PDA

View Full Version : Detail DX9 shader performance for Fx5700


Hellbinder
10-23-03, 01:10 PM
From Dave at B3D..

http://www.beyond3d.com/forum/viewtopic.php?t=8623

It would appear from this lot that the configutation of 5700 Ultra in terms of its fixed function pipelines is the same as 5600 - 4x1 in single / no texturing circumstances, but 2x2 in any multitexturing case, so it seems like only two of the pipelines are able to loopback and have any shader functionality as well.

We can see that the DX9 PS2 performance of 5700 Ultra is significantly improved from 5600 Ultra, well beyond the clock speed increases in some cases. This would suggest that NVIDIA have done exactly as they did do with NV35 over NV30, and changed the integer units to smaller float ops that help out with common PS2 instructions (not that the simple PS2 test is only at difference over the 5600, which is their clock rate differences) - the difference being that this time the 52.16 drivers are better able to make use of this extra functionality, where the 44.03 drivers couldn't really show it with nv35 initially. Note, that the DX8 performance also stay inline with the clock rate increases.

Note that in the Fill-rate tester results a couple of the PS2.0 results take a dump in performance in comparison to the 5600 Ultra - the PS2.0 Longer is 17% behind and the Longer 4 Reg case is the same performance (despite the extra FP performance in 5700 Ultra). this is like down to differences in register space.

Despite the improved PS2.0 performance they are still quite significantly behind ATI's "Pure" DX9 performance, and in the shadermark tests the new MS 2_a compiler for the FX series isn't making much of a difference - this likely suggests that the compiler optimiser that is now in the 52.16 drivers is already getting close to the performance of the HLSL compiled code in the first place, not not their optimal performance for Sahder Assembly reordering. Despite the 5700 being a new chip, entirely designed an built after DX9 was finalised they haven't altered the FX architecture at all do improve some of the missing areas - still no float buffer support and still no MRT's etc.

Great to see that Nvidia took some steps to improve the DX9 performance of the FX5700.

Unfortunate to see that the New version of DX9 compilers coupled with the new dets do not deliver 60% increase in anything.

Uttar
10-23-03, 01:48 PM
Originally posted by Hellbinder
From Dave at B3D..

http://www.beyond3d.com/forum/viewtopic.php?t=8623

Great to see that Nvidia took some steps to improve the DX9 performance of the FX5700.

Unfortunate to see that the New version of DX9 compilers coupled with the new dets do not deliver 60% increase in anything.

Please explain me why you didn't say "Great to see that Nvidia took some steps to improve the DX9 performance of the FX5900." several months ago then? :P

The FX5700 has practically NO improvements, DX9-wise, compared to the FX5900 - it's just half a FX5900, really ;)


Uttar

Hellbinder
10-23-03, 01:54 PM
Originally posted by Uttar
Please explain me why you didn't say "Great to see that Nvidia took some steps to improve the DX9 performance of the FX5900." several months ago then? :P

The FX5700 has practically NO improvements, DX9-wise, compared to the FX5900 - it's just half a FX5900, really ;)


Uttar
Well I guess the obvious answer is we never SAW the results of these improvements until Today. As pointed out already by some the improvements in the Pipelines were hidden by unoptomized drivers.

It is also notable that The FX5900U had/has much less impressive Gains when compared to the 9800pro than the FX5700U compared to the 9600pro.

digitalwanderer
10-23-03, 01:58 PM
Originally posted by Uttar
Please explain me why you didn't say "Great to see that Nvidia took some steps to improve the DX9 performance of the FX5900." several months ago then? :P

The FX5700 has practically NO improvements, DX9-wise, compared to the FX5900 - it's just half a FX5900, really ;)


Uttar
I wish you fanboys would quit knocking nVidia on this nVidia fansite so much, why don't you go to Rage3D or something and do your little ATi love-in/circle-jerk? :eek2:








































;) Sorry, couldn't resist...I thought it was "opposite-day" or something. ;)

volt
10-23-03, 02:05 PM
IIRC, Dave mentioned earlier that there is no difference (performance wise) between new and old compiler.

B&R
10-23-03, 02:11 PM
Originally posted by Uttar
Please explain me why you didn't say "Great to see that Nvidia took some steps to improve the DX9 performance of the FX5900." several months ago then? :P

The FX5700 has practically NO improvements, DX9-wise, compared to the FX5900 - it's just half a FX5900, really ;)


Uttar

Thats because the DX9 benchies are only good enough for a mainstream card not a high-end.:p

schuey74
10-23-03, 05:32 PM
Isn't the only difference (other than clocks) between the 5600 and the 5700 cores the extra vertex pipeline?:confused:

Hellbinder
10-23-03, 05:37 PM
Originally posted by schuey74
Isn't the only difference (other than clocks) between the 5600 and the 5700 cores the extra vertex pipeline?:confused:
No...

