|
|
#1 | |
|
Registered User
Join Date: Mar 2003
Location: Tulsa
Posts: 5,101
|
Wanting to take a deeper look into the performance variables of 3dmark, Paticularly its vertex emphasis, I decided to collect and gather data of how well 3dmark scales with pixel pipelines and vertex units.
Here is the chart. ![]() Game Test 1: This result actually surprised me. The vertex shader emphasis is relatively small here, And this test is definately pixel shader bound. Game Test 2: This is yet another interesting result, The emphasis on vertex performance is rather astounding. In some cases disabling vertex units has much more profound impact than disabling 4 pixel pipelines. It is my subjective opinion that this test is unrealistic in regards to gaming and polygon counts. GameTest 3: This test is very balanced between vertex/pixel shader performance. Losing fairly fair shares of performance with each disabled. Conclusion: Why this conclusion is half based and only based off a night's work. I have to say that denying 3dmarks' extreme load on vertex performance is folly. This benchmark is as vertex limited as it is pixel limited. Whether this vertex performance will be a realistic interpretation of future gaming remains to be seen. I personally have my doubts we'll be seeing games with over 1 million drawn polygons per scene anytime soon. Even Unreal 3.0 will only draw 400,000. P.S. This information oughta be useful for your soft modders, give ya an idea of what to expect ![]()
__________________
|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 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 Last edited by ChrisRay; 10-23-04 at 05:24 AM. |
|
|
|
|
|
|
#2 | |
|
Anisymbolic
Join Date: Aug 2004
Location: Wellington, New Zealand
Posts: 1,365
|
Hrm... I'm actually interested in why the NV40's vertex performance is FAR more reflective of its clock speed than its Pixel performance is. In other words - the NV40 does a lot better in terms of pixel fill per clock than vertex fill. Or at least, that's what it looks like to me.
Interesting results though - top effort in producing this!
__________________
Dr Possible: Core 2 Duo E6400 on Gigabyte GA-965P-DS4. Galaxy GeForce 7600GT. 2GB Corsair XMS 2 DDR2-6400 RAM (CL5). ATi Theatre 550 Pro. Windows XP MCE. All stored in Piano black Antec Sonata II, with a broken door. Mobile: ASUS M2400N, Pentium M 1.5 GHz. 512 MB DDR RAM. Intel EXTREME graphics. Windows XP SP 2 / Ubuntu 5.10. Ridiculous DOES not have an 'e' in it. It comes from "ridicule" and has less than nothing to do with the colour red. |
|
|
|
|
|
|
#3 |
|
Registered User
Join Date: Jan 2004
Posts: 2,553
|
Nice work!
|
|
|
|
|
|
#4 | |
|
R420, NV40
Join Date: Jan 2003
Location: In Front Of My PC
Posts: 164
|
3DMark05 is pixel limited too. Here are my X800 PRO scores:
ASUS AX800 PRO/TVD 256MB (475,900) 12x1 / 6 VS Units = 4600 3DMarks ASUS AX800 PRO/TVD 256MB (475,900) 16x1 / 6 VS Units = 5111 3DMarks ATI RADEON X800 XT Platinum Edition (520,1120) 16x1 / 6 VS Units = 5795 3DMarks
__________________
AMD Athlon64 X2 6000+ | ASUS CROSSHAIR | 2 x 1GB Corsair XMS PC2-6400 DDR2 800MHz | ATI RADEON HD 3870 512MB (800,2340) | PNY GeForce 8800 GTS 320MB (513,1584) | ATI RADEON HD 2900XT 512MB | Leadtek WinFast GeForce 8800 GTS 640MB (500,1584) |
|
|
|
|
|
|
#5 | |
|
Registered User
Join Date: Mar 2003
Location: Tulsa
Posts: 5,101
|
Quote:
__________________
|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 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 |
|
|
|
|
|
|
#6 | |
|
CoD4!
|
Very intresting results, espically the diffrene between 8x1 adn 16x1
![]()
__________________
Sony Vaio SB: i7, 8Gig, Intel 320 300gig |
|
|
|
|
|
|
#7 | |
|
Registered User
Join Date: Oct 2004
Location: Austin, TX
Posts: 115
|
Quote:
That could explains why I only got about 300-400 point more when I enabled the extra 4 pixel units on my BGF 6800 OC. I wonder if higher clock speed will reduce the difference?
__________________
Gigabyte EX58-UD3R| Intel Core i7 920 | BFG GTX 260+ | 6 GB CORSAIR DDR3 1600| 1TB WD Black SATA |750W Corsair |
|
|
|
|
|
|
#8 | |
|
Registered User
Join Date: Mar 2003
Location: Tulsa
Posts: 5,101
|
Quote:
to me this implies a bandwith restriction even on the shader units, It makes me wonder if even the Nv4x series are seeing diminishing returns at 16 pipes (compared to 8x1 to 12x1) I wonder how bandwith limited the X800 is. Which would of course raise several other questions, Clock effiency, Shader efficiency, ect
__________________
|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 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 |
|
|
|
|
![]() |
| Thread Tools | |
|
|