Mau1wurf, I really appreciate the time and effort you spend in benchmarking!

My attic is a mess right now (can barely walk in it!) but when I have time (and need to test 9 P1 mmx 233 cpu's soon anyway 😉 ) I'll do some benchmarking on a super 7 platform with Intel, K6, M2 and if I can get them to work, winchip and Rise. But it'll take a while, 🤣!

Sorry for the noob question but should't the ammount of cache memory present on the motherboard affect the final result? some motherboards like the "Soyo Sy-5ehm" have 1mb cache, most socket 7 boards have only 512kb.

Not a noob question, more cache is never a bad thing, so to speak.

My findings are more a guide to give you an idea of what kind of slow down you can achieve. The figures aren't meant to be an absolute reference, as there are too many other factors that can affect the result (BIOS settings, speed of the RAM, differences between boards).

So yes, cache can affect the results, but so can different types of RAM, type of board and BIOS settings.

W98 is likely the main environment where lots of cache would come in handy. But there is still the issue of the size of cachable area, e.g. many boards can only cache 128MB memory. If you are interested in running W98 on a socket 7 machine, it is highly recommended to get a K6-2+ or K6-3+ as they have on chip cache, just like current CPUs.

So it all depends on the context. For retro gaming, the size of the cache doesn't really matter. The biggest challenge is usually getting your machine slow enough for old games. If you need more grunt, just pick the next best thing (e.g. 486 > Pentium)

Mau1wurf1977 wrote:

Got around to testing the Pentium 200 chip (NON MMX) today and included it in my new "Timemachine - FSB Scaling" spreadsheet. […]
Show full quote

Got around to testing the Pentium 200 chip (NON MMX) today and included it in my new "Timemachine - FSB Scaling" spreadsheet.

We see some nice FSB scaling with L2 cache enabled. It also shows us how the CPU architectures differ.

Without Cache AMD K6-2+ is the slowest chip, Intel is doing better but Cyrix comes out on top 👍

Missing the 100 MHz FSB scores for the Cyrix. Will source a higher rates Cyrix chip soon to get that score. Also missing the 75MHz FSB for the K6-2+ but that's not a big deal...

So in a nutshell most Socket 7 cpus give you a 486DX-2 when you disable L1 cache and a 386DX when disabling L1 and L2.

In this case, though, isn't slower better? The Cyrix chip should lose this test since it runs faster than all the others. The AMD should be the winner since it clocks down to a 386-25 and the Cyrix only clocks down to a 386-40. That slight difference might be enough to run some older programs that crash at 40mhz.

Yes! AMD K6 FTW

Bump to this fine thread 😀

I can't think of any game / software that runs on 25MHz 386 that can not at 33MHz 386 , so the Pentium is as good as K6 for this purpose.

Also you mention the Pentium not being MMX, would this matter? Old software will not know the difference right?

Tetrium , were you able to benchmark those Winchip chips?

Filosofia wrote:

Bump to this fine thread :happy: […]
Show full quote

Bump to this fine thread 😀

I can't think of any game / software that runs on 25MHz 386 that can not at 33MHz 386 , so the Pentium is as good as K6 for this purpose.

Also you mention the Pentium not being MMX, would this matter? Old software will not know the difference right?

Tetrium , were you able to benchmark those Winchip chips?

In games written prior to the introduction of MMX it wouldn't matter, but are you planning on running your Socket 7 system at 386/486 speeds full time? If you ever want to play any games that are contemporary to the hardware you are using, then the lack of MMX might make a difference. I can't see building a Socket 7 system and using it exclusively to run 386/486 games when you can just buy an OEM 386/486 machine a lot cheaper than building one from parts.

sliderider wrote:

Filosofia wrote:

Bump to this fine thread :happy: […]
Show full quote

Bump to this fine thread 😀

I can't think of any game / software that runs on 25MHz 386 that can not at 33MHz 386 , so the Pentium is as good as K6 for this purpose.

Also you mention the Pentium not being MMX, would this matter? Old software will not know the difference right?

Tetrium , were you able to benchmark those Winchip chips?

In games written prior to the introduction of MMX it wouldn't matter(...)

Ok cool, this way I may change the Cyrix PR200 for the MMX Pentium 200MHz, because in addition to a "386 33MHz mode" I want the be able to play the early 3D games 😀

What I'd like to see is Doom benchmarks or some other relevant dos games, about 15.5-16 is about right for 3DBench for a 386DX-40 but a typical speedsys result would be around 9.2(without FPU), obviously its not a perfect science but it would be interesting and possibly pcpbench scores too, I still find it amazing that disabling the cache cripples the system so much.

dirkmirk wrote:

What I'd like to see is Doom benchmarks or some other relevant dos games, about 15.5-16 is about right for 3DBench for a 386DX-40 but a typical speedsys result would be around 9.2(without FPU), obviously its not a perfect science but it would be interesting and possibly pcpbench scores too, I still find it amazing that disabling the cache cripples the system so much.

Yeah, it makes me wonder what would happen if you were to mount a 286 chip on a small circuit board with some cache memory on the same board and then plug it into a 286 system how much faster it would run. I have an upgrade in one of My Atari Mega ST's that does just that with a 68000 chip and the speed difference is remarkable. Memory was expensive then so only a small amount of cache was included but at today's memory prices adding larger capacity chips would really boost the cache hit/miss ratio and speed it up even more.

sliderider wrote:

dirkmirk wrote:

What I'd like to see is Doom benchmarks or some other relevant dos games, about 15.5-16 is about right for 3DBench for a 386DX-40 but a typical speedsys result would be around 9.2(without FPU), obviously its not a perfect science but it would be interesting and possibly pcpbench scores too, I still find it amazing that disabling the cache cripples the system so much.

Yeah, it makes me wonder what would happen if you were to mount a 286 chip on a small circuit board with some cache memory on the same board and then plug it into a 286 system how much faster it would run. I have an upgrade in one of My Atari Mega ST's that does just that with a 68000 chip and the speed difference is remarkable. Memory was expensive then so only a small amount of cache was included but at today's memory prices adding larger capacity chips would really boost the cache hit/miss ratio and speed it up even more.

I think the speedup with such a 286 upgrade would be even more dramatic when you consider the speed and width of the interface between cpu and memory.... even a few kilobytes of cache in physical locality to the chip would see rather significant improvements imo.

I just keep bumping this thread 😉

Results are shown for the K6-2+, would K6-2 results be much different?

Filosofia wrote:

I just keep bumping this thread 😉

Results are shown for the K6-2+, would K6-2 results be much different?

I believe the K6-2+ is a K6-2 with on-die L2 cache. So it shouldn't be different in terms of slowing things down with L1 and L2 😀

EDIT: If you disable L1 cache in BIOS with a K6--2 or 3 + installed, it will disable L1 and on-die L2. The mainboard cache is now L3, but the BIOS option for L3 is still called L2.

this seems to be one of the techniques that moslo uses (only realised that today 😵 ).

Mau1wurf1977 wrote:

I believe the K6-2+ is a K6-2 with on-die L2 cache. So it shouldn't be different in terms of slowing things down with L1 and L2 😀

EDIT: If you disable L1 cache in BIOS with a K6--2 or 3 + installed, it will disable L1 and on-die L2. The mainboard cache is now L3, but the BIOS option for L3 is still called L2.

K6-2+ is a K6-III+, with half of the L2 cache disabled. I just don't know if all K6-2+'s are from chip harvesting, in other words, K6-III+'s that couldn't work properly with the full 256kb cache.