You could build a Windows 95 / DOS gaming PC for the late DOS games / very early 3D games with a Voodoo card (not Voodoo 2). Here 64 MB is plenty. A MMX 233 should also work well.

I will just quickly comment on the P3 saying its a nice build 😀

I have a Compaq Presario 4766, which is the same model just a small localization difference (my had ISDN card, yours probably 33.6k modem). See link to build here: My Compaq Presario 4766 - DOS / Win9X machine
First off, you can install a K6-III which greatly increases performance (the + version does not work, you need the original 2.4V)! Maximum 80MB SDRAM and maximum 66mhz FSB. The integrated S3 Trio64+ card is actually very decent in DOS with great compatibility and speed. It has ESS 1888 integrated audio which works very well together with SoftMPU. It is also pretty quick in VGA in PCPBench, Quake and 3DBench compared to other Intel boards. See some benches here.

Alright. I just installed the K6-2 because it was there. The computer previously had a 166MMX that overclocked perfectly to 200MHz. The only reason I wanted more RAM was that 98 was swapping quite a bit more with that amount of RAM than I remember on 32MB, but I may be viewing my first actual PC with nostalgia-goggles.

Yea Windows 95 is a lot easier on RAM and resources. And ideal for these very early games right at the transition from DOS to Windows.

There is also the issue of cache-able RAM. Many boards have a 64 MB limit so it pays off to stay with 64 MB in such cases.

I didn't realize that the cacheable limit was that small, especially given a lot of the lovely SS7 builds in this forum. I just remember Windows 98 running way better on worse hardware, plus that computer specifically didn't come with Windows 95 in this case as I traded my old P4 for it (Good riddance!) Given that the DOS games that I'd play would be on the order of DOOM, Quake, Duke3d, SC2k, and Transport Tycoon, should I reinstall the Pentium or stay with the K6? It looks like more modern out of order architectures don't really run DOS games all that well, while in order architectures like the Pentium should.

If you get a AMD K6-III you don't have to worry about the cache limit as it has on-die L2 cache. I ran 80MB in my Compaq for years as I was unaware of it and I personally think the cache issue is highly overrated here on Vogons.

Quake, Duke3D likes the K6 if you're going for higher resolutions.

Nice build. How well does the Rage128 run OpenArena?

Make sure you get some photos of your K6 build 😀

Nice build. How well does the Rage128 run OpenArena?

I'd say that it runs the game pretty well. I get around 20-30 fps in 800x600 when there's a lot of action. 1024x768 is on the edge of playable. I'll see if I can post some benchmarks later, but for now I just have this Sandra benchmark:

I'm happy 🤣 (this is only when sandra is allowed to run on more than one core)

Ah you have an IBM Model M13-- the rarer version (most M13s came in black).

Indeed I do. I love that keyboard. I managed to get it locally for something like $10 😀

Some benchmarks:
OpenArena (Very subjective and inaccurate as there are no specific numbers):
All of these values are with affinity set to the faster CPU.
640x480: Anywhere from 40 to 9 fps. Less playable than 800x600...somehow.
800x600: Idles at around 30fps, drops to 20s and 10s based on the action. In general it is very playable at this resolution.
1024x768: Idles at around 20fps, drops to the low 10s based on the action. A bit sluggish but still playable.

Sandra '99:
Affinity set to CPU 0:
CPU Benchmark -
Dhrystone 1688
Whetstone 807
Multimedia Benchmark -
MMX 1432
FPU 869
RAM Benchmark (Will not be repeated for other two settings) -
Sandra crashed. 😵
Affinity set to CPU 1:
CPU Benchmark -
Dhrystone 1698
Whetstone 590
Multimedia Benchmark -
MMX 1438
FPU 872
Affinity set to both CPUs:
CPU Benchmark -
Dhrystone 1702
Whetstone 812
Multimedia Benchmark
MMX 1434
FPU 6 ( 😕 )
Second run of Multimedia Benchmark
MMX 10
FPU 874

Reference Pentium II@450 (Should match the slowest CPU at the CPU benchmark):
CPU Benchmark -
Dhrystone 1220
Whetstone 590
Multimedia Benchmark
MMX 1130
FPU 655.

