feipoa wrote:

kool kitty89 wrote:

couldn't you at least do some quick preliminary tests for basic stability (ie will with system at least boot and maybe run some basic tests/benchmarks -like speedsys, etc)

You could, but it wouldn't prove long-term stability. I have found that CPUs can pass all the benchmarks, pass Windows98 installation, and still fail in the long-run. Have you been looking for a SS7 and a Cyrix MII? Perhaps you could run this test. I'm busy with the 686 benchmark comparison and don't want to go too far astray.

I currently only have a MII 366 (2.5x100) 2.9V I got on ebay about a month ago. That should certainly pass 2x100 MHz fine (and give realistic benchmarks for a hypothetical 2x100 MII), but it wouldn't address the overclockability of 66/75/83 MHz bus parts. (though I could test to see if its stable at 3.5x75 or 4x66 -I doubt it will do much more than that given the general limits of the 2.9V parts)

The only socket 7 board I have now is a FIC503A, though (assuming the performance is similar to the 503+) that should be a good one to benchmark with.

I've been checking weirdstuff for more boards and CPUs, but I haven't seen any socket 7 boards since that 503+ a couple months back, and they don't have any socket 7 CPUs stocked/sorted in the front part of the warehouse either. (they probably have a lot more old boards/systems/CPUs in the back warehouse section, but that's employees only and they've been too busy to go back and look, at least the days I've tried) -I'm still keeping an eye out for an FIC 502 though. (on that note, it might be worth contacting the Red Hill to see if they have records of their FIC 502 with 166 MHz 6x86MX-200 tests mentioned on their museum page -they mentioned integer performance on par with a Pentium MMX 233)

They up until a little over a month ago (early february), the wierdstuff website had listings of CPUs from 808x up through modern Athlons/Xeons/Pentiums (as well as some 68k models and a few RISC parts), but they seem to have removed everything prior to socket A (I don't even see 370 on there). So either they sold all their current stock of older parts, or they stopped selling the older parts online at all. (though it would seem pretty odd to abruptly go from offering almost every category down to early 80s and then cutting off everything pre 2001-ish)
The last time the site had socket 5/7 parts, it was still limited to K6-II/+ and Pentiums (a couple dozen K6 parts iirc and a handful of Pentium/PMMX parts).

But, again, it's still quite possible that there's a lot more unlisted stuff in the back warehouse (and most of the customer-access section has stuff that's not on the site as well -especially a huge amount of "as-is" parts). I saw a guy donate/sell an old PC-AT class system (huge IBM-style desktop box) with a K6 sticker on it several weeks ago, and I haven't seen it in the front section either. (probably back in the warehouse)

My MII-366GP 2.9V unit was only able to handle up to 233 MHz (66x3.5). I tried it at its rated 250 MHz, but the system would hang on the benchmarks (even the DOS ones). I had to replace it with a MII-400GP to run the 250 MHz benchmarks. I ran my MII-366GP 2.9V at 83x3.0, not the marked 100x2.5, which makes me wonder if some oddball Cyrix chips have a minimum FSB requirement as well.

Unfortunately, I do not currently have any MII's slower than a MII-366GP, so I am unable to test 83 MHz pieces for an upper limit on the FSB. I do have a Cyrix MII-333GP on the way, which is rated at 83x3.0 (250 MHz), so I'll test it at 2.5x100 when it comes in and when I am at that stage in the benchmark comparison. I still need to complete the Socket 7 chips before moving onto the Super7's.

The next closest chip I have currently is a Cyrix and IBM 6x86MX-233PR, one marked as a Cyrix 75*2.5 (188 MHz) and one marked as an IBM 66x3 (200 MHz). The IBM piece handled 200 MHz without issue and the Cyrix piece couldn't even finish the 166 Mhz tests. All the 2.9 V Cyrix chips I've tested get awefully warm to the touch, unlike the 2.2 V pieces. I wonder if there is something wrong with the 2.9 V setting on the motherboard. I suppose I should remeasure that. Two Cyrix chips not passing benchmarks at their rated speeds seems curious, however I should point out that my IBM 120 MHz chip was able to run all the benchmarks overclocked to 133 Mhz without issue. If the Cyrix-marked pieces are this bad, I'd hate to try an SGS Thomson variety.

feipoa wrote:

My MII-366GP 2.9V unit was only able to handle up to 233 MHz (66x3.5). I tried it at its rated 250 MHz, but the system would hang on the benchmarks (even the DOS ones). I had to replace it with a MII-400GP to run the 250 MHz benchmarks. I ran my MII-366GP 2.9V at 83x3.0, not the marked 100x2.5, which makes me wonder if some oddball Cyrix chips have a minimum FSB requirement as well.

