Um, kay?

I have one spare AT case, I can't decide whether to build a fast Tillamook MMX system, or something built around that M-II.

I didn't literally leave the CPUs on my keyboard, but short of putting them away into an anti-static bag, I'm not really sure what to do with em. They both work, I do know that! 😀

Leave them on top of a CRT, somewhere in the vicinity of the magnetic field.

what is the magnetic field going to do to a CPU? Maybe you're thinking of a floppy disk.

DonutKing wrote:

Leave them on top of a CRT, somewhere in the vicinity of the magnetic field.

what is the magnetic field going to do to a CPU? Maybe you're thinking of a floppy disk.

Iron Man was turned to steel in a great magnetic field, but of course that was when he traveled time for the future of mankind.

Can he see or is he blind?

SquallStrife wrote:

Um, kay?

I have one spare AT case, I can't decide whether to build a fast Tillamook MMX system, or something built around that M-II.

I didn't literally leave the CPUs on my keyboard, but short of putting them away into an anti-static bag, I'm not really sure what to do with em. They both work, I do know that! :)

I say go for the Cyrix MII for the obscurity factor. Best source a MII-400GP or better though. A good review on the MII-433GP can be found here,
http://www.pcstats.com/articleview.cfm?articleid=1187

It is only about as good as a Pentium Celeron 300 in ALU operations; so much for its 433 MHz Pentium rating. I'm not sure what the FPU equivalent of the 433GP is, perhaps a Pentium 233MMX. When I begin the ultimate socket 7 comparison, I'll be better able to answer this. I am still waiting on a few parts before casually starting this massive project.

DonutKing wrote:

Leave them on top of a CRT, somewhere in the vicinity of the magnetic field.

what is the magnetic field going to do to a CPU? Maybe you're thinking of a floppy disk.

If you don't think it will do anything, then go ahead and try it.

OK, I will. I have a CRT here and a shoebox full of common 486 CPUs. I'll leave the CPU there for a day or two and test it again. My money is on a perfectly functional CPU afterwards.

feipoa wrote:

I say go for the Cyrix MII for the obscurity factor. Best source a MII-400GP or better though. A good review on the MII-433GP […]
Show full quote

SquallStrife wrote:

Um, kay?

I have one spare AT case, I can't decide whether to build a fast Tillamook MMX system, or something built around that M-II.

I didn't literally leave the CPUs on my keyboard, but short of putting them away into an anti-static bag, I'm not really sure what to do with em. They both work, I do know that! 😀

I say go for the Cyrix MII for the obscurity factor. Best source a MII-400GP or better though. A good review on the MII-433GP can be found here,
http://www.pcstats.com/articleview.cfm?articleid=1187

It is only about as good as a Pentium Celeron 300 in ALU operations; so much for its 433 MHz Pentium rating. I'm not sure what the FPU equivalent of the 433GP is, perhaps a Pentium 233MMX. When I begin the ultimate socket 7 comparison, I'll be better able to answer this. I am still waiting on a few parts before casually starting this massive project.

I've seen that before too, and I wonder how Cyrix established that 433 rating. Different motherboard combinations and settings could play a significant role in performance (but probably not THAT much), but a bigger possibility would be that the rating was established by a processor other than a contemporary Celleron and/or a different benchmark in general was used. (in particular, the weighted performance could vary significantly for benchmarks with greater emphasis on 16-bit code -which the 6x86 architecture is especially fast at- rather than 32-bit -where the P6 pentiums were considerably more optimized, more so than the P5 and far more so than the 486 -which had much greater emphasis on 16-bit instructions, with 32-bit stuff being little to no faster than on a 386 clock for clock)

Having benchmark software that was specifically coded/compiled with the Pentium architecture in mind would also be problematic. (whereas less specifically optimized coding/compilers would be less selective, including 486-optimized compilers for that matter -since they'd tend to generate code catering to the 6x86's strengths more, and in the 6x86's heyday, the vast majority of software was optimized for 486 or lower)

I've also seen some explanations on the CPUworld forums that Cyrix never moved away from the P5 architecture for the PR benchmarks, so performance may have been compared to older pentiums on a per-clock basis. (though that wouldn't seem to make much sense for the late gen Cyrix chips)

As for FPU performance, yes, I guess a Pemtium MMX 233 would be about right compared to the Celeron 333 vs 300 MHz MII (433 at stock settings) benchmarks in that review and relative FPU benchmarks on Pemtium MMX and later P6 processors. (including the nice comparison thread here: 133 MHz Challenge - 5th/6th gen CPU per clock performance -too bad there's not a Cyrix CPU in there . . . )

DonutKing wrote:

OK, I will. I have a CRT here and a shoebox full of common 486 CPUs. I'll leave the CPU there for a day or two and test it again. My money is on a perfectly functional CPU afterwards.

