomas

 

[Doomtrooper]

that test 4 isn't executed and this has a huge impact on the overall scores

No different than 3Dmark 2001 and PS 1.1 Nature test... Geforce 3 cards were getting the same inflated scores vs Radeon, GTS, GF2 ULTRA and they should, its a DX8 test..this being a DX9 test is the same.

DX9 tests require DX9 hardware.

 

[Ichneumon]

The updated Rage3D Tweak with the force pixel/vertex shader tweaks is up on the download page.

 

[Ichneumon]

I'll release 3D-Analyze 2.0 tomorrow. There you can force the app to 1.1 or 1.4 shaders. You can also disable the two sided stencil test. If you use these features, you will see, how bad the overall 3dmark score computation is, because the only difference with a r9700 pro and 1.4 and 2.0 shaders is, that test 4 isn't executed and this has a huge impact on the overall scores

This is news?

3dm2k1 raped your score if you couldn't run the nature test... and only the GeForce3 could run it when 3dm2k1 came out.

Don't see what the difference is this time around... this time its simply that Nvidia doesn't have cards available for people to buy that can run Nature... and they're whining about it.

 

[Joe DeFuria]

because the only difference with a r9700 pro and 1.4 and 2.0 shaders is, that test 4 isn't executed and this has a huge impact on the overall scores:

Um, yes...the impact is roughly 20% hit in the final 3DMark score, which is what one would expect by reading the 3DMark white-paper on how they calculate the score. :?

This is how the 3DMark score "quantifies" the advantage for hardware for having advanced feature support.

Do you think 20% penalty target is really out of line? In other words, if Card A lacks DX9 support, but is otherwise 20% faster than Card B across the DX7 and DX8 benchmarks, the total 3DMark score will be about equal. Is that not reasonable?!

As Ichneumon said, this is no different than how 3DMark01 treated the Game 4 test wrt DX8 pixel shaders. (Though I don't know the exact percentage break-down off hand...)

 

[tb]

I'll release 3D-Analyze 2.0 tomorrow. There you can force the app to 1.1 or 1.4 shaders. You can also disable the two sided stencil test. If you use these features, you will see, how bad the overall 3dmark score computation is, because the only difference with a r9700 pro and 1.4 and 2.0 shaders is, that test 4 isn't executed and this has a huge impact on the overall scores

This is news?

3dm2k1 raped your score if you couldn't run the nature test... and only the GeForce3 could run it when 3dm2k1 came out.

Don't see what the difference is this time around... this time its simply that Nvidia doesn't have cards available for people to buy that can run Nature... and they're whining about it.

No, thats not new, but I had the hope they've learned from the past It's okay to include some points just for features, but this makes the 3dmark's useless for performance meassuring(Which is the main purpose of this program!). If they want scores for features, they could add up the dx9 caps + max shader versions, which would give a nice feature score....

3DMark 2k1 score = (Game1LowDetail + Game2LowDetail + Game3LowDetail) * 10 + (Game1HighDetail + Game2HighDetail + Game3HighDetail + Game4) * 20

game 4 impact: 18 %

3DMark03 score = (Game Test 1 frame-rate x 7.3) +
(Game Test 2 frame-rate x 37) +
(Game Test 3 frame-rate x 47.1) +
(Game Test 4 frame-rate x 38.7)

test 4 impact: 24 %

I see a lot of reviews on the net and even in print magazines, which uses the overall scores and that's simply wrong from a performance stand point.

P.S. The graphix and the music are well done. The single tests are nice and they show what's possible with dx9 hardware and a new card(R400+). So, you can take ATi / NVIDIA tech demos or/and 3DMark03... and many real gaming benchmarks.

Regards,
Thomas

 

[Ichneumon]

3DMark 2k1 score = (Game1LowDetail + Game2LowDetail + Game3LowDetail) * 10 + (Game1HighDetail + Game2HighDetail + Game3HighDetail + Game4) * 20

game 4 impact: 18 %

3DMark03 score = (Game Test 1 frame-rate x 7.3) +
(Game Test 2 frame-rate x 37) +
(Game Test 3 frame-rate x 47.1) +
(Game Test 4 frame-rate x 38.7)

test 4 impact: 30 %

I see a lot of reviews on the net and even in print magazines, which uses the overall scores and that's simply wrong from a performance stand point.

Regards,
Thomas

It isn't 30%.
It's more like 20%
This is specifically talked about in the whitepaper. The weights are set in such a way to give game test four roughly 20% of the score on current high end cards. The weights take into account frame rates that tests get across different hardware. Thats why such a low weight on GT1... otherwise with its Much higher FPS it would dominate the score.

It may not be an ideal system, but I definately understand the logic behind why they did it.

I'll still wager money that in a year the hierarchy of card performance shown by 3dmark03 scores from the aggregate of hundereds of thousands of benchmarks will show very close parity to exactly how you would line up cards by performance on your own.

3Dmark2k1 does this quite well right now going back 3 or 4 generations of hardware now. I expect the same will be the case for 3dmark03.

3DMark will never tell you how well a card will perform on a specific game, but I am impressed with its ability to line up cards in a performance hierarchy that quite closely reflects the reality of how the cards perform when looked at across a large number of games.

I also think it is a singularly excellent tool for testing an individual systems performance under different configurations/drivers/settings/etc for purposes of getting the most out of your system, but thats a little beside the point in this discussion.