VOGONS


First post, by Mau1wurf1977

User metadata
Rank l33t++
Rank
l33t++
Hi, I am Phil and in my spare time enjoy working on projects that revolve around old computers. Specifically DOS and Windows PCs […]
Show full quote

Hi,
I am Phil and in my spare time enjoy working on projects that revolve around old computers. Specifically DOS and Windows PCs that run old games. I hope you enjoy reading this document!

In this document I am publishing the results of my Voodoo 2 processor scaling project. What is Voodoo 2? Voodoo 2 is a graphics card from the company 3dfx that launched in 1998. It is the successor to the Voodoo and, at the time, offered leading performance in 3D computer games.

A total of 25 processors have been used to collect data. All the way from a Pentium 100 to a very fast Pentium III-S 1400. Voodoo 2 cards scale very well with processing speed but I always wanted to have a better idea of how much exactly.

The file is too large to upload here (10 MB), so please download it here from my Dropbox account:

http://www.philscomputerlab.com/voodoo-2-and- … ng-project.html

This was a LOT of work and my first time publishing something in written form. So go easy on me 🤣

My website with reviews, demos, drivers, tutorials and more...
My YouTube channel

Reply 1 of 37, by mwdmeyer

User metadata
Rank Oldbie
Rank
Oldbie

Very nice work! Just had a quick read over it.

If I were using a Voodoo 2 with or without SLI, I'd really want to use 800x600 or higher, so based on your testing I'm thinking a Pentium 2 400+ is a good match for it.

For the Pentium 1s there is no point in SLI at all, which isn't really a surprise.

Vogons Wiki - http://vogonswiki.com

Reply 2 of 37, by darksheer

User metadata
Rank Member
Rank
Member

😳 Now that's what I call a huge interesting work about Voodoo 2's cpu dependence.
Using a PIII 800 EB for my Voodoo 2 SLI wasn't a dumb idea after all 🤣

Just one thing to say about BIOS optimisation : Why would you activate Shadow Video Bios and Cacheable Video Bios for Windows specific tests only ?
Even if it's negligible, in most cases it will just slow down performances a bit, even used in DOS system only it does not always improves framerate and can reduces it instead.
Also, I don't know if you were able to change it or not, but system run a bit faster when spread spectrum is disable.
Disable COM1/COM2 and LPT is also a good idea for freeing up ressources.

I Think that all of that were just some minor omissions because you had a very big amount of hardware to test, just wanted to point it because I'm an annoying person 🤣

It's for sure some solid informations for those who hesitate about what to use for a Voodoo 2 dedicated rig. 😀

Can you just try for fun "if possible" (because I don't crearly see any improvement between P III 800 an P III 1.4 with V2 SLI @ 90 Mhz in 1024x768 in the benchs) if underclock a V2 SLI @ 75 Mhz gives a significent boost
with the PIII 1.4 over the PIII 800 ? (for me it does run faster at 75 than @ 90 Mhz with a PIII 800, I gain +- 166 pts under 3dMark00 default settings 1024x768).
I will try some timedemos just to see if it really makes a difference with games (that could be fun 🤣) Less heat, a bit faster and increasing life expectancy, that would be a really cool thing if totally averred (It's also possible that it can help PII or K6's cpu's as well) . 😎

Reply 4 of 37, by Gamecollector

User metadata
Rank Oldbie
Rank
Oldbie

Interesting.
Can you add info about a MiniGL used with GLQuake/Quake II. 1.47/1.49/other?
Plus only 3 games are using glide, Incoming/Forsaken are d3d only. So - if someone try to extend yours results to Win2k/Xp - there are troubles.
Tnx for this testing.

Asus P4P800 SE/Pentium4 3.2E/2 Gb DDR400B,
Radeon HD3850 Agp (Sapphire), Catalyst 14.4 (XpProSp3).
Voodoo2 12 MB SLI, Win2k drivers 1.02.00 (XpProSp3).

Reply 5 of 37, by Mau1wurf1977

User metadata
Rank l33t++
Rank
l33t++

@ mwdmeyer

Thanks 😀 Yes at high resolutions it doesn't take a powerful CPU. However in games often later levels are more demanding with more enemies. So having a bit of a buffer in regards to CPU performance wouldn't hurt.