Did you try 75x3.5 on your 366?

No. I'm not sure what information could be concluded from that test. 262 MHz (75x3.5) exceedes the rated speed of 250 MHz (100x2.5). I should also point out that one benchmark test failed on the MII-366GP even at 66x3.5, 233 MHz (it was the Final Reality - Software test, which had to be run on the MII-400GP). All other 95 benchmark tests completed successfull on the MII-366GP at 233 MHz. I would like to see how the MII-366GP does at 250 MHz on a different motherboard though.

Cyrix has certainly lived up to its reputation of being a poor overclocker. I have 3 Cyrix MII-400GP processors, so hopefully one will be a stable overclocker. I'm hoping for a 350 MHz stable unit, but will settle for 333 or 300 MHz if that is all it can handle. From my experience with overclocking 5x86's, some features overclock well, and others not well. Luckily Cyrix features can be enabled/disabled in software.

feipoa wrote:

No. I'm not sure what information could be concluded from that test. 262 MHz (75x3.5) exceedes the rated speed of 250 MHz (100x2.5). I should also point out that one benchmark test failed on the MII-366GP even at 66x3.5, 233 MHz (it was the Final Reality - Software test, which had to be run on the MII-400GP). All other 95 benchmark tests completed successfull on the MII-366GP at 233 MHz. I would like to see how the MII-366GP does at 250 MHz on a different motherboard though.

If it's stable at 75x3.5, it might show that the bus speed is the main stability limit (and Cyrix did introduce the 75x3.5 333 significantly earlier than the 366 -implying yields were higher for that speed range). The same would be true for testing 66x4, though it would be strange if it was actually more stable at 4x66 than 3.5x66 (as you mentioned, there being some minor problems at 233 MHz).

Which board did you use for those tests?

I'll definitely let you know my results with my 366 when I get around to that. (perhaps the FIC board will fare better)

There's also some interesting benchmark figures on cpu-world, though some odd cases like the 83 MHz PR333/350 and PR266 actually performing worse (integer) than the 66 MHz PR233 (3x66). This wasn't the case for the 75x3.5 333 or 2.5x100 366, or any other >200 MHz core clocked parts. (ie, all other parts ranked fairly evenly by clock speed)
It also shows a significant per-clock advantage over the K6 for integer performance, but a per-clock disadvantage for PII/PIII integer (pacing the K6 even further behind), which seems strange. (maybe they used very P6-biased tests)

http://www.cpu-world.com/CPUs/MII/Cyrix-MII-233GP.html

www.cpu-world.com/CPUs/MII/Cyrix-MII-40 ... 2.2V).html

www.cpu-world.com/CPUs/MII/Cyrix-MII-33 ... CPGA).html

I am using an i430TX board for these tests, a Biostar MB-8500TTD. This is noted on the bottom of the test matrix. Certainly not an expensive big name board, but it is branded and the company is still around. The board seems well made for a Taiwanese-class motherboard.

I am not at a stage where I want to delve into FSB tolerances on the Cyrix chips; this will have to come later, however I encourage you to begin your own investigation. Cyrix 6x86MX and MII chips are relatively cheap still, but this won't last forever. You can usually get them for $5/6 from guys on CPU-world, or for $6-10 from eBay scrappers. eBay scrappers will normally part out specific chips from a mass gold recovery auction if you are willing to pay more than a gold buyer for a particular chip. I have quoted the low-ends of what I've been able to get, for stubborn sellers and chips you are really insistant on getting, you may need to double those ranges.

