PC-Engineer wrote on 2020-09-02, 11:55:

I have got 26.8fps with V1.08. with the system in my signature. With SIS496 PCI setup i have got 26.6fps and with a UMC8881 setu […]
Show full quote

I have got 26.8fps with V1.08. with the system in my signature. With SIS496 PCI setup i have got 26.6fps and with a UMC8881 setup 26.3fps - all with 1024kB cache.

But the SV2GX4 had DMA issues with L1WB and some Windows 95 issues with transparent mode - so i switched back to L1WT and synchronize mode. With this setup i have got 25.1fps

The S3 864 was a little bit faster than the ET4000 - 26.8fps vs. 26.6fps. Maybe with a ARK VLB card or a Diamond Trio64 VLB you can gain a little bit higher values.

AND the version 1.08 has a improvement from appr. 7% over the 1.06. So the 26.8fps in V1.08 should result in 25.1fps in V1.06

What was your screen size set to in-game when testing? I wonder what’s holding the three mobos I’ve tested back other than a lower 256k cache?

feipoa wrote on 2020-09-02, 23:28:

The variance in DOS results is due to a few factors. […]
Show full quote

The variance in DOS results is due to a few factors.

a) Some people use v1.08 instead of the v1.06 demo version. v1.08 returns higher results.
b) Some people are using Phils benchmark pack, which if I recall right, has made some adjustments to settings which alters results.
c) Other installed hardware, particularly USB, and to a lesser extent network, cards may reduce the benchmark results.

Before Phils pack, most people would benchmark using v1.06 demo at 320x200 with sound disabled. After Phils pack, there has been more uncertainty in what people are using. I prefer the original approach, that is v1.06 demo with sound off, booting directly into DOS with few drivers loaded, no EMM386, or mouse drivers, etc.

For sure some inconsistencies in testing methodology. I’m using 1.08, no drivers loaded, no sound, screen size one notch away from max.

Using v1.06 is the norm, not v1.08.

Now that you mention it, I recall Quake being fairly selective to L2 cache size. Some nice charts showing how it scales from 256K up to 2048K. Re: Upgrading Socket 4 L2 cache to 2048k (ECS SI5PI MB)

Chadti99 wrote on 2020-09-03, 00:20:

PC-Engineer wrote on 2020-09-02, 11:55:

I have got 26.8fps with V1.08. with the system in my signature. With SIS496 PCI setup i have got 26.6fps and with a UMC8881 setu […]
Show full quote

I have got 26.8fps with V1.08. with the system in my signature. With SIS496 PCI setup i have got 26.6fps and with a UMC8881 setup 26.3fps - all with 1024kB cache.

But the SV2GX4 had DMA issues with L1WB and some Windows 95 issues with transparent mode - so i switched back to L1WT and synchronize mode. With this setup i have got 25.1fps

The S3 864 was a little bit faster than the ET4000 - 26.8fps vs. 26.6fps. Maybe with a ARK VLB card or a Diamond Trio64 VLB you can gain a little bit higher values.

AND the version 1.08 has a improvement from appr. 7% over the 1.06. So the 26.8fps in V1.08 should result in 25.1fps in V1.06

What was your screen size set to in-game when testing? I wonder what’s holding the three mobos I’ve tested back other than a lower 256k cache?

Full screen with UI, 320x240x8, nosound, no EMM386, V1.08

I ran about 2 dozen PCI graphics cards on my MB-8433UUD with a POD at 100 MHz. The best score I could achieve in Quake v1.06 was 25.9 fps. In Quake v1.08, that's 27.9 fps.

feipoa wrote on 2020-09-03, 13:22:

I ran about 2 dozen PCI graphics cards on my MB-8433UUD with a POD at 100 MHz. The best score I could achieve in Quake v1.06 was 25.9 fps. In Quake v1.08, that's 27.9 fps.

That’s a great score! What PCI Graphics cards would push the best scores? I’m testing with a Tseng Labs ET6000.

feipoa wrote on 2020-09-03, 00:51:

Using v1.06 is the norm, not v1.08.

Now that you mention it, I recall Quake being fairly selective to L2 cache size. Some nice charts showing how it scales from 256K up to 2048K. Re: Upgrading Socket 4 L2 cache to 2048k (ECS SI5PI MB)

That’s pretty cool data! I went ahead and ordered sram to bump up to 512k cache. Maybe I’ll try the 1024k mod eventually.

The cards performed about the same. I think the CPU is holding back the scores. I only tested what I have in my drawer. There's some others that I have boxed or in systems that would be nice to add to the list as well, but it is too much work for me to dig them out. I did notice that on this board, running L2 in WB mode slowed down the results (25.4 fps) vs. running it in WT mode (25.8 fps).

Excel chart attached.

