Still no bar chart to represent the data?

I am also going to run this challenge, albiet on a much larger scale. The 686 benchmark comparison, which is again underway, will test 120 socket 7 CPUs (and a few slot 1's) using 97 different tests. To date, I've completed 42 CPUs.

For 133 Mhz CPUs, I'll be testing these (numbers refer to frequency, not PR):

Cyrix 5x86-133 (2x66)
Intel DX4-133 (2x66)
AMD X5-133 (4x33)
Cyrix 6x86-133
Cyrix 6x86MX-133
AMD K5-133
AMD K6-133
AMD K6-2-133
AMD K6-2+-133
AMD K6-3-133
Intel Pentium 133
Intel Pentium 133 MMX
IDT Winchip C6 - 133
IDT WInchip2 - 133
Rise mP6 - 133

No P-Pro or PII's at 133 MHz.

Note that the 6x86MX has the same performance as the MII, so there'd be no need to run both.

@swaaye
What is the name of the software utility you used to enable write allocate on the K5 and how did you confirm that it was not enabled by default by the BIOS?

feipoa wrote:

@swaaye
What is the name of the software utility you used to enable write allocate on the K5 and how did you confirm that it was not enabled by default by the BIOS?

I think it was:

Write Allocate Monitor II (works with K5 & K6)
http://k6plus.com/index.php?name=Downloads&re … wdownload&cid=1

How did you run your WinChip2 at 133 MHz (66x2)? The version of the chip I have, W2A, has non-conventional clock multipliers, in particular, the 2x multiplier is remapped to 3.33x. The other oddities are that 5x is remapped to 2.33x and 5.5x is remapped to 2.66x.

The only way I can run it at 133 Mhz is to do 50x2.66 = 133 MHz, but then that is a very unfair bus-wise.

This is the second CPU which has thrown me a curve ball for a comparison at 133 MHz. The other one was the AMD K6-2, but I have sourced a proper K6-2 chip now. I may need to source an older Winchip2. Coud you let me know what your chip looks like? If you don't have a photo, perhaps you could point out which one it is on chipdb,
http://www.chipdb.org/cat-200-276.htm

I don't have the 2A. The model I have is the original Winchip 2 240MHz (60x4) with only integer multipliers.

I believe this is it. I found it as part of a CPU upgrade kit (perhaps Evergreen).
http://www.chipdb.org/img-idt-winchip2-w2-3de … f-logo-3304.htm

That's what I figured. Do you know anyone in Canada who has an original W2? 133 MHz is kicking my butt!

I'm sorry to say that I have only one and I don't know anyone else with a Winchip.... You never know when one will appear on ebay though.

I'm asking around with CPU-World sources now. I only need it for for the Ultimate 686 Benchmark Comparison at 133 MHz. Usually, if a collector in Canada has it, I'd just borrow it if I cannot find it for sale. Otherwise, I'll synthesise the results, which takes 3x as long as actual CPU testing.

By exploring benchmark data from 66 MHz FSB CPUs, and in particular the Pentium MMX, I have found the results to extrapolate linearly to a precision that is indistinguishable from original CPU testing. The exception seems to be for the calculation of pi, which follows a power curve very nicely. The only issue with this method is that I'll need to test my W2A Winchip2 at more 66 MHz FSB frequencies than I was planning on. This is necessary so that I have sufficient data to extrapolate with.

At any rate, I'll get the 133 MHz data some how. I think 133 Mhz was really the sweet spot for benchmark interests. This frequency allows for 486 data capture as well as all socket 7 CPUs. There will be 97 benchmark tests per CPU. I will probably make bar charts for average ALU, FPU, and select tests, like Quake 1, Quake 2, Speedsys, Sandra99, etc.

I am quite excited to see the results of your testing. It will probably be the most thorough comparative testing ever performed on these CPUs outside of the original engineering teams.

Indeed. The Ultimate 486 Benchmark Comparison was insanely thorough, great job on that feipoa, can't wait to see the 686 Benchmark Comparison!

I view the 686 benchmark comparison somewhat as a test of mental endurance. I am constantly questioning my motives for this work; feeling like it is an absolute waste of my precious spare time. I don't think many people would have the patience, nor determination to gather nearly 15,000 data points by hand - and all for free! The retro virus has planted its roots deep and there's no stopping me now!

I have only a few more Rise mP6 speeds to tally before moving onto another motherboard. It will be time to retire this very faithful 430TX socket 7 board for a super 7 board. Unfortunately, I needed to do the comparison double-boarded because Super7 boards don't contain 50/55 MHz FSBs. 79 CPU's have been completed to date.

I'm lacking a socket 4 board, so unfortunately the 60/66 Pentium won't be included. I suppose I could synthesise that data, but it would be from a 430TX board, which is undoubtedly faster than s4 boards, so I probably won't include this data at all. I'm also lacking a NexGen board/CPU and a s8 board. If anyone has one of these 3 systems and would like to volunteer your time, please PM me for the Excel sheet.

I have already synthesised the data for the Winchip2 at 66x2. For CPU-only benchmarks, the results were identical to actual benchmarks at 50x2.66 (133 MHz). So I am confident the linearisation works well and will probably cease active interests in finding an original W2.

I'll probably be reaching the PII soon, and knowing my thirst for completeness, I'll want it run at 133 MHz as well. I only have a PII-400, which is likely multiplier locked. Does anyone have an old or extra PII-233/266 they don't want?

Swaaye, what is the s-spec on your Klamath?

feipoa wrote:

I'll probably be reaching the PII soon, and knowing my thirst for completeness, I'll want it run at 133 MHz as well. I only have a PII-400, which is likely multiplier locked. Does anyone have an old or extra PII-233/266 they don't want?
Swaaye, what is the s-spec on your Klamath?