I have never actually clicked the benchmark tab on cpu-world; I just sorta figured it was empty - but it is not! The only cases I've seen of ever so slightly worse integer performance for a higher frequency Cyrix 6x86 is when select tests make eavier use of the FSB, particularly for comparisons between 75 and 83 MHz FSBs. These are usually with heavily graphic tests and implies to me that it is the overclocked PCI bus which is leading to the higher 83 MHz score.

I tend not to give too much weight to benchmarks picked out from the internet because the range of sample benchmark programs is not very large. For the Ultimate 686 Benchmark comparison, I am using 23 different benchmark programs and 97 tests. The programs are a mix of synthetic and real programs and will average ALU, FPU, Graphics-based, and overall system performance using the same system with identical hardware. Every possible 686-class CPU will be tested on the same system, so the results are directly comparable. I may even take select benchmark programs, like Quake 1/2, pcpbench, etc, and make specific comparison charts for them as well.

I aim to finish up the Cyrix MII 262 (75x3.5) and MII 292 (83x3.5) sometime today or tomorrow. This will finish off the Cyrix chips for the Socket 7 platform, though not yet the Super7 platform.

Are any of your chips the black-top variety? Those appear to be flip-chips with the die on top (like K6 and most/all newer CPUs) and may have better cooling properties than the traditional under-mount PGAs.

I have a black-top IBM 6x86-233MX.

luckybob wrote:

I HATE THAT CASE. I hate them with the burning passion of a 1000 suns. My high school used them in all teh computer labs and I cant count how many times I cut myself on them. Last time I got one used I took it to the mountains and shot it for target practice. I think I saved some pictures somewhere...

My high school had two rooms of those but desktops, one set with Intel 815 boards and the other set with later Intel 845 boards.

My high school had a room full of Apple IIe computers, and perhaps one or two 286's.

Could you possibly add a Celeron300A with it's legendary Overclock to 450 MHz on a 100 MHz FSB? I really have fond memories of this particular CPU back in 1999.

It would be nice to see where a Rise mp6 fits in that comparison. There's someone on ebay selling some now, but they're a bit pricey. Does anybody have any they can test?

I have a mp6 266 2x100, well i have two but only one of em works.

That score seems pretty respectable for 200 MHz. Can you run speedsys at 66 x 3 as well? This will make it more directly comparable with an Intel Pentium 200.

[GPUT]Carsten wrote:

Could you possibly add a Celeron300A with it's legendary Overclock to 450 MHz on a 100 MHz FSB? I really have fond memories of this particular CPU back in 1999.

I am no longer actively sourcing CPUs for this benchmark comparison, however I am accepting donations. The only Celeron I have is a 400/66 (SL39Z). Will this be sufficient? Anything special about this Slot1 Celeron?

I just noticed something odd regarding the difference between the 2.9 V MII and the die shrunk 2.2 V MII; per-clock, the 2.9 V MII seems to perform 2 - 6% better in benchmarks compared to the 2.2 V version. These results come from Winbench99 CPUmark, Sandra99, and MDK Performance (Software). I didn't bother checking more.

Thinking that I must have changed hardware or some other settings, I re-ran the 2.2 V benchmarks, but they were still 2 - 6% lower than the 2.9 V Cyrix MII. The only MII 2.9 V CPU I have that would run reliably at 3.5 x 75 (262 MHz) was an IBM 6x86MX-333PR, which was the last IBM MII. I am not sure if IBM made any secret mods to their last MII or if Cyrix's die shrunk 2.2 V version had any features removed for the purpose of increasing yields at higher frequencies.

Another idea is that IBM simply changed some of their default register settings for slightly improved performance. I probably should have checked for that before putting the machine away...

The performance also increased in Speedsys, 1% for L1, 13% for L2, and 5% for the RAM. Memory bandwidth is listed as 225 MB/s for both CPUs. I used 3.5 x 75 = 262 MHz for these tests.

Speedsys
Cyrix MII 2.2v ---> IBM MII 2.9v

Score: 187.13 ---> 188.24 (This is probably where the 1% for L1 comes in)
L1: 877.76 ---> 887.48
L2: 196.35 ---> 222.29
RAM: 98.45 ---> 103.11

Puzzling. Does anyone know with any certainty if Cyrix made any technological changes to the 2.2 V die shrunk MII?

