VOGONS


More DOS PCI Benchmarks

Topic actions

First post, by clueless1

User metadata
Rank l33t
Rank
l33t

Following in kanecvr's footsteps, I got two new PCI cards this week: the Millennium II and S3 Virge 325, so I decided to rebench all my PCI cards with Phil's new DOS benchmark kit. I haven't gotten the hang of quickly producing nice looking tables and graphs, so I present you with a simple spreadsheet:

DOSVGABENCH2.jpg
Filename
DOSVGABENCH2.jpg
File size
101.87 KiB
Views
2222 views
File comment
Synthetic and Game Benches
File license
Fair use/fair dealing exception
DOSGAMEBENCH.jpg
Filename
DOSGAMEBENCH.jpg
File size
65.77 KiB
Views
2222 views
File comment
Game Benches
File license
Fair use/fair dealing exception

The system is a Packard Bell Multimedia C110. It has a 200Mhz Pentium Overdrive MMX and 32MB of 72-pin edo. PCI Burst is enabled in the BIOS. This system's main weaknesses are lack of L2 cache and an older (Triton) chipset. The same cpu in an MVP would probably be significantly quicker overall.

No real surprises. The Voodoo3 and Millennium II perform very well, and nearly identically. The TNT2 M64 cements itself as the clear leader of the pack, with exceptional performances in 3DBench, Wolfenstein 3D, Doom, and Quake 640x480. It also is a couple of steps ahead in the SVGA portions of Chris' and PC Player. EDIT: The Rage 128 GL performs very similarly to the Voodoo3 and Millennium II.

The Virge is just a step or two behind the Mil2 and V3, and the Cirrus Logic brings up the rear. But honestly, in VGA resolutions, the CL performs perfectly fine, not noticeably slower than the others. It really struggles in Quake 640x480 compared to the others, but really, none of them are performing all that great due to the system being a bottleneck.

Thanks for reading!

EDIT: added ATI Rage 128 GL PCI and included Speedsys VESA Memory Speed in the first chart, also added a 2nd chart with only game benchmarks, including Duke Nukem 3D and Descent 2. In the 2nd chart, I took out the S3 and CL results without VBE2.0 TSR.

Last edited by clueless1 on 2016-12-17, 16:44. Edited 3 times in total.

The more I learn, the more I realize how much I don't know.
OPL3 FM vs. Roland MT-32 vs. General MIDI DOS Game Comparison
Let's benchmark our systems with cache disabled
DOS PCI Graphics Card Benchmarks

Reply 1 of 23, by Imperious

User metadata
Rank Oldbie
Rank
Oldbie

All Hail The TNT2 M64 !!

Atari 2600, TI994a, Vic20, c64, ZX Spectrum 128, Amstrad CPC464, Atari 65XE, Commodore Plus/4, Amiga 500
PC's from XT 8088, 486, Pentium MMX, K6, Athlon, P3, P4, 775, to current Ryzen 5600x.

Reply 6 of 23, by brostenen

User metadata
Rank l33t++
Rank
l33t++

Hmmm....
I might pull my Creative TNT2-M64 out of my P-III Compaq deskpro EN then!!
The problem now, is what to put into that machine instead.

Don't eat stuff off a 15 year old never cleaned cpu cooler.
Those cakes make you sick....

My blog: http://to9xct.blogspot.dk
My YouTube: https://www.youtube.com/user/brostenen

001100 010010 011110 100001 101101 110011

Reply 7 of 23, by elianda

User metadata
Rank l33t
Rank
l33t
James-F wrote:

TNT2 M64 FTW!
It's got speed, compatibility, funky text modes, VBE3.0... one of the best PCI video cards for DOS gaming.

While it has VBE3.0 it supports only 8,16 and 32 bit modes.
A typical late S3 Virge supports 8,15,16,24 and 32 bit modes with it's VESA BIOS including the low res modes.

