I'm prepared to do a "marathon" PCPBench test. Please can someone suggest a testing strategy for me to implement. Thanks a lot.

🤣 don't go out of your way just for me, I was asking to see if it had been done before and there was a list of results handy...

But if you're really keen, main thing it tests is video card performance- so you would stick with the same processor, use the same BIOS settings and leave all your cache enabled, and just swap video cards.

I found with PCPBENCH a lot of cards wouldn't work in the default mode without UNIVBE loaded.

If you have any VLB cards, I'd be ecstatic if you'd bench 2 or 3 of them and tell me the results 😀

Ah pity - I don't have any VLB boards any more. 🙁 I "standardised" on PCI a couple of years ago. I think Vogons user BastlerMike has lots of graphics cards. Perhaps he would be willing to bench some of his (VLB) graphics cards using PCPBench?

I have a working 486/100 or 5x86/133 setup with a couple of VLB cards iirc... will report what are they tomorrow and see if there is any interest in benchmarking them 😁

DonutKing wrote:

But if you're really keen, main thing it tests is video card performance-

That depends the chosen video mode and on the speed of the video card relative to the speed of bus and CPU.

Some video mode suggestions:

• Standard VGA (/VGAMODE parameter)
• 320x240x32 - Mode for high framerates, useful to find bottlenecks
• 640x480x8 - Similar to the default mode, but runs on more cards
• 640x480x16 - Well supported among VESA cards - even crappy ones
• 1024x768x32 - For fast CPUs and 4MB cards
• 1600x1200x32 - VESA torture mode 😈

At the early stages of my Retro experience, I did some PCPBENCHing as well. I posted the results underneath...

5u3 gave me the tip regarding FASTVID, which unlocks graphics performance on Slot 1 machines. I got quite a nice boost, so it's definitely recommended...

DonutKing wrote:

I'm wondering if anything similar has been done for PCPBench? Main reason I ask is because my 486 is pretty sluggish in SVGA modes

For SVGA I highly recommend a Slot 1 Pentium 3 system. Most of these games run on very fast machines (e.g. Duke Nukem 3D will run on a Phenom II Quadcore 3.2 GHz 🤣) Just like 3DBENCH, PCPBENCH is really a CPU benchmark, so changing your graphics card in your 486 won't do a thing...

The P3 1GHz score is quite impressive. I don't have a Tualatin capable adapter, but RG100 does 🤣 Seeing a P3 1.4 GHz results would be awesome. But yea you need to run FASTVID beforehand, otherwise it won't work. I believe I linked FASTVID in the beginning of this thread.

I was referring to the standard mode (from memory 640x400)- in my experience you seem to get wildly varying results simply by swapping videocards. I'm specifically interested in comparing video cards at these sorts of high resolutions (well, high for DOS games anyway 😜 )

The standard results are under "PCPBENCH". Well to me standard is the mode PCPBENCH uses when you start it without any options rights? Pretty sure it's 640 x 400...

Well if you want to compare video cards, then you got to do it like the hardware sites test video cards. Best cpu you can afford, OC it and bench the cards. That way you make sure the CPU isn't holding back and it's all up to the video card.

If you are testing AGP well it would have to be a P4 or AMD64 Muahahahahha

Well, that would be ideal... but you can do it with any CPU, as long as your cpu and the rest of the system remains consistent across benching runs simply swapping video cards will give you comparable results on the same system. If you start getting identical results with different cards then you can assume you have a bottleneck elsewhere such as the CPU or memory.

With PCPBENCH on my 486DX2 I found that I got varying results with different video cards but an otherwise identical system. I did record the results but I don't have them handy at the moment.

I'm specifically interested in VLB cards anyway in high res VESA modes, not PCI or AGP. This is why I was hoping there would be a database of these benchmarks somewhere 😀

I also looked... A Lot and didn't find any PCPBENCH results. It was a benchmark created by a German computer magazine (PC Player), so it never got a lot of use outside of Germany (until now it seems 🤣)...

VLB yea that's 486 territory right there 😀 I guess AMD 133 is the top dog in that league...

Thanks Keropi for offering to test some VLB video cards.

As I don't have that kind of hardware, would it be a good idea to run a series of PCPBench tests on a P3 @ 1.4 GHz (or perhaps 1 GHz and also 1.4 GHz), using a variety of video cards? I could select some PCI and AGP cards. I am due to sort out my retro boxes soon. I could put aside a selection of video cards, and then begin testing...

I'm interested on a 1.4 GHz result with a decent video card (AGP Geforce or something like that). But remember to load VIDSPEED beforehand! You can try a run without it and see that your scores are much lower.

The file is linked on the first page... If you need PCPBENCH let me know!