I'm not sure about 2.9 vs 2.2v cores in general, but I Cyrix did make modest modifications/tweaks that slightly improved performance for later revision parts: there's a bit on that mentioned here:
http://www.sysopt.com/forum/showthread.php?t=44340

Not all MII are created equal - rev. 08h is faster than older ones at the same clock speed, rev. 14h is faster again, and even newer ones might be further optimized. The ratings table got changed several times to take this into account.

Also note that the MII was made in both 250 nm and 180 nm as 2.2V parts (and fewer rated at 2.0v -possibly 2.1 as well), though apparently most/all of the PR400 (and 433) chips are 180 nm.

Also, addressing some recent comments on the 6x86 I made here:
Re: VIA C3/C7 Discussion on the 16-bit code performance.

It seems I was both right and wrong with my perception of the bias for 16-bit code on Cyrix chips. The 6x86 and MII are quite good at 16-bit code, and seem to be more competitive there than the Pentium II (much more so compared to the PPro), and the MII did improve 32-bit code performance over the M1 design.
And it's specifically mentioned in this article that MII winstone scores are significantly better in Win95 than NT: http://www.azillionmonkeys.com/qed/cpuwar.html#6x86MX

However, it seems the 6x86 classic is actually significantly faster at running 32-bit applications than the P5 Pentium and has a significantly bigger advantage for Winstone 32 than Winstone 96:
http://www.realworldtech.com/altcpu/subpages/ … d/myupgrade.htm
Showing the PR166+ (133/66) 6x86 as consistently faster than the P5 166, but somewhat slower than the Pentium 166 in Winstone 96.

That's not to say the 16-bit application performance wasn't actually better than 32-bit on the 6x86 (since actual speeds aren't given), but at least that 32-bit 6x86 performance was better than Pentium, implying that it has more balanced all-around performance than the Pentium classic. (it also implies that Cyrix didn't design the 6x86 expressly to cater to a narrow range of mainstream applications at the time of its inception, but to be fast per-clock in most/all integer operations -and at least much better than the 486 for floating point)

feipoa wrote:

That score seems pretty respectable for 200 MHz. Can you run speedsys at 66 x 3 as well? This will make it more directly comparable with an Intel Pentium 200.

Here ya go. My MSI board dosent like the mp6 so i switched to my epox one, it has 1mb lvl2 cache so it may have a inpact on the score.

Does it run Win98SE and Win2000?

I've been lurking around for the past few weeks but anyway does anyone know what the fsb wall is for most Cyrix M2? I've been tinkering around with a few samples that I got left with a p5a rev 1.03 and gotten some running at 2x120 on stock volts. However I am not good at overclocking these 😒 As for getting to 300mhz I can't get even remotely close most of the time. Still need to find a home for my Cx486DRx²66mhz. Still got the old badge and cooler to match.

nforce4max wrote:

I've been lurking around for the past few weeks but anyway does anyone know what the fsb wall is for most Cyrix M2? I've been tinkering around with a few samples that I got left with a p5a rev 1.03 and gotten some running at 2x120 on stock volts. However I am not good at overclocking these 😒 As for getting to 300mhz I can't get even remotely close most of the time. Still need to find a home for my Cx486DRx²66mhz. Still got the old badge and cooler to match.

Only the 2.2V models are good for clocking that high. The older 2.9V (350 nm) parts don't tend to go much beyond the rated 250~262 MHz ratings on the fastest commercially released parts, even with boosted voltage. (the fastest preproduction part on that process was the 3x90=270 MHz PR350)

Still, in a best case, you might get at least a bit closer to 300 MHz with a relatively good CPU example along with boosted voltage (and adequate cooling to match).
The multipliers aren't locked either, of course, so you've got a good deal of flexibility there too, especially coupled with a board with numerous FSB speed options (like the Asus P5A with 105/110/115/120 -or less so for the MVP based FIC boards with 112/124 -not sure which offered 90, but that would be interesting to test as well)
112x2.5 (280 MHz) might be doable on some chips. (I'm going to try that with my MII 366 and FIC 503A . . . eventually -I'm also planning on playing around with lower multipliers and higher bus speeds too: 124x2=248 MHz would mean a slightly lower core clock than the rated 250, but a considerably faster bus and L2 cache -if it worked stable)