Take a look here for some benches I've made for a specific motherboard: ASUS PCI/I-P54TP4 Socket 5 motherboard thread/review

Phil also have more videos on the topic, like this one: https://www.youtube.com/watch?v=fcAqRbFFQPU

I can give you a short version of what I experienced:

Socket 7:

Intel MMX: Very fast without cache: very fast 386DX and average 486DX2
Cyrix: Very fast without cache: very fast 386DX and average 486DX2
Intel Pentium: fast 386 DX and fast 486DX
AMD: Slow without cache: average 386 DX and average 486DX

I did do some testing with Slot 1, the faster the processor the closer you get to a 386 DX. Around 1 GHz or so Wing Commander is quite playable.

Here is my Pentium 133 compared to CPUs in the Norton Sysinfo database:

1CPU	                 NUSI
2Pentium 133	        421.3
3Pentium 66	         211.4
4486DX2-66	          141.7
5486DX-33	            71.2
6386DX-33	            35.9
7Pentium 133 L1D	     16.2
8286-12	               8.9
9286-8	                4.4
108088	                 1

This is something I'm going to work on for myself with my systems. 😀 I don't have numbers at the moment, but with L1 disabled my Celeron 333 and PII-333 score around 10 on this test. I plan on gathering your standard VGA results along with these.

I highly recommend not to use just Norton System Information.

Benchmarks that scale well are 3dbench 1.0c, Quake timedemo and Doom. I found these to be spot on when doing cache turning off experiments 😀

PhilsComputerLab wrote:

I highly recommend not to use just Norton System Information.

Benchmarks that scale well are 3dbench 1.0c, Quake timedemo and Doom. I found these to be spot on when doing cache turning off experiments 😀

Yep. I plan on using your benchmark suite along with this. I was just using Sysinfo as an illustration of what I was envisioning because it had a small database of other CPUs that I don't own which give you a ballpark of how the P133 performs with L1D. I also like Speedsys. It seems to be accurate with scaling down to 386 speeds.

PhilsComputerLab wrote:

Benchmarks that scale well are 3dbench 1.0c, Quake timedemo and Doom. I found these to be spot on when doing cache turning off experiments 😀

Quake wouldn't even load for me with L1 cache disabled...

zerker wrote:

PhilsComputerLab wrote:

Benchmarks that scale well are 3dbench 1.0c, Quake timedemo and Doom. I found these to be spot on when doing cache turning off experiments 😀

Quake wouldn't even load for me with L1 cache disabled...

Yeah, it's very slow to load on my P133 with L1D. It took many minutes just to load up, then many more minutes to complete the benchmark. I ended up with 1.7 FPS.

Yes it just takes a very long time 😀

clueless1 wrote:

zerker wrote:

PhilsComputerLab wrote:

Benchmarks that scale well are 3dbench 1.0c, Quake timedemo and Doom. I found these to be spot on when doing cache turning off experiments 😀

Quake wouldn't even load for me with L1 cache disabled...

Yeah, it's very slow to load on my P133 with L1D. It took many minutes just to load up, then many more minutes to complete the benchmark. I ended up with 1.7 FPS.

That's around 386DX-40 with Cyrix FPU.

I'm collecting benches with and without L1 on 3 different systems at every available clock speed: Pentium 1, Pentium II, and K6-2. The same benches as Phil's, but with Speedsys and Norton Sysinfo added, with and without L1. So far the results are very interesting. 😉 Might be another week before I'm done though. Some of those Doom and Quake runs without L1 can take quite awhile.

Nice, looking forward to your take on the whole thing 😀

clueless1 wrote:

I'm collecting benches with and without L1 on 3 different systems at every available clock speed

I'd create a special bench theme with a table of scores. P3 1GHz gives 15 XT in Speedtst wich has good correlation with Doom on CPU up to Pentium.

The same benches as Phil's

Quake and Doom would run too long.

Tertz wrote:

I'd create a special bench theme with a table of scores. P3 1GHz gives 15 XT in Speedtst wich has good correlation with Doom on […]
Show full quote

clueless1 wrote:

I'm collecting benches with and without L1 on 3 different systems at every available clock speed

I'd create a special bench theme with a table of scores. P3 1GHz gives 15 XT in Speedtst wich has good correlation with Doom on CPU up to Pentium.

The same benches as Phil's

Quake and Doom would run too long.

