Just did some RAM tweaking. DRAM Timing 70ns -> 60ns, Fast RAS to CAS Delay 3 -> 2, Fast MA to RAS# Delay 2 Clks -> 1 Clk, SDRAM CAS latency/RAS-to-CAS 3/3 -> 2/2, SDRAM Speculative Read Off -> On.

Final Reality showed no difference, results were within margin of error. Half-Life felt subjectively just a tiny bit smoother but it could have been my imagination, or some other factor I'm unaware of.

Now for the strange part: Every time I open speedsys with reboot to MS-DOS I get that same 124.67 MB/s memory throughput, but when I go into DOS with ctrl while booting Windows I get 132.72 MB/s. It can't be a concidence since I did each benchmark three times.

d1stortion wrote:

Just ran Speedsys 4.78 w/ caches disabled: got a CPU score of 13.89. DX-33 to DX2-40 range? With L2 on I get 36.39 which comes close to a 5x86-133. Now that I think about it, I really shouldn't be this using this version since it's MMX aware... no wonder the results are skewed, when I tried Descent in Windows w/ both caches disabled it wasn't even fast enough to pass the MIDI data properly. When I ran the game in DOS it freezed with an "HMI Driver Not Found" error.

I think you have too much RAM for decent.

Speedsys is not reliable for comparing slowed down computers. 3dbench works very well, so does PCPBENCH, DOOM and Quake.

With LCDs it's not easy because the 4:3 screens are usually very old panels, or expensive and too big (1600 x 1200). The 5:4 models are the easiest to find, cheap and newer. Not perfects but good enough IMO.

Running from DOS w/ ctrl menu selection:

3DBENCH2: 16.6 with no cache and a freeze. After that I set back all the RAM tweaks and got 14.0. Results are reproducible, so the tweaks did have an effect after all. 36.7 with L2 on.

PCPBENCH: 1.9 with no cache, 5.2 with L2 on. Dead on DX2-66 if I believe the chart. Wonder how it compares to an actual 486 though, i.e. would the framerate be as consistent/inconsistent etc...

d1stortion wrote:

Running from DOS w/ ctrl menu selection:

3DBENCH2: 16.6 with no cache and a freeze. After that I set back all the RAM tweaks and got 14.0. Results are reproducible, so the tweaks did have an effect after all. 36.7 with L2 on.

PCPBENCH: 1.9 with no cache, 5.2 with L2 on. Dead on DX2-66 if I believe the chart. Wonder how it compares to an actual 486 though, i.e. would the framerate be as consistent/inconsistent etc...

Oh forgot to mention. For slow scores, 3dbench 1.0 should be used. 1.0 is unreliable for high scores, 1.0c is unreliable for slow scores.

But the scores look pretty good.

No more comments/suggestions? 🙁

I thought you were satisfied with how you have setup the system, is there something bothering you?

I don't think there is something like perfect satisfaction with retro builds 🤣

Nah, I'm overall satisfied, just still a bit curious why Turok underperforms compared to the tomshardware review... but maybe 10 FPS is the expected difference between a TX platform with 512K and a VIA chipset with 1M cache.

I never trust any of the benchmarks I see that were reported back when the hardware was new...
My mobo of choice for p1 builds is the HOT-591p : it's MVP3 based with 512kb cache.
I also have a similar mobo with MVP3 and 1MB cache: a P5BV3+

I did some testing and I didn't really found that the extra cache made a difference... with the same cpu/vga the difference was nothing important

Confirmed now that it's just the left USB port on the bracket that doesn't work... quite weird really. Need to see about fixing this.

I also was honestly impressed when I tried the Forsaken demo on the ATi card. It runs solid 15-25 FPS with bilinear filtering off in 640x480 and looks correct save for the blur effects and glitchy perspective correction. This is one well coded D3D game no doubt. It even looks like VSync is on so probably it drops frames cuz of it yet obviously not as bad as Voodoo. For that time, not too bad for a budget card running a 1 year newer game I think. Filtering costs 5-9 FPS or so on this chip so not worth it especially considering the card can't do it in decent quality anyway. This is probably the catch with this card as at least in this game it makes the difference between playable and unplayable. I think I have to take my opinion on it being useless for 3D back, even though other options would be still preferable 🤣

Should maybe try some CIF stuff on it since I have the card anyway. Fun with glitchy drivers etc 😀

At a bit of a loss with OpenGL on this PC... 3dfx MCD works fine but can't get the ICD to work. Quake 3 fails to start with a "could not load OpenGL subsystem" message, for instance. Things I tried to fix it include reinstalling all video card drivers, trying Diamond's Monster 3D drivers (which suck hard, they don't allow setting maximum refresh rate and disabling Direct3D acceleration) and copying 3dfxogl.dll into Windows\System.

IMO you are playing way too "new" games on this machine.

Easy of of use and excellent DOS compatibility is the strength of Socket 7.

Even Quake 2 I wouldn't play on this machine but move to a Slot 1 BX440 PC.

I haven't done any real playing on this PC yet out of time constraints... but I also do like comparing hardware of different periods, quick testing, seeing how far you could go and what would have been playable on what hardware. I know that all the DOS games will run for the most part anyway, so when testing it is usually more interesting how later stuff Windows runs 😀

OK, now this is interesting... when I inserted the RAM in another slot it actually got detected as 128 MB. It's a Hyundai PC100 stick, presumably from November 2000. With some RTFM I see that the original placement is undocumented, but it looks like a good way to limit RAM amount... the previous owner could have thought that 32 MB is faster on this board than 128.