p5-90 problems
39 messages in this thread
Hi Gary,
I just received my Gateway P5-90. It has the Matrox board, IDE CD-ROM, sound
card and I've added a second hard drive, ethernet card, modem and SCSI card. So
far, 3DS is working fine. The most difficult part of getting it up and running
was messing with int's to make all the boards work. I'm using DOS 6.21's memory
managers and have the latest QEMM on order. I'll keep you posted with any
conflicts.
– Steve Ormonde
Steve, have you done any specific benchmark timings re your old machine?
– G
Yes,
The P5-90 is 224% faster than our Gateway 486-66 w/ same amount of RAM. The
486-66 uses QEMM for memory management and the P5-90 uses DOS 6.21's MM's. As
soon as I receive QEMM 7.2, I will rebenchmark (is that a word?) and let you
know.
BTW, the benchmark model we use is an interior view of a typical residential
home with multiple shadow casting lights, flat reflections and about 56K
faces.
– Steve Ormonde
Steve, could you do a benchmark for me? Please render the chevy.3ds file at
320×200. On my DEC P5/66, it takes exactly 60 seconds. Thanks,
– G
Are you referring to the "57chevy.3ds" on the CD-ROM? If so, at 320×200, it
take 23 seconds to render.
– Steve Ormonde
That's with Metal shading and AA on? Wow! Great time! Those 90's really
scream… very encouraging. We're going to be using Micron P5/90's at the
Siggraph booth, along with Liquid Speed. I'm really looking forward to using
3DS on one of those systems.
– G
Yep, Metal and AA were on!
– Steve Ormonde
As Alan said, there are two chevy models that come with 3DS…
Chevy.3ds (on the magnetic media, with the brick wall) and
57chevy.3ds (on the CD-ROM). Which one was it?
As Alan says, I was referring to the "57CHEVY.3DS" (I don't keep the samples on
my HD and assumed the CD-ROM was the same). I will dig up the "CHEVY.3DS" and
try again.
– Steve Ormonde
There may be a bit of confusion here, Gary. Weren't you referring to
the Chevy.3ds file on the r3 shipping disks? If so, it's rendering
times are considerably longer than the 57chevy.3ds that is in the
Viewpoint directory on the CD-ROM that I believe SteveO may be
reffering to. If I am off-base about this, just tell me to shut-up
and go away, and I will…<g>.
Anyway, here's my times for the DEC Pentium 60 (AA on, Pixel size
1.1), if anyone is interested (ie JNJ):
Chevy.3DS (Shipping disk file) 320×200 — 55 seconds
Chevy.3DS (Shipping disk file) 640×480 — 2 minutes, 23 seconds
57chevy.3DS (CD-ROM file) 320×200 — 31 seconds
57chevy.3ds (CD-ROM file) 640×480 — 59 seconds
The two files' difference in rendering times between resolutions is
quite radical, eh? I would guess this is due to the lack of backround
objects in the 57chevy.3ds file. I figure larger resolution makes
less of a difference 'cause of the proportionally less scene-covering
objects it has to render.
Steve, if you see this, could you benchmark the Shipping Disk file at
640×480, please? I think Corky's 90s were around 1:40 somewhere.
That p90 Chevy render benchmark got to me as well. Are you referring
to the copper Chevy with the brick background & graffiti ? If so,
that's the one I used on my 486-66 – which is pretty fast for a 486.
I was about to post some questions when I saw this message, as my
machine rendered this file at 1m23s under these conditions:
A/A ON, Pixel Size 1.1, Dith. T/C ON, Shadows & Filtering ON, Bartlett
Filter, Metal Shad. ON, etc..
Therefore, I found it overwhelming on a P90 at 23 seconds (wasn't that
the number ?) ~= 400 % faster. I'd like to propose that someone
define all imaginable equalizing render attributes which might
possibly skew the numbers, as I am very curious about this. For
example, when I pull the maps from an old erasable optical disk, the
render time goes up to 1m42s compared to pulling them off the hard
drive. Also, what about scenes with layered transparency, such as
might be generated with VAPOR/FLAME axp use. This area seems to be
3DS's Achilles (sp?) rendering heel (as well as most renderers').
