Screenshot from the animation showing fps, the number of vertices rendered and the amount of frame buffer memory used.
|
Advice Beginners BIOS Guide CPUs Links Mainboards Memory Network Storage Video/Sound Cards Contact Forum SiteMap Sponsors WebNews Home |
. | . |
Prices: Mainboards ABIT ASUS Chaintech Shuttle Soyo Tyan CPU Intel P4 2.4C-800 P4 2.6C-800 P4 2.8C-800 P4 3.0-800 P4 3.2-800 AMD AthlonXP XP 1700+ XP 2000+ XP 2400+ XP 2500+ XP 2700+ XP 3000+ XP 3200+ Athlon64 Athlon64 3200+ Athlon64 FX-51 Opteron Opteron 240 Opteron 242 Opteron 244 Opteron 246 Memory Corsair Crucial Kingston Mushkin OCZ |
LOSTCIRCUITS |
|
| nVidia Quadro FX4500 Pushing the Professional Envelope with SLI | |
|
(Review by MS, February 1, 2006) |
| nVidia Quadro FX4500 SLI |
Springmark and SLI
So far, the results we showed are essentially single GPU results, simply because there is no SLI support in the specific applications used. At least, we could not see any evidence for it. The situation changes a bit when we are running a synthetic graphics benchmark that uses a spring model with each loop consisting of 1 million vertices that take up 36 MB of frame buffer. It is easy to see how 10 coils will use 360 MB on-board memory, 20 coils will gobble up as much as 720 MB. Keep in mind though that SLI does not allow sharing of the frame buffer between the cards, in other words, each card is on its own. This does not appear to change with split frame rendering - we did not expect it to but we tried nonetheless anyway.
Screenshot from the animation showing fps, the number of vertices rendered and the amount of frame buffer memory used.
Frames per second per coil (million vertices). Higher is better: At the tail-end (20 million vertices / 720 MB memory) the resolution does not suffice to see the sudden drop in performance after exceeding the frame buffer density.
Vertices processed per second: Higher is better. If we plot the number of vertices / sec, it is easy to see what is going on. As long as the model fits into the frame buffer, the actual vertex processing performance is fairly constant but drops immediately when the frame buffer is exceeded. The fact that the "real world performance" does not reach the 180 million vertices / sec listed in nVidia's specifications merely reflects the difference between theory and practice or lack of optimization for the particular model used.
|
Creative Labs Wildcat Realizm 800 |
Next Page: => Viewperf 8.1 =>
If you enjoyed reading this article and found it useful, please consider making a small donation to LostCircuits.