VESAINFO of a S3 Virge/GX2

Super VGA BIOS extension version 2.00 is installed
OEM string is 'S3 Incorporated. 86C357'
Device capabilities are 0

Mode Attr Window WindowA WindowB Mapping Bytes Resolution Char BPP Banks Type
Grn Siz Seg att Seg att Function /line Cell Planes

0109 IBCT 32 32 B800 RW C000:67EB 264 132x25 8x16 4 4 1 Text
010A IBCT 32 32 B800 RW C000:67EB 264 132x43 8x8 4 4 1 Text
012E I CG 64 64 A000 RW C000:67EB 320 320x200 8x8 8 1 1 VGA
010D I CG 64 64 A000 RW C000:67EB 640 320x200 8x8 15 1 1 06
010E I CG 64 64 A000 RW C000:67EB 640 320x200 8x8 16 1 1 06
0128 I CG 64 64 A000 RW C000:67EB 960 320x200 8x8 24 1 1 06
010F I CG 64 64 A000 RW C000:67EB 1280 320x200 8x8 32 1 1 06
0131 I CG 64 64 A000 RW C000:67EB 320 320x240 8x8 8 1 1 VGA
0132 I CG 64 64 A000 RW C000:67EB 640 320x240 8x8 15 1 1 06
0133 I CG 64 64 A000 RW C000:67EB 640 320x240 8x8 16 1 1 06
0135 I CG 64 64 A000 RW C000:67EB 960 320x240 8x8 24 1 1 06
0134 I CG 64 64 A000 RW C000:67EB 1280 320x240 8x8 32 1 1 06
0141 I CG 64 64 A000 RW C000:67EB 400 400x300 8x8 8 1 1 VGA
0142 I CG 64 64 A000 RW C000:67EB 800 400x300 8x8 15 1 1 06
0143 I CG 64 64 A000 RW C000:67EB 800 400x300 8x8 16 1 1 06
0145 I CG 64 64 A000 RW C000:67EB 1200 400x300 8x8 24 1 1 06
0144 I CG 64 64 A000 RW C000:67EB 1600 400x300 8x8 32 1 1 06
0151 I CG 64 64 A000 RW C000:67EB 512 512x384 8x16 8 1 1 VGA
0152 I CG 64 64 A000 RW C000:67EB 1024 512x384 8x1615 1 1 06
0153 I CG 64 64 A000 RW C000:67EB 1024 512x384 8x1616 1 1 06
0155 I CG 64 64 A000 RW C000:67EB 1536 512x384 8x1624 1 1 06
0154 I CG 64 64 A000 RW C000:67EB 2048 512x384 8x1632 1 1 06
0100 I CG 64 64 A000 RW C000:67EB 640 640x400 8x16 8 1 1 VGA
011C I CG 64 64 A000 RW C000:67EB 1280 640x400 8x1615 1 1 06
011D I CG 64 64 A000 RW C000:67EB 1280 640x400 8x1616 1 1 06
0129 I CG 64 64 A000 RW C000:67EB 1920 640x400 8x1624 1 1 06
011E I CG 64 64 A000 RW C000:67EB 2560 640x400 8x1632 1 1 06
0101 I CG 64 64 A000 RW C000:67EB 640 640x480 8x16 8 1 1 VGA
0110 I CG 64 64 A000 RW C000:67EB 1280 640x480 8x1615 1 1 06
0111 I CG 64 64 A000 RW C000:67EB 1280 640x480 8x1616 1 1 06
012A I CG 64 64 A000 RW C000:67EB 1920 640x480 8x1624 1 1 06
0112 I CG 64 64 A000 RW C000:67EB 2560 640x480 8x1632 1 1 06
0102 IBCG 64 64 A000 RW C000:67EB 100 800x600 8x8 4 4 1 EGA
0103 I CG 64 64 A000 RW C000:67EB 800 800x600 8x8 8 1 1 VGA
0113 I CG 64 64 A000 RW C000:67EB 1600 800x600 8x8 15 1 1 06
0114 I CG 64 64 A000 RW C000:67EB 1600 800x600 8x8 16 1 1 06
012B I CG 64 64 A000 RW C000:67EB 2400 800x600 8x8 24 1 1 06
0115 I CG 64 64 A000 RW C000:67EB 3200 800x600 8x8 32 1 1 06
0104 I CG 64 64 A000 RW C000:67EB 128 1024x768 8x16 4 4 1 EGA
0105 I CG 64 64 A000 RW C000:67EB 1024 1024x768 8x16 8 1 1 VGA
0116 I CG 64 64 A000 RW C000:67EB 2048 1024x768 8x1615 1 1 06
0117 I CG 64 64 A000 RW C000:67EB 2048 1024x768 8x1616 1 1 06
012C I CG 64 64 A000 RW C000:67EB 3072 1024x768 8x1624 1 1 06
0118 I CG 64 64 A000 RW C000:67EB 4096 1024x768 8x1632 1 1 06
0161 I CG 64 64 A000 RW C000:67EB 1152 1152x864 8x16 8 1 1 VGA
0162 I CG 64 64 A000 RW C000:67EB 2304 1152x864 8x1615 1 1 06
0163 I CG 64 64 A000 RW C000:67EB 2304 1152x864 8x1616 1 1 06
0165 I CG 64 64 A000 RW C000:67EB 3456 1152x864 8x1624 1 1 06
0164 I CG 64 64 A000 RW C000:67EB 4608 1152x864 8x1632 1 1 06
0106 I CG 64 64 A000 RW C000:67EB 160 1280x1024 8x16 4 4 1 EGA
0107 I CG 64 64 A000 RW C000:67EB 1280 1280x1024 8x16 8 1 1 VGA
0119 I CG 64 64 A000 RW C000:67EB 2560 1280x1024 8x1615 1 1 06
011A I CG 64 64 A000 RW C000:67EB 2560 1280x1024 8x1616 1 1 06
Show last 5 lines
012D I CG  64  64 A000 RW         C000:67EB 3840 1280x1024  8x1624 1 1     06
0120 I CG 64 64 A000 RW C000:67EB 1600 1600x1200 8x16 8 1 1 VGA
0121 I CG 64 64 A000 RW C000:67EB 3200 1600x1200 8x1615 1 1 06
0122 I CG 64 64 A000 RW C000:67EB 3200 1600x1200 8x1616 1 1 06