Also, I'm curious about benchmarks for SCREEN REDRAWS, KEYFRAMER
PLAYBACK SPEED, FLIC PLAYBACK SPEED, etc., are these positively
affected as well ? Very, very curious for accurate information.
I've never seen a renderer in my LIFE that doesn't come to its knees
when multiple overlapping layers of transparent polygons hit it.
According to the benchmarks some of our beta sites who also use SGI
machines did last year during the r3 beta period, our transparency
rendering is up there with the fastest scanline renderers in the
world. FYI. <g>
I didn't mean to imply that 3DS was skewed out of the norm wrt any
other renderer wrt layered transparency, that's why I included this
line:
<<…rendering heel (as well as most renderers'). >>
I just meant wrt the same scene w/o transparency in 3DS. I didn't say
"*ALL* renderers", because I really haven't used or benchmarked all
renderers – especially recently. OK ? <g>
OK! I'm just extremely proud of how quickly we do multi-layer
transparent rendering vis a vis other systems.
Bill, I'm assuming you are talking about that rendering at 320×200,
right? My P60 does it in 55 seconds. If your 486/66 does it in 83
seconds, that *does* seem pretty dang fast to me. What's the time
with 640×480?
I feel that most of these benchmarks assume a good HD controller with
a decent HD for map access. Certainly, something like the optical
disk can throw a wrench into the works. As particle systems like
Vapor and Flame are primarily special effects, they probably don't
represent the "typical render" very broadly.
Many of us are using Chevy.3ds/640×480/Metal/AA on/Pixel size 1.1 (the
default) for our rendering benchmarks. The 486/66s were coming in at
over 5 minutes, with the p60s at the 2:23 mark (GusG's too). I've
seen P90s at around the 1:40s.
I've been doing redraw benchmarks with the chevy.3ds file, too. I make
all four viewports the camera view and redraw all of them (shift+`).
My 486/66/ViperVLB was around 12 sec. and 6 with Liquid speed. My DEC
P60/ViperPCI/Liquid Speed now does it in less than 3 seconds!
Needless to say I'm gettin' a lot more done per hour these days.
-Alan
Hello Mr. Alan "Benchmark" Iglesias!
GW P5/66, Chevy.3ds @640×480: 2:18 (nice model!)
Redraw, ATI/PCI (no LS): 6 sec
GW 486/66, too lazy to benchmark
You goin' to Siggraph, right?
You goin' early?
You scuba?
PS: Man, I forgot to tell you: we vandalized your Chevy – broke in and
stole the seats, rearview mirror, sun visors, and tires for a recent
project. I hope you have insurance. I'll send you some JPG's if
you're interested.
Hey Johnny!
That P5/66 seems to running nice, huh? When ya gettin yer 90??
I'm sure you realize that Viewpoint scanned in that Chevy model…It woulda'
took me 2 years to do that by hand. I'll betcha' YOU coulda' done it, though!
Funny, but the only car parts I DID model were the outside mirrors and
windsheid wipers (don't look at 'em too close!<g>).
Yup, MartinF and I will arrive in sunny (and hot, and humid, and…) Orlando on
the 23th, as to get the ole' Saturday stay-over don't 'cha know. So I guess
that's early, eh? Haven't SCUBA-dove in years, but I'll be glad to man the
pumps for ya<g>! We'll definitely have to get together before the show if you
have the time, I think we will. Let's schedule something…
-Alan
Hi Alan,
<< What's the time with 640×480? >>
Don't know now, I'll get back to you with that. May I suggest using 720-756 X
486 X 24 & a Pixel Size of 1.5-2.0 for real-world benchmarks ?
<< I feel that most of these benchmarks assume a good HD controller with a
decent HD for map access. >>
Yes. Maybe using a RAM Disk for this evaluation would eliminate that variable ?
