|
|
Thread Tools | Search this Thread | Rate Thread | Display Modes |
#106
|
|||
|
|||
OSDM benchmark data for Dell PWS360 workstation
Benchmark system ...
# "Graphics test sec" , "Viewbench test milliseconds/frame" , "Modeling/CPU Tests sec" # 0 , 211 , 0 For further details, see the attached benchmark results file ...
__________________
John Struthers Product Design Consultant & CAD App Specialist |
#107
|
||||
|
||||
Some might have noticed that the results posted here often differ much. There are two main reasons for this:
1. Most drivers are preconfigured to run in "sync with vertical retrace" method. This means that you'll never get frame rates beyond the number of screen refreshes which your monitor supports. To address this, you can disable vertical sync in the driver. However, this is only useful for benchmarking. For interactive usage, it can introduce graphical artefacts which many people find unpleasant. 2. The test models are fairly small and simple by today's standards, and today's graphics cards process them easily. As a consequence, the bottleneck is now closer to the system bus and CPU again. By enabling displaylists in OSDM, the CPU load will decrease sharply. I ran some tests recently and found that with today's graphics cards you will typically get a 2x performance improvement by simply enabling display lists in OSDM's graphics settings. Displaylists, however, do cost some RAM, so if you're already maxing out your system, this option is not for you. Claus |
#108
|
|||
|
|||
We just got in a Dell 670 box with the nVidia FX3400 chip and so far, it doesn't look that much better than the older Dell 650 boxes. I tried running Viewperf 8 (just released today at the www.spec.org site) and it refuses to run, complaining of a Pixel format error.
Has anyone seen how the PCI-Express cards run? I haven't been impressed so far with the limited testing I've done with this box. We may opt to get these boxes in the future with 3DLabs' PCI-Express offerings instead, since the price is comparable. Results are attached for the following specs: P4 Xeon 2.8 GHz nVidia FX3400 1 GB RAM Win2K w/ SP4 OSD 12.01D |
#109
|
||||
|
||||
May,
according to your results, almost all of the viewbench test models max out at a framerate of ~60 Hz which is probably your monitor's refresh rate; see my previous message. Turn off vertical sync, then compare again. Claus |
#110
|
|||
|
|||
Thanks for the tip, Claus. I reran the Dell 670 and 530 with V-sync disabled (same 530 I submitted before). The numbers look much improved, though workplanes are invisible in the viewport (but not Structure Browser) when it is disabled.
I'll try to rerun the 650 I submitted before to see what effect disabling V-sync has on it. We normally run these boxes at either 75 or 85 Hz. |
#111
|
|||
|
|||
OSDM benchmark on older hardware ...
# "Graphics test sec" , "Viewbench test milliseconds/frame" , "Modeling/CPU Tests sec" # 0 , 563 , 0 Here is the result of enabling the "Display Lists" option in OSDM's Graphics Settings and re-running the benchmark test ... # "Graphics test sec" , "Viewbench test milliseconds/frame" , "Modeling/CPU Tests sec" # 0 , 587 , 0 I noted that the "Display Lists" setting was cleared after the benchmark test was run??
__________________
John Struthers Product Design Consultant & CAD App Specialist Last edited by John Struthers; 08-12-2004 at 02:06 AM. |
#112
|
||||
|
||||
viewbench disables the displaylists by default to make sure that it runs in a reproducible environment. You'd have to change the viewbench code to fix that, I guess. Hmmm...
Claus |
#113
|
||||
|
||||
Quote:
__________________
John Scheffel |
#114
|
||||
|
||||
Displaylists are off by default only because of the additional memory load they cause.
However, since graphics cards these days come with 128 MB of RAM or more, they can store a lot of displaylists in their own RAM, so my guess would be that for many models not a lot of main RAM is consumed. However, I haven't run tests in this area for some time, so I cannot be sure. There is also one known bug in the area of changing part colors while display lists are on which will be fixed with the next release. That bug was discovered not too long ago, though, so it didn't play a role in the decision about the displaylist defaults. Claus |
#115
|
||||
|
||||
I discovered that the Display Lists setting is not persistent, so it will not be remembered when you exit and restart Modeling. If you want it turned on at startup, you can add the following line to one of your startup files.