I have run a series of PcpBench tests. The hardware used was: Slot 1, Abit AB-BH6, Tualatin 1.4GHz CPU, Mobo+CPU: 133 FSB, Voodoo3 2000, 1 stick of 256MB PC-133 SDRAM. All tests automatically ran in "LFB mode". FastVid was not required. The first number in each row is PcpBench's mode value.

Edit: The scores below can be ignored. Please see my post below for the explanation. 😦

8 bit colour
100 (640x400) 65.8
101 (640x480) 54.9
103 (800x600) 35.9
105 (1024x768) 22.2
107 (1280x1024) 13.4

180 (320x200) 238.1
181 (320x240) 199.0
184 (400x300) 133.2
187 (512x384) 83.8

16 bit colour
18a (640x400) 52.0
111 (640x480) 43.3
114 (800x600) 28.2
117 (1024x768) 17.5
11a (1280x1024) 10.5

10e (320x200) 190.5
182 (320x240) 158.5
185 (400x300) 105.5
188 (512x384) 66.2

Oh dear. After I ran all of those tests above, I double-checked some similar benchmarks here on Vogons. They were a lot higher, for similar hardware. I removed my DOS HDD, and replaced it with a Win98 HDD, and reran PcpBench. The score obtained on the default 100 test was about 3 times higher. When I ran PcpBench under DOS, it said "LFB", and so I assumed that I didn't need FastVid or MTRRLFBE. I will rerun every test above, but run them on my Win98 HDD. Well, CF actually.

That's better. Running PcpBench under Win98SE seems to produce favourable results.

8 bit colour
100 (640x400) 188.4
101 (640x480) 161.1
103 (800x600) 117.7
105 (1024x768) 81.3
107 (1280x1024) 53.9

180 (320x200) 473.1
181 (320x240) 412.3
184 (400x300) 310.5
187 (512x384) 220.1

16 bit colour
18a (640x400) 134.5
111 (640x480) 114.8
114 (800x600) 81.5
117 (1024x768) 54.5
11a (1280x1024) 35.1

10e (320x200) 365.0
182 (320x240) 316.1
185 (400x300) 234.0
188 (512x384) 161.3

Edit: I just reread 5u3's PcpBench suggestions above. They include 32 bit colour modes. When I queried PcpBench for its available modes (/modes switch), it only listed 8 and 16 bit colour modes. Edit 2: Ah, I know why. I'm using a Voodoo3 card!

Should it be like that ? I mean we all know the Voodoo3 does 16-bits in 3D modes, but PCPBENCH doesn't use the 3D hardware, it's a VGA benchmark that uses VESA modes, and you should have 24 or 32-bit modes available, as you have them to set your Windows desktop resolution. Isn't the Voodoo3 VESA 3.0 capable ? Are you referring that you used the Windows version of PCPBENCH ? I didn't get that part. Are you using a PCI or AGP Voodoo3 ?

Also, to prevent PCPBENCH from using LFB modes, just use /NOLINEAR switch.

retro games 100 wrote:

That's better. Running PcpBench under Win98SE seems to produce favourable results.

That's because the windows video driver sets the MTRRs for faster framebuffer access. Under plain DOS you have to run FASTVID/MXK6OPT/MTRRLFBE to enable this feature.

retro games 100 wrote:

When I queried PcpBench for its available modes (/modes switch), it only listed 8 and 16 bit colour modes. Edit 2: Ah, I know why. I'm using a Voodoo3 card!

Yep. The Voodoo3/4/5 cards are not particularly good for this. Their VBE 3.0 BIOS supports 24bit truecolor modes, but PCPbench only tests 32-bit modes.

Impressive scores!

Honestly, I tried FastVid before I swapped my DOS HDD for a Win98 HDD. However, when I ran it, I said "Yes" to "Write Posting". I think that confused it, because I'm using an i440BX chipset mobo, and you don't need WP for this chipset. So, I tried FastVid again, and now I get better results if I say "No" to WP. For the Voodoo3, in 640x400 (mode 100) I get 192.7, which is a good score. I removed the V3 card, and replaced it with a GF3 card. Mode 100 now gets me 188.9. I continued with the GF3 for the following tests:

* Standard VGA = 372.0 (BTW, the display did not say "LFB")
* 320x240x32 = 283.3 (LFB)
* 640x480x8 = 162.4 (LFB)
* 640x480x16 = 115.6 (LFB)
* 1024x768x32 = 12.5 (The display did not say "LFB")
* 1600x1200x32 = No mode option available!

unmei220, please note that there is no Windows version of PcpBench. I simply ran the .exe from within Windows 98. Also, the Voodoo3 2000 card I was using was AGP.

Yes, there is a Windows version of PCPBENCH that test Direct3D. It's exactly the same as the DOS version (I mean, the landscape and all that. Also, it includes sound), except it's for Direct3D. I was referring to that one. Sorry if I didn't was more clear.