You will also need a similar i440FX board to run the P-II at 133MHz with L2 cache. On a i440BX board the L2 cache gets disabled at that speed.
In case you also want to test Deschutes or Coppermine cores, see my test on page 2 of this thread. In between these two there is the Katmai core.

There are a few deschutes PII-400's which are not fully locked. I have a PII-333 like that, which actually contains L2 cache rated for 200MHz normally found on PII-400.

I needed to draw the line somewhere, so I'm stopping with Katmai 600 MHz. If anyone pressures me for a coppermine at 600 MHz, I'm a sucker for adding more CPUs, but I'm not going beyond 600 MHz.

I was actually planning on using a VIA slot 1 board for the Klamath 133 MHz test. Does the VIA 694/596 chipset disable the cache at 133 MHz? I'll be using an ASUS P3V4X for the slot 1 tests. Will this board work? It supports 66 MHz FSB and 2x multipliers.

I don't have a i440FX board. I have a spare Tyan dual 440BX board that I can put a terminator into the second slot, but you just mentioned i440BX misbehaves at 133 Mhz.

My PII-400's are from week 48 of 1998, so they are most likely locked.

@feipoa put your results in a nice PDF and I'll pay money for it!

feipoa wrote:

I needed to draw the line somewhere, so I'm stopping with Katmai 600 MHz. If anyone pressures me for a coppermine at 600 MHz, I'm a sucker for adding more CPUs, but I'm not going beyond 600 MHz.

Coppermine starts at 500MHz normally. But a 533EB has just a 4x multiplier. It is kinda interesting because the integrated L2 on the coppermine was told to be such an improvement, but my limited benchmarks do not show much improvement.

feipoa wrote:

I was actually planning on using a VIA slot 1 board for the Klamath 133 MHz test. Does the VIA 694/596 chipset disable the cache at 133 MHz? I'll be using an ASUS P3V4X for the slot 1 tests. Will this board work? It supports 66 MHz FSB and 2x multipliers.

Don't know about the L2 behaviour on that one. I have a VIA Socket 370 mainboard but I noticed benchmarks being underwhelming compared to intel mainboards. I learned to distrust anything VIA.

How would you account for the chips that can't be run at 133mhz like the 60-66mhz Pentiums or the Nexgen chips? It wouldn't be a fair comparison to include them at all. You can;t always make a straight line extrapolation as to how fast these chips would be if if you could bump up the frequency because there are other factors to performance besides CPU frequency alone.

i have coppermine and tualatin engineering samples with unlocked multiplier, but yes, the minimum multiplier you can set is 4x. i have no mendocino ES so i am not sure if its the same way, and if it has do with socket370 signal definitions.

noshutdown wrote:

i have coppermine and tualatin engineering samples with unlocked multiplier, but yes, the minimum multiplier you can set is 4x. i have no mendocino ES so i am not sure if its the same way, and if it has do with socket370 signal definitions.

So they shouldn't be used for purposes of this test because there is no way to clock them down to 133mhz. It's going to be very difficult to do this test at all because most of the fastest chips that could be included are going to be frequency locked.

noshutdown wrote:

i have coppermine and tualatin engineering samples with unlocked multiplier, but yes, the minimum multiplier you can set is 4x. i have no mendocino ES so i am not sure if its the same way, and if it has do with socket370 signal definitions.

It has to do with how the processor interprets the multiplier 4-bit signal.
My tualatin ES does 4x. My coppermine ES does 3x. I read about a certain mendocino celeron ES doing 2x.

sliderider wrote:

How would you account for the chips that can't be run at 133mhz like the 60-66mhz Pentiums or the Nexgen chips? It wouldn't be a fair comparison to include them at all. You can;t always make a straight line extrapolation as to how fast these chips would be if if you could bump up the frequency because there are other factors to performance besides CPU frequency alone.

I'm not! I'm only doing CPU's which will run at 133 MHz. For all other CPUs, refer to the Ultimate 686 Benchmark Comparison (in progress). I will not be doing the P60 at 133 MHz. The only CPU values I have synthesised so far is for the P300 MMX.

The reason why it might be of benefit to have results for the P60/66 and NexGen at 133 MHz would be to have a quick and dirty means of comparing CPU architecture vs. frequency. But for NexGen, you could always compare a 102 MHz NexGen to a 100 MHz Pentium or an AMD K5. Or even an 83 MHz NexGen to a POD83.

I was not aware that the coppermines went down to 533 MHz. Is that in the s370 form-factor? I was only going to use a slot 1. If my slot 1's disable the Klamath's L2 cache, then I won't include this CPU on the 133 MHz sub-section.

Yes, there will be neat printable sheets, just as with the Ultimate 486 Benchmark Comparison.

feipoa wrote:

I was not aware that the coppermines went down to 533 MHz. Is that in the s370 form-factor? I was only going to use a slot 1. If my slot 1's disable the Klamath's L2 cache, then I won't include this CPU on the 133 MHz sub-section.

Coppermine Pentiums went down to 500/100 and 533/133 MHz (the latter being the lowest multiplier at 4x) and Celerons down to 533 (8x), but those bottom end speed grades are rather uncommon iirc. Parts using 100 or 133 MHz bus could also be clocked down to 66 MHz, so you could get some speeds significantly lower still. (the slowest possibility being a 266 MHz PIII at 66x4)

Edit, looking on ebay, it seems there's actually a few 533 slot 1 PIIIs on there right now, albeit not super cheap.