This is a really weak spot of the NVidia cards. They were released at a time where DOS (VBE) support got weaker.
For DOS the gain of VBE3 over VBE2 is negligible since the only feature really employed there was setting custom refresh rates. However before cards brought their own tools for this (like S3Refresh or CLMode etc.).

Retronn.de - Vintage Hardware Gallery, Drivers, Guides, Videos. Now with file search
Youtube Channel
FTP Server - Driver Archive and more
DVI2PCIe alignment and 2D image quality measurement tool

Reply 8 of 23, by clueless1

User metadata
Rank l33t
Rank
l33t

@elianda

How would those missing modes effect the TNT2 negatively in actual use?

I had to run Univbe 5.3a with the S3 and CL in order to get access to 640x480 in Quake. The other cards natively supported that res.

The more I learn, the more I realize how much I don't know.
OPL3 FM vs. Roland MT-32 vs. General MIDI DOS Game Comparison
Let's benchmark our systems with cache disabled
DOS PCI Graphics Card Benchmarks

Reply 9 of 23, by elianda

User metadata
Rank l33t
Rank
l33t
clueless1 wrote:

@elianda
How would those missing modes effect the TNT2 negatively in actual use?

Applications that expect this VESA mode will not work or fallback to a compatibility mode.
Or you have to load an additional TSR (like S3VBE20 / UniVBE) to add those modes.

Retronn.de - Vintage Hardware Gallery, Drivers, Guides, Videos. Now with file search
Youtube Channel
FTP Server - Driver Archive and more
DVI2PCIe alignment and 2D image quality measurement tool

Reply 10 of 23, by clueless1

User metadata
Rank l33t
Rank
l33t
elianda wrote:
clueless1 wrote:

@elianda
How would those missing modes effect the TNT2 negatively in actual use?

Applications that expect this VESA mode will not work or fallback to a compatibility mode.
Or you have to load an additional TSR (like S3VBE20 / UniVBE) to add those modes.

Sorry, I meant specifically which applications uses these modes so I can see if I would be effected. If I never will use these, then it should not be an issue for me.

The more I learn, the more I realize how much I don't know.
OPL3 FM vs. Roland MT-32 vs. General MIDI DOS Game Comparison
Let's benchmark our systems with cache disabled
DOS PCI Graphics Card Benchmarks

Reply 11 of 23, by PhilsComputerLab

User metadata
Rank l33t++
Rank
l33t++

That's really it, it's all about what games are you actually playing.