Code:
set_displaylist_mode :on
__________________
John Scheffel |
#116
|
||||
|
||||
Effect of Z Buffer Depth on Benchmark
I have been running some benchmark tests on a PC with Intel Extreme graphics (part of the 865G chipset). This is the graphics in the standard "business" PC offered by our IT department. I know it's not a great choice for ME CAD, but I wanted to see how well it works with Modeling since it is a cheap and easy to order option. Here are the specs.
PC: HP/Compaq Evo D530 CPU: P4 2.6 GHz with 533 MHz FSB RAM: 512 MB Graphics: Intel Extreme 2 O/S: Windows 2000 Pro SP4 OSDM: v12.01D Viewbench: v1.68 At first the results were surprisingly good, with a score of 343. Then we discovered a graphics issue. When laying down infinite construction lines, the lines did not appear until the OK button was clicked. Some testing found that changing the Z buffer depth from the default of 24 bits to 16 bits fixed this problem. However, it had a significant effect on the benchmark. With 16 bits the score was 788. This is the second graphics issue we have found that can be fixed by forcing the Z buffer depth to 16 bits (the other was with ATI Radeon graphics in a laptop). Just be aware that doing this may impact graphics performance.
__________________
John Scheffel |
#117
|
||||
|
||||
Interesting observations - thanks for sharing them!
Changing the video resolution, color or Z buffer depth can indeed force the driver into slower execution paths in which it uses software to do part of the rendering. Maybe this is what happened in your case as well when switching to a 16-bit Z buffer. Claus |
#118
|
||||
|
||||
Improvement with Graphics Card Upgrade
Quote:
PNY NVIDIA Quadro FX500 (model VCQFX500-PB) - 128 MB DDR - dual monitor support, one analog and one DVI out - max resolution with 32bit color - 2048x1536@75Hz - AGP 8X (supported by D530) - DirectX 9 - OpenGL 2.0 I purchased one of these cards and installed it in my D530 with the latest reference drivers from the NVIDIA web site (not the drivers from the CD). I did not find any graphics issues, and didn't expect to since the NVIDIA chip set was certified by CoCreate. There was no need to change the Z-buffer depth as was necessary for the standard Intel Extreme graphics. Here are the results. Intel Extreme - 343 (before graphics fix for 2D line issues) Intel Extreme - 767 (after force 16 bit Z-buffer to fix line issues) PNY Quadro FX500 - 214 (default as installed settings) The driver for this card actually has a "OneSpace Designer" setting that is supposed to optimize for Modeling, but it made no difference in the benchmark so I went back to the defaults. All of these tests were run at a 75 Hz refresh rate. Based on the comments above from Claus, I raised it from 75 Hz to 85 Hz on my PC and was able to reduce the score to 195. The PNY card can go higher than 85 Hz, but I could not test it because my old monitor does not support it. One other advantage of this card is that it supports Occlusion Culling (not available with the Intel Extreme graphics) which should also boost the speed if turned on. I assume this is disabled by the benchmark test for consistency the same as Display Lists. Beyond the benchmarks, after installation of this card the 3D graphics felt faster and more responsive, and seems to be free of any graphic anomolies. Overall I'm very happy with the combination. So bottom line is that putting a relatively inexpensive 3D CAD graphics card in a business PC that is pretty cheap and low end by today's standards, you can turn it into a capable ME CAD PC.
__________________
John Scheffel |
#119
|
||||
|
||||
Viewbench Results Table Updated
I have updated the results table with the results which have been posted since the last update in May.
http://www.cocreateusers.org/misc/viewbench_results.html
__________________
John Scheffel |
#120
|
||||
|
||||
HP/Compaq NC8000 Notebook
PC: HP/Compaq NC8000
CPU: P4 1.6 GHz RAM: 1000 MB Graphics: ATI Radeon 9000 with 64 MB DDR O/S: Windows 2000 Pro SP4 OSDM: v12.01D Viewbench: v1.68 The initial score was very good at 206. Unfortunately the graphics in this model also suffer from the sluggish mouse cursor tracking discussed in another thread, so it was necessary to force the Z-buffer depth to 16 bits to fix it. After that the score went up to 765, a pretty dramatic difference.
__________________
John Scheffel |
Currently Active Users Viewing This Thread: 2 (0 members and 2 guests) | |
Thread Tools | Search this Thread |
Display Modes | Rate This Thread |
|
|