Dan386DX wrote on 2026-05-28, 22:49:

In that case, have I misunderstood your ultimate comparison results where the POD achieved what looked like the best scores at 20.8 stock + 24.4 @ 100?

Or did you manage the 21.8 a bit later?

It looks like you are mixing up the Pentium Overdrive results with that of an Am5x86-180. This thread is asking about the best DOS Quake results when using an AMD 5x86, not a Pentium Overdrive in a socket 3. An overclocked POD will always beat an overclocked Am5x86 in DOS Quake. If you are interested in POD results, look for threads in the 110-125 MHz range.

feipoa wrote on 2026-05-28, 23:25:

Dan386DX wrote on 2026-05-28, 22:49:

In that case, have I misunderstood your ultimate comparison results where the POD achieved what looked like the best scores at 20.8 stock + 24.4 @ 100?

Or did you manage the 21.8 a bit later?

It looks like you are mixing up the Pentium Overdrive results with that of an Am5x86-180. This thread is asking about the best DOS Quake results when using an AMD 5x86, not a Pentium Overdrive in a socket 3. An overclocked POD will always beat an overclocked Am5x86 in DOS Quake. If you are interested in POD results, look for threads in the 110-125 MHz range.

Hold up, yeah I got that we're not talking about POD here.

What I'm saying is, in the ultimate 486 comparison, the best Quake result seems to be 20.8 at stock - and that is scored by a stock POD.

But it looks like the 180MHz 5x86 scores in this thread score higher than that?

An overclocked, and system optimised, Am5x86 at 180 MHz will outperform a non-overclocked Pentium Overdrive 83 MHz in DOS Quake. The 486 benchmarks do not include system optimised Am5x86 benchmarks at 180/200 MHz.

Also, most of those benchmarks were run on an M919 which implements an automatic 2/3 PCI frequency reduction when run with an FSB of 40/50/60 MHz. This was not known when running those benchmarks. To further the separation in results, the POD benchmarks were not on M919 (they were run on an MB-8433UUD) because the M919 doesn't work properly with the POD in L1:WB mode.

bertrammatrix wrote on 2026-05-28, 19:48:

21.2 on mine @ 180mhz. Score goes up a tad at 200 but since I normally use it at 180 that's the "official" score for me

M919, 1mb L2, matrox g200

Also a very nice result!

Dan386DX wrote on 2026-05-27, 21:35:

Correct, the Pentium's strong FPU gives it a huge advantage over most 486s

It's less the performance of the FPU itself (though that certainly helps), but more the superscalar design of the processor as a whole. This is what prevented other Pentium-class processors from AMD and Cyrix from performing as well in Quake.

Hi,

My highest score is 24.0 fps in Quake 320x200 with an AMD 5x86-133@200 4 volts (Peltier cooling) on ​​a Pcchips M919, Pancake Puppies 1MB cache stick with 10ns chips, 32MB EDO, TNT2 M64 pci@ 66mhz pci bus and the fastest timings in the bios possible.

I have tested 150+ memory modules to get the setup stable at the fastest timings.

Greets Sebastian

feipoa wrote on 2026-05-29, 04:26:

An overclocked, and system optimised, Am5x86 at 180 MHz will outperform a non-overclocked Pentium Overdrive 83 MHz in DOS Quake. The 486 benchmarks do not include system optimised Am5x86 benchmarks at 180/200 MHz.

Also, most of those benchmarks were run on an M919 which implements an automatic 2/3 PCI frequency reduction when run with an FSB of 40/50/60 MHz. This was not known when running those benchmarks. To further the separation in results, the POD benchmarks were not on M919 (they were run on an MB-8433UUD) because the M919 doesn't work properly with the POD in L1:WB mode.

Thank you for explaining. That makes good sense!

SquallStrife wrote on 2026-05-29, 15:20:

Dan386DX wrote on 2026-05-27, 21:35:

Correct, the Pentium's strong FPU gives it a huge advantage over most 486s

It's less the performance of the FPU itself (though that certainly helps), but more the superscalar design of the processor as a whole. This is what prevented other Pentium-class processors from AMD and Cyrix from performing as well in Quake.

Interesting. AFAIK, the 6x86 is superscalar on the integer side, but its FPU was essentially a 387-class design and obviously far behind Pentium. In Quake,which is heavily FPU bound, I'm sure that FPU was the 6x86 choke point?

Sebas1983 wrote on 2026-05-29, 23:25:

Hi, […]
Show full quote

Hi,

My highest score is 24.0 fps in Quake 320x200 with an AMD 5x86-133@200 4 volts (Peltier cooling) on ​​a Pcchips M919, Pancake Puppies 1MB cache stick with 10ns chips, 32MB EDO, TNT2 M64 pci@ 66mhz pci bus and the fastest timings in the bios possible.

I have tested 150+ memory modules to get the setup stable at the fastest timings.

Greets Sebastian

Wow!
That must be the new record !!!

