CompuServe Thread

#P5-90 optimize

12 messages in this thread
#121835From: Ian ChristieSep 2, 1994 9:45 PM
I remember a thread recently concerning speed optimizing Pentium 90 machines. I just bought a Pentium 90 mb and it turns out to be a little less than twice the speed of my 486-66 for rendering (no disk swap.) Is this all I can expect? I was hoping for more. If anyone could repeat their P5 optimizing experience I'd be much obliged. Ian.
#121842From: Yost GroupSep 2, 1994 10:07 PM
A 90 ought to be 3x a 486/66, according to my benchmarks. Anyone else out there with their own numbers? – G
#121847From: Jamie Clay [DPh]Sep 2, 1994 11:10 PM
Welcome to the 90s! One time benchmark – From 33 to 90 This scene used 44mb of memory to render at full video rez. The average was based on all frames rendered by that CPU type. All 486s are Gateways, the P5s are Microns. CPU Average Render Time 486/33 50:12 486/66 24:22 P5/60 12:05 P5/66 10:29 P5/90 08:04 ======= Also note that P5-60s are really getting cheap! You can get a basic P5-60 now for less than $1.5k. Add 32mb of memory @ $1.3k and you have a decent render servant! J-me
#121885From: Ian ChristieSep 3, 1994 9:51 AM
Darn, my P5-90 is choking seriously! Ian.
#121915From: Brian McHenrySep 3, 1994 1:34 PM
I don't get it? Is Autodesk that inefficient? I find with math intensive Fortran Simulation Programs Compiled with Fortran Powerstation 1.0 using Phar-Lap Dos extendor ( I believe the same AD uses for 32-bit DOS code) on identically configued systems: 486/66 vs Pentium/60 both with 16 meg and same disk & video drivers that I get 12X the speed with the pentium. Let me repeat, The Pentium/60 is 12 times as fast at number crunching as the 486/66!!!!! Obviously AD is somehow I/O locked and not optimizing the rendering. Since the actual rendering should be numbercrunching mania and very limited video until their through! Anybody else get these kinda numbers?
#121932From: Jamie Clay [DPh]Sep 3, 1994 3:18 PM
>>Anybody else get these kinda numbers? Not even the standard benchmarks like DOSmark, WinStone or WinBench return that kind of result. (12x) Almost ALL benchmarks show the 60mhz Pentiums to be only twice as fast as DX2-66s and 3DS echos those findings. So it looks like you have one of the few applications that sees that type of performance improvement. Good for you, perhaps you should consider writing a renderer! 😉 J-me PS -Perhaps you didn't mean it, but it sounded like you BLAMED Autodesk for the fact that 3DS+Pentium doesn't = Fortran app+Pentium. If you did, come on, man – get real.
#121936From: Brian McHenrySep 3, 1994 3:45 PM
Are the "standard benchmarks" using dos extenders like Phar-Lap so there testing 32-bit code? And/or are they somehow I/O bound. I know all the tests from the PC magazines are severely I/O bound. My test is a ~2 meg program which includes I/O but isn't burdened by it. My implication about Adesk and there apparent lack of utilization of the POWER of the Pentium is that the number crunching end of the 3DS must be written in 16-bit C or the like if it doesn't benefit from what I see as a much greater than 2x speed increase. I've gotten real, man. Real fast man which apparently Adesk hasn't yet.
#121944From: Yost GroupSep 3, 1994 5:42 PM
Excuse me? Please name a faster commercial production renderer on the Pentium. – G
#121964From: Brian McHenrySep 3, 1994 7:50 PM
G – My comment was related to the fact that any math intensive program should realize a speed gain of greater than the only 2x 3DS does from going from the 486/66 to the Pentium/60. I assumed that 3DS is math intensive during the rendering stage? It's math procedures were obviously written in 16-bit C or the like and therefore do not truly exploit the power of the Pentium over the sluggish 486-66. With true 32 bit compilers on both machines, as I earlier stated, my math intensive routines realize a 12x increase. So excuse u all u want!
#122026From: Yost GroupSep 4, 1994 11:29 AM
The question was… please name a faster commercial production renderer for the PC, Mac, Amiga, etc. – G PS: 3DS is a true 32-bit app, written with the Metaware HighC 32b compiler.
#122042From: yann bertaudSep 4, 1994 12:57 PM
I've been doing some tests in the last couple of days comparing a p5-90 to a p5-66 and a 486/66. I rendered the antique car which I had done about 2 years ago. You can see a rendering of it here on the forum. It has a lot of autoreflecting parts and is therefore very computation intensive. Here are the results: (in minutes:Seconds) P5-90: 16:55 p5-66: 20:40 486/66: 43:23 All have 64 megs ram, used about 32 megs, image is 1024×768 TGA. Basically, the 90 is about 3 times as fast the dx2/66. Yann Bertaud – Marketing Support, Autodesk.
#122136From: Ian ChristieSep 5, 1994 8:54 AM
Thanks, Sounds like I have a bum motherboard. I'll see if I can replace it. Ian.