These scores certainly are weird to say the least. From these results it appears that the slower PIII is matching the faster one in all benchmarks.

Does anyone have an explanation? This is both intriguing and confusing at the same time.

meisterister wrote:

These scores certainly are weird to say the least. From these results it appears that the slower PIII is matching the faster one in all benchmarks.

Does anyone have an explanation? This is both intriguing and confusing at the same time.

You're running mismatched processors, the slower 450 is holding you back. I'm surprised this runs at all. The 600 has been down-clocked to run at 450Mhz.

Slot-1 P3s are pretty cheap now, I suggest finding a matching processor for the 600Mhz or finding a pair of the fastest 100Mhz FSB P3s you can get for that board.

You're running mismatched processors, the slower 450 is holding you back. I'm surprised this runs at all.

That's the weird thing. According to Intel and backed up by HP http://h30499.www3.hp.com/t5/ProLiant-Servers … rs/td-p/2525877, the faster CPU is supposed to match the clock rate of the slower CPU. Based on these benchmarks, that clearly isn't happening. It appears that the slower CPU is actually matching the speed of the faster one. I decided to swap the processors and found that they properly defaulted to 450mhz. After switching them back, 600. The system has actually been a tad bit unstable but still usable, so the clock speeds incorrectly matching might be the cause. It looks like I just got a free PIII/600 out of this. 🤣

^Are those benchmarks even multi-threaded? (Or are you just getting scores for a single 600)
Is your 450 getting overclocked (if that's even possible, since the multiplier ought to be locked)?

(Well Katmai 600 was basically a factory overclocked chip but that's another matter)

The benchmark is single threaded but as I noted, I set the processor affinity so that I could benchmark each CPU individually. I also tested with no change to the affinity setting so that windows 2000 would bounce the benchmark between the CPUs. As you can see, each core performed on par for a 600 MHz Pentium III (http://www.pcstats.com/articleview.cfm?articleid=187&page=3). It does appear that in multiprocessor setups, the Pentium III has an unlocked multiplier. It would be cool to try to exploit this with an even faster CPU, but the 600 MHz Pentium is the fastest Katmai part.

Interestingly, CPUz correctly detects the two CPUs as 600 and 450mhz.

Welcome to the forum. I love slot-1 builds.
Those test results do look weird, but I'm still feeling skeptical about the chance of that multiplier being able to change under any circumstances. It just seems like that would be an exploit that the motherboard manufacturers would have found ways to take advantage of.

Your test results do show matching performance for each CPU, where the performance matches whatever is the primary.
Have you confirmed that the dual processors are definitely working? I know Windows is seeing them, but I'm wondering if at the hardware level everything is actually just going to the primary CPU. That might be how the board/chipset/whatever reacts to this situation.
I'm not sure what benchmark to suggest offhand, but it might be worth verifying that you actually have the simultaneous performance throughput of two processors. I have a feeling the secondary is just sitting there, despite what Windows thinks.

meisterister wrote:

That's the weird thing. According to Intel and backed up by HP http://h30499.www3.hp.com/t5/ProLiant-Servers … rs/td-p/2525877, the faster CPU is supposed to match the clock rate of the slower CPU.

That Compaq page says it only applies to Xeons. For P2/P3 CPUs, they still claim it can't change the speed of individual CPUs because of the multiplier lock.

The core frequency of each processor in the server is set to the frequency of the slowest processor. Switches on the system boar […]
Show full quote

The core frequency of each processor in the server is set to the
frequency of the slowest processor. Switches on the system board
(or the ROM) can be used to reset the frequency of Pentium Xeon
processors. However, Pentium III processors and Pentium II
processors are locked to the frequency ratio at which Intel intends
them to run. Since they are frequency locked, Compaq does not
support mixing of Pentium II or Pentium III processors with
different frequencies.

Your 450MHz could be unlocked, but I think that's really rare for Katmais. I think that would only happen with engineering samples.