@darksheer

Regarding the BIOS, I just load the BIOS defaults for any test and benchmarks. Not much of a tuner. But if you really want a nice boost, especially on BX440, is get some CL2 memory and tighten the timings. Instant boost 😀

PIII 800EB is very powerful, a solid choice for V2 SLI.

You know usually I disable COM and LPT.

Ok I tried 75 MHz on V2 SLI with Tualatin 1400, performance dropped as expected.

@Artex

Thanks 😀

My website with reviews, demos, drivers, tutorials and more...
My YouTube channel

Reply 6 of 37, by Mau1wurf1977

User metadata
Rank l33t++
Rank
l33t++
Gamecollector wrote:
Interesting. Can you add info about a MiniGL used with GLQuake/Quake II. 1.47/1.49/other? Plus only 3 games are using glide, Inc […]
Show full quote

Interesting.
Can you add info about a MiniGL used with GLQuake/Quake II. 1.47/1.49/other?
Plus only 3 games are using glide, Incoming/Forsaken are d3d only. So - if someone try to extend yours results to Win2k/Xp - there are troubles.
Tnx for this testing.

I knew that the choice of games will never please everyone. I used a mix of D3D and glide games covering a range of release years. Also they are easy to benchmark and scale well. I found it quite hard to find information on how to benchmark old games. But I hear you, there is always improvement to add more games and tests. But this project is simply too large for that.

I might do a V2 SLI vs. V3 project using just two or three processors and expand the range. What games would you have liked to see?

The other thing is that a lot of games people mention are very new and straight away GPU bound. While V2 SLI is quite decent it only took a year or two for games to demand a lot more graphics performance.

I put a screenshot of the version numbers for GLQuake at the end. I hope this is what you're looking for.

Quake II is the original disc version. Version number also at the end of the document. It comes with everything, no need to install a MiniGL driver yourself.

If this doesn't answer your question please let me know how / what I need to do to look it up 😀

My website with reviews, demos, drivers, tutorials and more...
My YouTube channel

Reply 7 of 37, by darksheer

User metadata
Rank Member
Rank
Member
Mau1wurf1977 wrote:

Ok I tried 75 MHz on V2 SLI with Tualatin 1400, performance dropped as expected.

Thanks, will continue to use them u/c with my PIII 800 though, since I don't seem to see major differences and want them to last longer 😊

Reply 8 of 37, by Mau1wurf1977

User metadata
Rank l33t++
Rank
l33t++

In Incoming and Quake 2 at 1024 x 768 the fps dropped from over 60 to under 60.

My website with reviews, demos, drivers, tutorials and more...
My YouTube channel

Reply 9 of 37, by darksheer

User metadata
Rank Member
Rank
Member

Just remembered why I used 75 Mhz instead of 90 Mhz by looking my games settings... Ok I'm officialy super dumb, I forget that I was running my most demanding 3dfx games at 800x600 (graphics all the way up in-game) with voodoo variables tweaked for improving quality in the registry. It looks like 1024x768 with voodoo variables set for performance and has a overall more enjoyable framerate.
Just check with utbench :
UT at 1024x768 @90 Mhz Perf reg => Min 30 Max 54 Average 49
UT at 800x600 @75 Mhz Quality reg => Min 40 Max 60 Average 48
Shame on me 🤣

Reply 10 of 37, by meljor

User metadata
Rank Oldbie
Rank
Oldbie

AMAZING!!!!

Very nice reading and very complete. Funny that the cards stop scaling full fps at an early point but with every faster cpu they keep on going and add 0.2fps or something like that. And just don`t stop doing it right up to to the tualatin 1400.

Now i understand why you wanted to test them on p4`s 😀

Good to see that i made the right choice pairing my sli setup with a k6-3+@600. At 1024x768 it is a good match.

So much work....... very nicely done, thank you! We all do benchmarks and testing but this kind of stuff is KING.

(p.s. p3 sli scaling is missing some p3 cpu names, numbers are there but names are missing)

asus tx97-e, 233mmx, voodoo1, s3 virge ,sb16
asus p5a, k6-3+ @ 550mhz, voodoo2 12mb sli, gf2 gts, awe32
asus p3b-f, p3-700, voodoo3 3500TV agp, awe64
asus tusl2-c, p3-S 1,4ghz, voodoo5 5500, live!
asus a7n8x DL, barton cpu, 6800ultra, Voodoo3 pci, audigy1

Reply 12 of 37, by Mau1wurf1977

User metadata
Rank l33t++
Rank
l33t++

@darksheer That looks much better 😀

I found that some drivers set the clock speed to 93 MHz. This would always cause my system to lock up. So worth checking in either the registry or via V2 tools.