I also ran some one of the fastest ISA cards, and if run at 13 MHz, the result isn't all that far off from the slowest PCI card.

The ATI Rate128 VR is probably the cheapest and easiest to acquire from the list of 25.9 fps cards.

feipoa wrote on 2020-09-04, 23:06:

The cards performed about the same. I think the CPU is holding back the scores. I only tested what I have in my drawer. There […]
Show full quote

The cards performed about the same. I think the CPU is holding back the scores. I only tested what I have in my drawer. There's some others that I have boxed or in systems that would be nice to add to the list as well, but it is too much work for me to dig them out. I did notice that on this board, running L2 in WB mode slowed down the results (25.4 fps) vs. running it in WT mode (25.8 fps).

Excel chart attached.

I also ran some one of the fastest ISA cards, and if run at 13 MHz, the result isn't all that far off from the slowest PCI card.

The ATI Rate128 VR is probably the cheapest and easiest to acquire from the list of 25.9 fps cards.

PCI_cards_POD100_Quake.png

Excellent info! I do have a Riva 128 PCI and it sometimes scores 0.1 FPS higher than the ET6000 but not always. I’ll try again with WT mode as you mentioned!

Hey you were right, WT mode for L2 cache was faster on this Biostar board. Gained another half frame!

The ATI Rate128 VR is probably the cheapest and easiest to acquire from the list of 25.9 fps cards.

Speedy/overclocked S3 cards with S3 Speed VGA tweak are easier to find.

PC-Engineer wrote on 2020-09-03, 04:44:

Chadti99 wrote on 2020-09-03, 00:20:

PC-Engineer wrote on 2020-09-02, 11:55:

I have got 26.8fps with V1.08. with the system in my signature. With SIS496 PCI setup i have got 26.6fps and with a UMC8881 setu […]
Show full quote

I have got 26.8fps with V1.08. with the system in my signature. With SIS496 PCI setup i have got 26.6fps and with a UMC8881 setup 26.3fps - all with 1024kB cache.

But the SV2GX4 had DMA issues with L1WB and some Windows 95 issues with transparent mode - so i switched back to L1WT and synchronize mode. With this setup i have got 25.1fps

The S3 864 was a little bit faster than the ET4000 - 26.8fps vs. 26.6fps. Maybe with a ARK VLB card or a Diamond Trio64 VLB you can gain a little bit higher values.

AND the version 1.08 has a improvement from appr. 7% over the 1.06. So the 26.8fps in V1.08 should result in 25.1fps in V1.06

What was your screen size set to in-game when testing? I wonder what’s holding the three mobos I’ve tested back other than a lower 256k cache?

Full screen with UI, 320x240x8, nosound, no EMM386, V1.08

You sure you don’t mean 320x200? What is full screen with UI?

Pleased to report the UMC chipset on this BioStar supports DMA transfer much better than SIS. My Verite 2200 is hitting 35.9 FPS in vQuake!

After Phils pack, there has been more uncertainty in what people are using

Didn't noticed any difference in results.

The Serpent Rider wrote on 2020-09-05, 16:08:

After Phils pack, there has been more uncertainty in what people are using

Didn't noticed any difference in results.

See attached. 23.3 fps w/Phil vs. 25.9 fps run direct.

I ran the Biostar with only 256K double-banked and the score drops from 25.9 (512K double-banked) to 25.1 fps.

I did know about VQuake back in 96. Nice to see the Donovan/Abrash port running on a POD.

Regarding Phil's suite...
Strange - is the screen size the same or maybe he is setting up 320x240?
Depending on your video card setting low res VESA mode 320x200 may be faster than the normal MCGA/VGA mode.
I always use the command line parameters "-nosound -nocdaudio -nojoy -nolan"

I never understood why Quake and other games were benchmarked with the sound turned off. Was it perhaps due to some soundcards causing issues and lowering performance compared to others? That's the only rational explanation I can think of.
Benchmarking should be as close to real world as possible and who'd want to play without sound and the awesome Trent Reznor soundtrack anyway? 😁

feipoa wrote on 2020-09-06, 10:47:

I ran the Biostar with only 256K double-banked and the score drops from 25.9 (512K double-banked) to 25.1 fps.

I can confirm this speed based on same board, same cache, same Quake version, and the same screen size setting. Feeling better that I’m not that far off the mark.

Garrett W wrote on 2020-09-06, 11:18:

I never understood why Quake and other games were benchmarked with the sound turned off. Was it perhaps due to some soundcards causing issues and lowering performance compared to others? That's the only rational explanation I can think of.
Benchmarking should be as close to real world as possible and who'd want to play without sound and the awesome Trent Reznor soundtrack anyway? 😁

Yah not sure, someone started the trend long ago. You aren’t playing Quake if you aren’t enjoying it with the original Soundtrack. Been waiting ages for the vinyl release.