These old Keen games always get mentioned, I never play them 😀

But I do play Wing Commander III, and a lot of cards have this odd flicker when you go from room to room (it plays animations). There is an alternate video mode in the setup, that sometimes fixes the issue. I don't think I tried the TNT2, but on a FX, you always get a flicker. On a V3, with the alternate video mode, it is perfect.

In the Alone in the Dark installation, the text is corrupted. It uses a font, and that can be easily fixed by loading a tool such as FIX8X14 - 8x14 font fixer will resolve this. Again, my FX needs this tool, I don't know about the TNT2.

But yea, everyone has their own little test suite and that's good. Gona's VGA Matrix: http://gona.mactar.hu/DOS_TESTS/ is awesome, but pretty much none of the games interest me really 😊

YouTube, Facebook, Website

Reply 12 of 23, by clueless1

User metadata
Rank l33t
Rank
l33t
PhilsComputerLab wrote:
That's really it, it's all about what games are you actually playing. […]
Show full quote

That's really it, it's all about what games are you actually playing.

These old Keen games always get mentioned, I never play them 😀

But I do play Wing Commander III, and a lot of cards have this odd flicker when you go from room to room (it plays animations). There is an alternate video mode in the setup, that sometimes fixes the issue. I don't think I tried the TNT2, but on a FX, you always get a flicker. On a V3, with the alternate video mode, it is perfect.

In the Alone in the Dark installation, the text is corrupted. It uses a font, and that can be easily fixed by loading a tool such as FIX8X14 - 8x14 font fixer will resolve this. Again, my FX needs this tool, I don't know about the TNT2.

But yea, everyone has their own little test suite and that's good. Gona's VGA Matrix: http://gona.mactar.hu/DOS_TESTS/ is awesome, but pretty much none of the games interest me really 😊

I remember seeing Gona's chart before, but looking at it with a fresh eye, my S3 325 is about as compatible as there is on there. And not that much slower than the M64. But like you say, I'm not playing the games that have issues anyway. Thanks for linking that chart, Phil.

The more I learn, the more I realize how much I don't know.
OPL3 FM vs. Roland MT-32 vs. General MIDI DOS Game Comparison
Let's benchmark our systems with cache disabled
DOS PCI Graphics Card Benchmarks

Reply 13 of 23, by clueless1

User metadata
Rank l33t
Rank
l33t

I forgot that I had an S3 Trio64 V2/DX, so I benched it and added it to the chart in the OP. With the exception of Doom, it's noticeably faster than the Virge. I also separated out the S3 and CL results with and without VESA BIOS Extensions. So they get a ZERO on Quake 640x480 without VBE, and a new line item with the VBE TSR enabled.

The more I learn, the more I realize how much I don't know.
OPL3 FM vs. Roland MT-32 vs. General MIDI DOS Game Comparison
Let's benchmark our systems with cache disabled
DOS PCI Graphics Card Benchmarks

Reply 14 of 23, by clueless1

User metadata
Rank l33t
Rank
l33t

I added an ATI Rage 128 GL PCI to the chart in the OP. With the exception of Quake 640x480, it's a very fast card, about the same as the Voodoo3 and Millennium II. But it's about 11% slower in Quake 640x480 (12.7 fps vs. 14.3).

I also included Speedsys VESA memory speeds. They correlate very closely to the rest of the benchmark results.

Lastly, I added a separate chart with only game benchmarks, including Duke Nukem and Descent 2 in various resolutions. In Descent 2, 320x200 and 640x480 have the cockpit on. The other (cockpit off) resolutions show more differentiation between the cards, probably because the static cockpit reduces the amount of moving pixels that need to be rendered. So you will get a better idea comparing the cards by using the non-cockpit resolutions.

Some interesting notes:
1) the onboard CL-GD5430 is consistently 7.5% faster (8.6 fps vs 8.0) in Quake 640x480 with 2MB vs. 1MB.
2) the onboard CL-GD5430 gets consistently lower fps in PCPBench 640x480 with VESA 2.0 compared to VESA 1.2, using Scitech Display Doctor 5.3a.
3) the Millennium II cannot render Descent 2 in 800x600. The screen scrambles like an old TV channel with unadjusted antennas. I tried with and without UNIVBE 6.7, with no difference.
4) The Rage 128 GL is unusually slow in Quake 640 and Duke 640 and 800, but very fast in Descent 2 640 and 800.
5. Trio64 and Virge perform nearly identically in Descent 2 with the cockpit turned on, but when you turn the cockpit off the Trio64 pulls ahead, especially in 800x600.