Sebas1983 wrote on 2026-05-29, 23:25:

Hi, […]
Show full quote

Hi,

My highest score is 24.0 fps in Quake 320x200 with an AMD 5x86-133@200 4 volts (Peltier cooling) on ​​a Pcchips M919, Pancake Puppies 1MB cache stick with 10ns chips, 32MB EDO, TNT2 M64 pci@ 66mhz pci bus and the fastest timings in the bios possible.

I have tested 150+ memory modules to get the setup stable at the fastest timings.

Greets Sebastian

Oh wow, that's impressive!

Is that on a full system, or is it just barebones mobo + video?

Indeed, memory/cache combination plays a big role, definitely lots of alchemy involved. I could never get success at 0/0ws at 180/200 - my cache module only has 12ns chips and have only tried 30ish memories, so clearly there is still some room for improvement.

Is the ram something special (40-50ns) or just something regular?

Sebas1983 wrote on 2026-05-29, 23:25:

Hi, […]
Show full quote

Hi,

My highest score is 24.0 fps in Quake 320x200 with an AMD 5x86-133@200 4 volts (Peltier cooling) on ​​a Pcchips M919, Pancake Puppies 1MB cache stick with 10ns chips, 32MB EDO, TNT2 M64 pci@ 66mhz pci bus and the fastest timings in the bios possible.

I have tested 150+ memory modules to get the setup stable at the fastest timings.

Greets Sebastian

Is the system stable when mounted inside a case? Does it run Windows 9x games? I've seen many systems stable on the bench only to fail with a little increase in ambient temperature (inside the case).

The pentium is going off some abrash trick to do painting twice in one go. If DOS Quake were compiled without the d_ ASM, the results between the 486s and the Pentiums would be a bit more fair (though obviously slower for both)

(also i've grown tired of a lot of pentium snobbery having played Quake online a 1997 ago on an am5x86...)

bertrammatrix wrote on 2026-05-30, 16:33:

Oh wow, that's impressive! […]
Show full quote

Sebas1983 wrote on 2026-05-29, 23:25:

Hi, […]
Show full quote

Hi,

My highest score is 24.0 fps in Quake 320x200 with an AMD 5x86-133@200 4 volts (Peltier cooling) on ​​a Pcchips M919, Pancake Puppies 1MB cache stick with 10ns chips, 32MB EDO, TNT2 M64 pci@ 66mhz pci bus and the fastest timings in the bios possible.

I have tested 150+ memory modules to get the setup stable at the fastest timings.

Greets Sebastian

Oh wow, that's impressive!

Is that on a full system, or is it just barebones mobo + video?

Indeed, memory/cache combination plays a big role, definitely lots of alchemy involved. I could never get success at 0/0ws at 180/200 - my cache module only has 12ns chips and have only tried 30ish memories, so clearly there is still some room for improvement.

Is the ram something special (40-50ns) or just something regular?

It is just barebones mobo + video.

I tried various 45 and 50ns modules, but they were not stable. I achieved this score with a 60ns module with Hyundai chips.

feipoa wrote on 2026-06-01, 08:48:

Sebas1983 wrote on 2026-05-29, 23:25:

Hi, […]
Show full quote

Hi,

My highest score is 24.0 fps in Quake 320x200 with an AMD 5x86-133@200 4 volts (Peltier cooling) on ​​a Pcchips M919, Pancake Puppies 1MB cache stick with 10ns chips, 32MB EDO, TNT2 M64 pci@ 66mhz pci bus and the fastest timings in the bios possible.

I have tested 150+ memory modules to get the setup stable at the fastest timings.

Greets Sebastian

Is the system stable when mounted inside a case? Does it run Windows 9x games? I've seen many systems stable on the bench only to fail with a little increase in ambient temperature (inside the case).

I haven't tested that yet. I've only done some benchmarking in DosBench.

leileilol wrote on 2026-06-01, 17:04:

If DOS Quake were compiled without the d_ ASM, the results between the 486s and the Pentiums would be a bit more fair (though obviously slower for both)

shockingly no, 486Quake recently renamed to https://github.com/goshhhy/TurboQuake, did just that, replaced some assembler optimizations aimed at taking full advantage of pipelined FPU and results are quite surprising. It somehow runs faster on all CPUs equally 😮 that is Pentiums still somehow get faster despite replacing fxch tricks with more classical serial FPU code. K5/K6 get some gains, but those are miniscule and still get crushed by Intel.
results from the readme:

1cpu					| stock	| 486quake	| 586quake
2AMD 5x86/133		|	16.4|	18.4	|
3Intel 486DX4/100	|	9.6	|	10.9	|
4Cyrix 486DX2/66		|	6.3	|	7.1		|
5Pentium 133 (P54C)	|	35.6|	39.6	|	41.0
6AMD K6/166			|	32.5|	37.8	|	
7AMD K6/133			|	28.5|	33.0	|	32.2
8Cyrix 6x86MX PR166	|	26.3|	30.3	|	29.7
9AMD K5 PR133		|	25.8|	28.9	|	29.0
10Cyrix 6x86 PR166	|	23.9|	27.9	|	27.1
11Cyrix 6x86L PR166	|	23.5|	27.6	|	27.1