<< My 486/66/ViperVLB was around 12 sec. and 6 with Liquid speed. My DEC
P60/ViperPCI/Liquid Speed now does it in less than 3 seconds! >>
That is very encouraging. I'm using a Viper VLB as well. Tell me, what
about KF playback and Flic playback – are they improved as well ?
Also, what about using Liquid Speed *with* a P90 – what kind of
proportional results does that give ? What display are you using
with the P90 ? (or anyone). Thanks for the info & your time.
BILL
Hi Bill,
The higher res and Pixel size is an interesting suggestion, but by "real-world"
I guess I meant using very standardized, common and default values, used by
many 3DS folks who use it for much more than broadscast-video only. It's just
easier to run the test with common values. And when things are easier, you
usually get good data.
The RAM disk certainly would take the HD out of the loop, but I think in a 2-3
minute render, and decent HD and controller would be fairly comparable, and not
make a huge difference.
Now that the 90's are rendering it in well under 2 minutes, maybe we should go
for a higher resolution just to keep the test lasting long enough to make
decently-scaled comparisons
The bottom line with all this benchmark stuff (as far as I'm concerned, anyway)
is to provide forum folks with a dependable number with which to evaluate their
system, and make sure things are basically OK. Nobody wants to render thousands
of frames, then find out the machine is for some reason not rendering as fast
as it might. Also, it gives a good way for folks to see if a particular system
that they want to purchase is up to snuff before they buy it. Some folks had
reported slow performance with some 90's and I wanted to see if it had anything
to do with native 3.3v capability, as many forum members seem to be looking at
"voltage regulated" P90 systems. So a gathering of P90 benchmarks would help
that, I think.
I haven't tested flic and KF playback to the second yet, but AAProrunning
in-RAM flics is considerably faster than the 486 for sure. KF playback is also
considerably faster than the 486. My UI display is set to 1024×768.
I'm pretty sure the 23 second number was for the CD-ROM file.
-Alan
Well, now I've tested at 640×480 with all other parameters as before.
I also noted something else which is probably due to my lack of
experience with 3DS:
I was quoting the render time as given in the Render Status Dialog. I
have now noticed that CURRENT STATUS pull-down gives a very different
number, and actually timing the thing yet another number. So here's
the breakdown with my VERY OLD AMI MB 486-66:
At 320×200 :
Render Status shows 1m23s
Current Status shows 2m10s
Actual Time 2m07s
At 640×480 :
Render Status shows 4m01s
Current Status shows 4m48s
Actual Time 4m41s
Also, my Viper VLB is one of the very first shipped, and I'm using the
latest production Vibrant drivers at 800x600x8. With all VP's set to
camera view, screen redraw is 10 seconds ???? No Liquid Speed.
I'll have to search through the messages again, but was that 23 second
number for the copper Chevy with the brick wall ? Or the other one ?
If it's the other one, did anyone benchmark this one and *actually*
put a stopwatch on it ?
Anyway, thought I'd clear this up (probably made more confusing <g>)
Sorry for all the rambling. I dug out the faster rendering 57Chevy.3ds, but
I've changed it (my larger relection maps, etc.) so I guess benchmarks wouldn't
be valid. I'm *guessing* my altered version takes longer to render, but even
still the P90 numbers are approaching 300% faster. Worth the money, it would
appear. Now when's those 150Mhz shipping <g> ?
BILL
OK, now I've read through the messages again. Can anyone post the P90
time for the CHEVY.3ds file (actually timed) ? I guess I've put the
other Chevy out to tape BU, as I don't use a CD ROM reader on this
machine. The other one's blue right ? I know I've seen it.
Thanks,
BILL
Good point, Alan. I'm curious as to which version Steve was using as well.
– G
JimB's benchmark would indicate it is the 57chevy.3ds file. He got 23
seconds, too.