The more I learn, the more I realize how much I don't know.
OPL3 FM vs. Roland MT-32 vs. General MIDI DOS Game Comparison
Let's benchmark our systems with cache disabled
DOS PCI Graphics Card Benchmarks

Reply 15 of 23, by feipoa

User metadata
Rank l33t++
Rank
l33t++

For Descent 2 and Duke Nukem, how are you determining the frame rate? I don't recall these games having timedemos. Are you taking an instantaneous fps value in-game?

For the Rage 128 GL, did you use a Mac version and flash it with a PC BIOS? All the Rage 128 GL cards I could find when I was looking for them years back were Macintosh versions.

Plan your life wisely, you'll be dead before you know it.

Reply 16 of 23, by clueless1

User metadata
Rank l33t
Rank
l33t

Descent 2 and Duke Nukem have framerate displays. For Descent, type 'frametime' in-game; for Duke, type 'dnrate' in-game. Then after framerate shows, abort game and start new game so that placement in-game is consistent between runs. There is no timedemo, unfortunately. I've done this for a long time, but usually don't show the results when I post because people complain that it's not as accurate as a timedemo. But after Phil posted a video yesterday using these results, I felt encouraged to post my own results. Hey, it's better than nothing, right? And there are not a lot of DOS games that let you see framerates, so I'll take what I can get.

As far as I know, my Rage 128 GL is a PC version. But there is no specific marking on it other than "Rage 128". Speedsys identifies it as a Rage 128 GL, but when I install it in Win98, the driver detects it as Xpert 128. So who knows? I'll post pics of it in a bit.

edit: pics. the writing on the back is mine.

rage128gl_top.jpg
Filename
rage128gl_top.jpg
File size
1.2 MiB
Views
2194 views
File license
Fair use/fair dealing exception
rage128gl_bot.jpg
Filename
rage128gl_bot.jpg
File size
1.6 MiB
Views
2194 views
File license
Fair use/fair dealing exception

The more I learn, the more I realize how much I don't know.
OPL3 FM vs. Roland MT-32 vs. General MIDI DOS Game Comparison
Let's benchmark our systems with cache disabled
DOS PCI Graphics Card Benchmarks

Reply 17 of 23, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I was aware of the ability to display live framerates such as how you are doing it, but I thought that maybe you were able to hack in a timedemo with some trickery. Oh well...

I too plan to benchmark more games in a manner like you are doing.

Thanks for the photo. I have seen these on eBay, but they all say they are for Mac or Apple G3. How did you find your card?
http://www.ebay.com/itm/ATI-Rage-128-16MB-PCI … %3D262406040167

Plan your life wisely, you'll be dead before you know it.

Reply 18 of 23, by clueless1

User metadata
Rank l33t
Rank
l33t

I can't remember for sure how I found mine. 😊

I *think* it was in a box of hardware donated to me from a guy who used to run a PC repair shop in the late 1990's.

The more I learn, the more I realize how much I don't know.
OPL3 FM vs. Roland MT-32 vs. General MIDI DOS Game Comparison
Let's benchmark our systems with cache disabled
DOS PCI Graphics Card Benchmarks

Reply 19 of 23, by PhilsComputerLab

User metadata
Rank l33t++
Rank
l33t++

Nice stuff 😀

Yea the built-in FPS counters are a great addition to the already existing benchmarks. Like you said, they aren't perfect, but better than nothing and especially with different processors, I am seeing nice scaling.

My conclusions are similar, the CL seems to have a bit of a weakness at higher resolutions. I wonder if there are any CL made tools to speed things up?

The S3VBE20 tool is a must for higher resolutions on the S3 cards, but it does take up 4 KB of memory, at least it did on my machine. Two of my S3 cards have the black level bug, but thanks to switched on VOGONS members, there are fixes for this.

Other games with FPS counters are Wing Commander III and IV, although they are capped at 24 fps. Both are mostly CPU bound, especially inside the large ships.

What I'm seeing is that high resolution games seem to require quite a bit of CPU power, so much that usually there is an AGP interface in the machine.

YouTube, Facebook, Website