#Pentium & 3DS
25 messages in this thread
Just got a new Gateway P5-60 today with 32 megs of RAM. I was very pleased to
find that renderings were exactly twice as fast on the new machine as on my
Gateway 486/66V. I can't wait to see what happens when the software is
actually optimized for this puppy! –Rick Rodriguez
Rick,
What did you pay for the Pentium box?
How much did it cost?
How long did it take to get after ordering?
Bob
I paid $4795 for 32megs, 500meg HD, 2X CD-ROM and 17-inch monitor. Not a bad
deal, I think. I ordered it Oct. 12 and received it Nov. 16. –Rick Rodriguez
Rick,
Sounds like a great deal. I'm envious. I was told that pentiums were hard to
come by but it looks like misinformation to me.
Bob
Glad to see Compaq got reasonable with their price. I heard much higher prices
for the p5 boxes a while ago.
John Tissavary (La Luna cie)
I got my Pentium from Gateway, not Compaq. –Rick
Ah, no wonder the price was reasonable. I thought for a moment that Compaq had
lost their heads and were actually going to sell something for a competetive
price!
John Tissavary (La Luna cie)
>> exactly twice as fast
Hmmm … that's interesting. Don't I remember Ana F. saying that she was
getting almost 5x the performance out of her Pentiums? (or are these long
hours finally getting to the ol'boy). I seem to remember her uploading a who
list of tests. Ana … are you listening?
Ang
I never got anything over twice the speed. Most of the time it stayed at around
70% though…
Ana's tests were about 4x that of a 486/33, Ang.
-Alan
Yes, Ana tested a DX/33, a DX2/50, and a Compaq P5-60 all with 8MB,
and a Dell P5-60 with 16MB. I think these tests are why people are
getting this 4 to 5 times faster figure. But since the times of the
16MB P5 system were much faster than those of the 8MB P5 system, I
would conclude that the comparisons between the 8MB P5 system and the
8MB 486 systems were not valid because of memory swapping, and the
comparision between the 16MB Pentium and the other machines were
invalid for the same reason.
I think the reason why the 4-5 times figure stuck in alot of peoples'
heads (including mine until you pointed out the DX/33 and I went back
and looked at the original message) is that the 66MHZ Pentium is
supposed to be 4-5 times faster on floating point than the
486DX2/66. It is supposed to be only 1.8 times faster on integer
stuff, which would indicate 1.6 times faster for the P5-60).
Since 3DS uses a mix of integer and fp, a 1.7 times faster for P5-60
is rather dissappointing. Sometimes I wonder how
floating-point-bound things that are supposed to be
floating-point-bound really are. I once ran a POV raytracing
benchmark on a 486 DX2/66, a Sparc2, an SGI R3000 Indigo, and an
Intergraph C400 machine. The times were as follows:
DX2/66 1:46 min:sec
Sparc2 2:21
R3000 1:42
C400 1:30
So the 486 held it's own against the RISC machines which were, at the
time, the best you could get for less than $50K (no Sparc10 then an
R4000 was available only in Iris Crimson). I think the Sun was rated
on SpecFP at 2X the 486. So I wonder how floating point bound are
these things?
Bailey, (and Alan),
>> the 66MHZ Pentium is supposed to be 4-5 times faster on floating point than
the 486DX2/66.
When running Landmark's Speed Test ver. 2.00, I get an integer
performance about 50% better on a 60 mhz Pentium as compared to a
486/66, and a floating point performance only 100% better on the
Pentium. It sounds like 3DS utilizes the floating point performance
in the same way Landmark recognizes it. (Also, are you sure that 4-5
times F.P. performance wasn't also based on the 486/33? This would
also explain the results.)
That sounds about right. We did some tests with our pentium and found it to be
about 1.5 times faster than a dx2/66 and twice as fast a 486/33
Yann Bertaud – VPG, Autodesk.
Interesting thoughts, Bailey. Certainly, the hardware-based Z-buffer
is where it's at for *real* fast rendering times. I'm hopin' for a
future 3DS release to give us a bit of that sort of a thing for the
Pentium platform, or whatever ends up as the most widely used PC
hardware standard. It's nice to have that double-speed on-board FPC
in the 66 for now, though…
Alan, we're building our own 486/66 with 32meg of ram, a 1gig scsi, Hercules
Graphite card, Illuminator Pro and ethernet to our Abekas A66 disk recorder.
Could you tell me more about the "double-speed FPC"? If we can get more speed,
I'll go for it. Thanks.
Well, Mike, you *did* go for it! Sorry I wasn't clear enough, but by
"double-speed FPC" I was simply referring to the on-board Floating Point
Co-processor that is part of all 486DX2's. Not only does it run at a full 66mhz
(double the speed of a 33mhz motherboard), but works very quickly due to it's
residency in the primary processor, as compared to it's 386/387 counterpart.
-Alan Iglesias
PS — Just let me know when I can send you some TGAs for you to lay to the
A66<g> Happy Thanksgiving.
Ah. I guess I should have asked our engineer first! Thanks for clearing up the
"FPC" question. And, you bet I'll let you know when all is hooked up. We have
also decided to upgrade a 386 machine to a 486/66 by swapping out the
motherboard so we'll have another box for the rendering net. Talk to you
soon…
Sounds like y'all are gonna have a nice set-up there. Best of luck with it and
3DS.
-Alan
>> 4x a 486/33
Aaa-Haaa! That's what I was missing. Thanks.
Ang
Ang,
The results are like follows (taking the best results from all computers)
-Using QBICDEMO.3DS in 1024x768x24bits phong antialiasing high in 3dsr2 for
the still images and Birdwalk 320×200 fli for the animation (other tests for
database -using DBASE- and Wordperfect for Windows)
Processor Still Animation WPfW Dbase
—————-+——-+———-+———+——-+
Pentium/60 | 0:58 | 8:15 | 0:11 | 0:31 |
486DX2/50 | 2:19 | 34:54 | 0:17 | 0:52 |
486/33 localbus | 2:51 | 30:47 | 0:22 | 0:53 |
—————-+——-+———-+———+——-+
ana /krazy images
Ana, isn't speed wonderful? Of course, it just makes me want to go faster!
–Rick Rodriguez
Yes, speed is great. The problem is that nobody know how to stop in the render
race 🙂
ana /krazy images
Ana,
Thanks for the "recoup" of the performance spec's
Ang
PS
Can we assume that each machine had the same amount of RAM?
No, the Compaq Deskpro P60 has just _8_ megabytes, while the IBM
Localbus had 16. I will call again and ask for the exact RAM numbers.
Anyway, I don't think the RAM to be a major issue in this renders,
since in any case there was swap to disk. The only disk operation was
the writing of the frames for the animation. However, I can be wrong
in that issue 🙂
That's because they are using a Gateway, not a Compaq <g> Gateway
don't have distribution in europe, just in the US, so we never tested
it. I'm going to call Madrid for the full "real world" benchmarks
(from WordPerfect to DBase) and post them tomorrow.