The Celeron Myth: Real World 3DSMAX Performance
Added on: Wed Jun 16 1999
|Page: 1 2 3 4
HOWEVER...before you go out and start buying yourself a fleet of Celerons, you must also look at the point at which the Pentium II overtook the Celeron. This occured in the Viewport benchmarks, when the polygon count went above 50,000. The higher the viewport count, the larger the gap between Pentium II and Celeron performance. For those of you who don't know, as the viewport polygon count goes up the memory needed to create a Z-buffer increases, as does the L2 cache usage.
And thats where the Celeron begins to lose its strength. The 128k L2 cache fills up too quickly and system memory is forced to take over. Also, since the system memory runs at 100 megahertz instead of the chips 450 megahertz, that causes a performance hit although ever so slight (3-4%).
But being as this appears to be the Celerons only fault, I would still have to recommend a Dual Celeron System over a Dual Pentium II System. Not only is it cheaper, it slightly out performs the Pentium II, even in such a resource hungry hog as 3dsmax.","large",780);
I'd like to thank everyone for their feedback from the previous article. It helped me to create some additional tests so that everyone will be satisfied that all possible situations were dealt with.
I stand behind my results, but you must remember that the tests were performed on my machine alone. Hopefully someday I will be able to reproduce this test on a wider variety of machines so that the benchmarks reflect a wider variety of home computers.
For the next article I will try to address the differences between a Single dual processor system and Two networked systems of similar speed. Unfortuantly I don't have access to any Pentium III processors, or I'd also do another comparison test with them.
* This of course only applies to Intel based platforms. DEC and SUN both have multiprocessor systems which outperform Pentium II based systems.
** When Render Benchmarking Geometry.max I used frame 10 so that a situatible amount of collisions
had occured. You will also notice in the graph that Geometry seems to take far less time then High
poly. Though this is the case in the graph it is not the case in reality. Instead of making another
graph I merely changed minutes to seconds for Geometry. The graphics remain the same whether the
results are in minutes or seconds. The actual results are:
640x480 (DPII450) 43 minutes 45 seconds.
640x480 (DC450) 43 minutes 25 seconds.
1024x768 (DPII450) 95 Minutes 11 seconds.
1024x768 (DC450) 94 minutes 49 seconds.
*** The reason why the highpoly.max scene is similar across both processors is that the prep stage of 3dsmax is single threaded. (Unable to Make Use of Two Processors). Had textures or additional scene mechanics been added the results would have been different.
Windows NT is a registered trademarks of the Microsoft Corporation 1999.
3D Studio MAX is a registered trademark, of Autodesk, Inc. 1999.
Special Thanks to Pedja Ljubomirovic and Stefan Didak.