@meljor

Thanks! Yea V2 is a good match for a fast K6-III+. Shame about the P4. The platform is so reliable and easy to work with.

@F2bnp

Thanks! I learnt a lot. Whatever project I am doing next will be easier as I am more comfortable with creating the charts and all of that.

My next project is likely a V2 vs. V3 paper.

My website with reviews, demos, drivers, tutorials and more...
My YouTube channel

Reply 13 of 37, by Robin4

User metadata
Rank l33t
Rank
l33t

I didnt knew that katmai vs coppermine did made a difference. But swapping a katmai for an coppermine (i dont think thats a good idea) Because for that little kind difference, i think i dont have to worry to much.

Only on two or three points it would be faster..I can swap to coppermine, but otherwise these katmais would only laying on the shelf.. That would be a waste to my opinion.

~ At least it can do black and white~

Reply 14 of 37, by Mau1wurf1977

User metadata
Rank l33t++
Rank
l33t++
Robin4 wrote:

I didnt knew that katmai vs coppermine did made a difference. But swapping a katmai for an coppermine (i dont think thats a good idea) Because for that little kind difference, i think i dont have to worry to much.

Only on two or three points it would be faster..I can swap to coppermine, but otherwise these katmais would only laying on the shelf.. That would be a waste to my opinion.

The main difference is power draw and heat. Coppermine is built using a smaller process and runs on a lower voltage. But in terms of performance I wouldn't worry too much 😀

My website with reviews, demos, drivers, tutorials and more...
My YouTube channel

Reply 15 of 37, by JoeCorrado

User metadata
Rank Member
Rank
Member

Wow, Phil !!

To say that you did an excellent job here seems almost an understatement. Quite impressive! And the kind of stuff that would (could) only be produced by somebody who is truly dedicated to the subject.

People will be saying thank you for your efforts in producing, assembling. and presenting this wealth of information in a single place for years to come-

Allow me to be one of the first. Thank you! It is an excellent resource... I know that you had fun doing it! Good for you, and lucky for us 🤣

-- Regards, Joe

Expect out of life, that which you put into it.

Reply 17 of 37, by retrofanatic

User metadata
Rank Oldbie
Rank
Oldbie

I just did a quick read...absolutely incredible work Phil! Let me be one of the first to thank you as well!

So much valuable information!

When i get some time to revisit my 815e build this document will help greatly in selecting an ideal cpu for my setup.

Thanks again.

Reply 18 of 37, by Kahenraz

User metadata
Rank l33t
Rank
l33t

Good read. Also noticed an error: missing processor-type on the left side of page 17.

This makes a lot of sense as there is a fair amount of setup that must take place before the information is pushed to the GPU because there was no no API-aware file format like .x for Glide or OpenGL.

Additionally, there was no hardware TnL on these chips which meant that the more polygons on the screen translated into a processor-bound scenario where more time would be spent on the CPU than the GPU. This was the boon for DirectX 7 where transformation and lighting was essentially a "free" operation which meant that the application overhead dropped down to the efficiency of the program code.

TnL isn't noticable unless you start doing things with a lot of polygons and most games were already programmed defensively prior to its introduction. Additionally, you had to actually target DirectX 7 to enable the TnL unit on the card; which is why you it didn't really provide any tangible benefit upon introduction until people actually started using it.

If you were to find a Glide game which had less polygons, rather than the more taxing games you had chosen, then you would see scaling further down the list and into the slower processors.

Some history to go along with your report. 😀

Reply 19 of 37, by Mau1wurf1977

User metadata
Rank l33t++
Rank
l33t++

Fixed page 17, thank you for spotting this 😀

I remember testing Turok in Glide mode. That game would scale to the moon! Didn't have any SLI scaling though...

What I'm finding is that most people are really only interested in the 800 x 600 and 1024 x 768 results. Kinda makes sense 🤣 I guess these are the resolutions I will focus on for my next project.

My website with reviews, demos, drivers, tutorials and more...
My YouTube channel