DX9 shader processing does not all require the same kind of math function. They shifted some of their processing capabilities toward the more comon math used in PS 2.0. This coupled with better register usage in FP16 gives them a nice boost in performance over the older core.

Deathlike2
10-23-03, 06:36 PM
I think all of us would agree on that NVidia should've done this from the start... instead of the cheats they made...

ATi
10-24-03, 09:03 AM
Originally posted by Hellbinder
From Dave at B3D..
Unfortunate to see that the New version of DX9 compilers coupled with the new dets do not deliver 60% increase in anything.
the 60 % increase will be on dx9.1
________
vaporizer reviews (http://vaporizers.net/)

Voudoun
10-24-03, 09:27 AM
Originally posted by ATi
the 60 % increase will be on dx9.1

I'll enjoy eating that jam tomorrow. :)

Voudoun

Dazz
10-24-03, 11:39 AM
Originally posted by digitalwanderer
I wish you fanboys would quit knocking nVidia on this nVidia fansite so much, why don't you go to Rage3D or something and do your little ATi love-in/circle-jerk? :eek2:

;) Sorry, couldn't resist...I thought it was "opposite-day" or something. ;) I was going to say like, you are the one that tends to spearhead the assault against NV :D

Dazz
10-24-03, 02:04 PM
This is intresting.
http://www.firingsquad.com/hardware/evga_e-geforce_fx_5700_ultra_review/images/shader2.gif
http://www.firingsquad.com/hardware/evga_e-geforce_fx_5700_ultra_review/page4.asp

The Baron
10-24-03, 03:31 PM
Originally posted by ATi
the 60 % increase will be on dx9.1
ahem. the ps2_a compiler is exactly what supposedly gives the 60% increase.

StealthHawk
10-24-03, 04:06 PM
Originally posted by The Baron
ahem. the ps2_a compiler is exactly what supposedly gives the 60% increase.

The ps2_a compiler is "DX9.1."

The Baron
10-24-03, 04:09 PM
Originally posted by StealthHawk
The ps2_a compiler is "DX9.1."
uhm, isn't there like a PS3.0 compiler in DX9.1? or am I on crack again?

oh wait, I'll just go ask you :)

Dazz
10-24-03, 06:16 PM
Thats correct DX9 uses PS3.0 shaders. I think it's just wishful thinking that DX9.1 will give nVIDIA a 60% boost in games and got the lines crossed with the 'Forceware' drivers. After all it was posted by INQ lmfao

-=DVS=-
10-24-03, 07:36 PM
Originally posted by Dazz
Thats correct DX9 uses PS3.0 shaders. I think it's just wishful thinking that DX9.1 will give nVIDIA a 60% boost in games and got the lines crossed with the 'Forceware' drivers. After all it was posted by INQ lmfao

I think you guys a bit OFF

DX 8.0 = PS 1.0 / 1.3
DX 8.1 = PS 1.4
DX 9.0 = PS 2.0 !!!
DX 9.1 = PS 2.0+ maybe 3.0 microsoft doesn't mention it

No word on PS 3.0 no card have support for that !!!
Maybe DX 9.1 will introduce some 3.0 shader support wasn't it DX 10 that was suppose to have it ;)

volt
10-24-03, 07:38 PM
There won't be DX 10 anytime soon, so PS 3.0 should come sooner with an update.

-=DVS=-
10-24-03, 07:41 PM
Originally posted by volt
There won't be DX 10 anytime soon, so PS 3.0 should come sooner with an update.

Maybe in DX 9.2 once R420 and NV40 hit retail ? :thinker:

Lezmaka
10-24-03, 11:09 PM
The DX9 spec has VS/PS 3.0 in it already, I just don't think it's exposed.

-=DVS=-
10-25-03, 12:39 AM
Originally posted by Lezmaka
The DX9 spec has VS/PS 3.0 in it already, I just don't think it's exposed.

PS 3.0 doesn't realy matter if it is in DX9 or if it isn't , not like we have cards supporting it , or better yet games takeing advantage of it:( :rolleyes:

Its hard to get Game with heavy use of PS 2.0 to run fast , now PS 3.0 more complexity will worsen it , my opinion anyways :p
Whats new in PS 3.0 ? any documentation ?

Uttar
10-25-03, 09:00 AM
DX9.0. supports Pixel Shader 1 series, Pixel Shader 2.0., Pixel Shader 2.0.+, Pixel Shader 3.0. and PixelShader 3.0.+ - same support for Vertex Shader.

The SW mode exists, allowing you to test Pixel Shader 3.0. programs already in refrast, at absolutely awful speeds though.
Nick on the B3D forums got a very interesting project allowing VERY fast emulation of PS3.0. through a number of tricks and very good use of MMX/SSE. Amazing job he's doing IMO :) Too bad not enough people know about it!

However, the HLSL compiler does not support PS3.0. or VS3.0. - this could be part of DX9.1., I believe, but there's no reliable evidence to backup that speculation.


Uttar