Can you put a link to Speedtst? I'm not sure I'm familar with it.

I have noticed how long Quake and Doom take. 😉 But I'm about halfway through already and usually just let the benches run while I'm at work, in the shower, etc.

I'll post a screenshot of my partially filled spreadsheet this evening. Not sure what you mean by "special bench theme", but after looking at my screenshot maybe you can expand on that. I'm just a newbie here and I'd like whatever I do to pass the muster of the veterans here.

I will say that just for completeness sake, I have been running every test with and without fastvid (on the cpus that it works with), even the ones with L1D. maybe that is unnecessary, but i'd rather have too much data than have to go back and redo anything. When I get to the K6-2, there will probably bee some utility for it that I should consider. Thanks.

Here's what I've got so far. Any suggestions for adding/subtracting/changing content much appreciated. I have a K6-2 550 that I will be adding to the mix (along with a bunch of underclocked results of it) down the road. Is there a FASTVID equivalent for the K6-2?

clueless1 wrote:

Can you put a link to Speedtst? I'm not sure I'm familar with it.

link
I used it in my bench theme in the signature.
May to be useful any benches which were made with 286 in mind, as they run not too long on 286-386 cpus level of wich is expected by switching off the cache1. If a bench runs >15 min - it's bad, and becomes worse if there are several such.

Not sure what you mean by "special bench theme"

A thread, created for gathering and showing results.

I will say that just for completeness sake, I have been running every test with and without fastvid (on the cpus that it works with), even the ones with L1D. maybe that is unnecessary

if this gives a significant difference (>25%) in the results - it's interesting

Thanks, Tertz. I will check out Speedtest this evening hopefully. My ultimate goal is to document what each CPU will perform as with L1 disabled. I imagine it would only make sense to do this with Pentium and higher, even though I've read that SETMUL will work on 486 or higher. It would still be interesting to know how a 486 is affected. And collecting results from members here would be ideal to fill in the blanks for platforms I don't have, like VIA C3, Cyrix and the K6-III.

clueless1:

Your results are interesting. Your values of the P133 with full caches (did you also had 512kb L2?) are 22% below my 3DB-Bench result and 36% below my PCP-Bench result. But i had also the real clock of 136.8MHz and another graphics card.
For my RetroPC's Projekt: Comparison of Pentium vs. K6-2 vs. PentiumIII @ lowest speed

And then with L1 disabled, the gap is really big for the P133. Your values: 3DB-Bench 9.2 and PCP-Bench 2.6, my values : 3DB-Bench 26.9 and PCP-Bench 8.0
What is with your L2 cache. Regarding these values it should be disabled.

melbar wrote:

clueless1: […]
Show full quote

clueless1:

Your results are interesting. Your values of the P133 with full caches (did you also had 512kb L2?) are 22% below my 3DB-Bench result and 36% below my PCP-Bench result. But i had also the real clock of 136.8MHz and another graphics card.
For my RetroPC's Projekt: Comparison of Pentium vs. K6-2 vs. PentiumIII @ lowest speed

And then with L1 disabled, the gap is really big for the P133. Your values: 3DB-Bench 9.2 and PCP-Bench 2.6, my values : 3DB-Bench 26.9 and PCP-Bench 8.0
What is with your L2 cache. Regarding these values it should be disabled.

There is no L2 cache on the Pentium. The Pentium II has 512KB L2.

Edit: If you notice, I do have a Pentium II 133 on the list. It's a 333Mhz Deschutes underclocked with a x2 multiplier. Even it is much faster than my "real" Pentium 133, but that is due to a better chipset (440LX vs 430FX) and faster RAM (SDRAM vs edo SIMMs) so it's hard to tell how much is the CPU vs those other variables.

Regarding the scores, the CL-GD5430 is a bit slow, so that might account for some of the difference. I plan on retesting the P133 with a TNT2 M64 PCI card I just got. That will get things more consistent with the Pentium II and Celeron results, which use the similar GF2 MX.

I can get confusing with L1, L2 and L3 cache...

In the BIOS, L1 is best to be looked at CPU cache. Whereas the L2 BIOS options is controlling motherboard cache and cannot be set via software, BIOS only.

This applied to Socket 7 / Super Socket 7 boards and gives you 2 options to control speed.

L1 and L2 off ~ 386 performance.

L1 off, but L2 on ~ 486 to 486 DX2 performance depending on the chip you're using.