-Alan
Alan
57chevy.3DS (CD-ROM file) 320×200 — 63 seconds
—486-66 57chevy.3ds
(CD-ROM file) 640×480 — 118 seconds
57chevy.3DS (CD-ROM file) 320×200 — 23 seconds
—Micron P-90 57chevy.3ds
(CD-ROM file) 640×480 — 45 seconds
How sweet it is.
JB
Ooof!
That Micron just hums along, don't it Jim?
Have fun with it.
-Alan
PS -Are we gonna see ya at SigGraph this year?
>> SigGraph this year?
Well… it seems the more long distance planning I put into something,
the better the chance of some last second aberration squelching said
plans. (Particularly Vacations). This time it's my Aussie inlaws
arriving –
Surprise-July 26, for their 2.5 month (Did you say "Months"?) stay at
our house. Sooo…I hope to come (alone) but….???
Siggraph is not vacation, it's work………geez, you'll get us all
in trouble!
Bring 'em all!–But don't let them talk to Martin, we'll loose him for sure!
-Alan
Wow… Those P5/90's are FAST. Whoa.
James,
For another benchmark data point..
I have a Dell 466M (66Mhz 486) and 57Chevy renders in 104 secs
@640×480 and in 53 seconds @320×200. I'm getting ready to plunk down
my cash on the new Dell Dimension P90. Latest reports show it just a
bit faster than the Micron, although the optiplex system looks very
good with the built in SCSI 2 and EISA bus slots. But the price is a
lot more. Can't wait to add one of those beasts to the network…
Thanks, Alan, for the clarification, and you're right about the
resolution stuff. Was chevy.3ds your creation? I think I saw your
name on a chevy still somewhere… Thanks for the times too. How
much RAM do you have on that machine?
Hi John,
There is 64MB presently in my P60.
The picture of the chevy in the r3 reference manual is my *composition*.
Viewpoint scanned in the Chevy (no one modeled it by hand).
The file many of us have been using (all along) for benchmarks is the Chevy.3ds
file from the r3 shipping disks. It is basically my file that GaryY changed and
made "ship-able" before the release of r3, as my file was filled with IPAS
routines. The 57chevy file on the CD-ROM renders much faster and has less
detail in the scene.
-Alan
I have been reading these rendering benchmarks and I did a test of my own.
I have a Dell 486/66 with 16meg RAM and I was rendering chevy.3ds in 4 min 30
sec with a swap of 260 and page faults at 360 @640×480. Does that seem about
right?
What files should I be loading up in my config.sys and autoexec.bat?
Should I load HIMEM, EMM386, etc…??
Currently I am just loading my mouse driver and my CD-ROM driver.
What should I set my files and buffers at?
Andrew Sinagra
Andrew, although your times seem pretty much typical for the 486/66, swapping
can throw off the rendering times so much that it's probably a good idea to
stick with in-RAM benchmarks. On my system, the file uses just under 17MB, so I
guess your system's swapping isn't much of a suprise.
As for which Memory Manager to use, it seems to vary from machine to machine. I
always have folks benchmark their systems using different ones (including NO
memory manager) to see which one provides the best rendering performance for
their machine. Then it's a matter of considering the best configuration for
their complete system. The PAR, for instance, requires a MM to run pardrv.exe.
-Alan
Get Qemm and forget about himem.sys and emm386.exe.<IMHO>
thanks Alan
-JNJ
You're right, Alan! As my message to Gary said, I used the "57CHEVY.3DS" from
the CD-ROM. I will dig up the other and try again…
– Steve Ormonde
hello gary….the Xi pentium 90 Mhz just did 320×200 with all the default
settings at 51 seconds.
It does 640 x 480 at 2:06.
This is w/ bare min config…i took everything out of config.sys and autoexec
and now starting to put it all back in one by one.
It's a happy dayi can _see_ again..the Vibrant 103 works with the Stealth 64
like a dream come true.
Thanks to everyone who helped by sharing their knowledge andexperience.
thanks gus, Chris Hall @ Vibrant, Evan Yares, Xi Computer Systems in San
Clemente, CA and ViewSonic.