Don't forget to turn the monitor on.

kool kitty89 wrote:

I've seen that before too, and I wonder how Cyrix established that 433 rating. Different motherboard combinations and settings c […]
Show full quote

I've seen that before too, and I wonder how Cyrix established that 433 rating. Different motherboard combinations and settings could play a significant role in performance (but probably not THAT much), but a bigger possibility would be that the rating was established by a processor other than a contemporary Celleron and/or a different benchmark in general was used. (in particular, the weighted performance could vary significantly for benchmarks with greater emphasis on 16-bit code -which the 6x86 architecture is especially fast at- rather than 32-bit -where the P6 pentiums were considerably more optimized, more so than the P5 and far more so than the 486 -which had much greater emphasis on 16-bit instructions, with 32-bit stuff being little to no faster than on a 386 clock for clock)

Having benchmark software that was specifically coded/compiled with the Pentium architecture in mind would also be problematic. (whereas less specifically optimized coding/compilers would be less selective, including 486-optimized compilers for that matter -since they'd tend to generate code catering to the 6x86's strengths more, and in the 6x86's heyday, the vast majority of software was optimized for 486 or lower)

I've also seen some explanations on the CPUworld forums that Cyrix never moved away from the P5 architecture for the PR benchmarks, so performance may have been compared to older pentiums on a per-clock basis. (though that wouldn't seem to make much sense for the late gen Cyrix chips)

As for FPU performance, yes, I guess a Pemtium MMX 233 would be about right compared to the Celeron 333 vs 300 MHz MII (433 at stock settings) benchmarks in that review and relative FPU benchmarks on Pemtium MMX and later P6 processors. (including the nice comparison thread here: 133 MHz Challenge - 5th/6th gen CPU per clock performance -too bad there's not a Cyrix CPU in there . . . )

Hello Kool kitty and welcome to the forum. Nice analysis on the possible cause of the over-rated MII-433PR. I suppose we would need to get some 486- and 686-optimised benchmarks to determine this for ourselves. If the Cyrix Pentium rating is really entirely based off the P54C-MMX processors, I suppose we could run both the Intel and MII at 233 MHz and see how the two compare. This future comparison of all socket 7 CPUs has been proposed on page 5 of this thread, The Ultimate 486 Benchmark Comparison

feipoa wrote:

kool kitty89 wrote:

I've seen that before too, and I wonder how Cyrix established that 433 rating. Different motherboard combinations and settings c […]
Show full quote

I've seen that before too, and I wonder how Cyrix established that 433 rating. Different motherboard combinations and settings could play a significant role in performance (but probably not THAT much), but a bigger possibility would be that the rating was established by a processor other than a contemporary Celleron and/or a different benchmark in general was used. (in particular, the weighted performance could vary significantly for benchmarks with greater emphasis on 16-bit code -which the 6x86 architecture is especially fast at- rather than 32-bit -where the P6 pentiums were considerably more optimized, more so than the P5 and far more so than the 486 -which had much greater emphasis on 16-bit instructions, with 32-bit stuff being little to no faster than on a 386 clock for clock)

Having benchmark software that was specifically coded/compiled with the Pentium architecture in mind would also be problematic. (whereas less specifically optimized coding/compilers would be less selective, including 486-optimized compilers for that matter -since they'd tend to generate code catering to the 6x86's strengths more, and in the 6x86's heyday, the vast majority of software was optimized for 486 or lower)

I've also seen some explanations on the CPUworld forums that Cyrix never moved away from the P5 architecture for the PR benchmarks, so performance may have been compared to older pentiums on a per-clock basis. (though that wouldn't seem to make much sense for the late gen Cyrix chips)

As for FPU performance, yes, I guess a Pemtium MMX 233 would be about right compared to the Celeron 333 vs 300 MHz MII (433 at stock settings) benchmarks in that review and relative FPU benchmarks on Pemtium MMX and later P6 processors. (including the nice comparison thread here: 133 MHz Challenge - 5th/6th gen CPU per clock performance -too bad there's not a Cyrix CPU in there . . . )

Hello Kool kitty and welcome to the forum. Nice analysis on the possible cause of the over-rated MII-433PR. I suppose we would need to get some 486- and 686-optimised benchmarks to determine this for ourselves. If the Cyrix Pentium rating is really entirely based off the P54C-MMX processors, I suppose we could run both the Intel and MII at 233 MHz and see how the two compare. This future comparison of all socket 7 CPUs has been proposed on page 5 of this thread, The Ultimate 486 Benchmark Comparison

You should include the Pentium II and Celeron, too, as those were the current CPU's at the high end of the speed scale of the MII. A good test would be to run a 433GP against the PII and Celeron at 300mhz since that is the actual clock speed of the 433GP. Against the 300mhz PII and Celeron, the performance difference is probably not as much as one would think but those weren't the chips the 433GP was compared to, they were usually compared to chips in the 400-450mhz range against which they were usually completely outclassed. I'd throw in a 300mhz K6 and K6-2, as well, to see how they stacked up against AMD at the same speed.

On the note of 6x86/MII chips in general, I found this interesting listing thread on CPUWorld:
http://www.cpu-world.com/forum/viewtopic.php?t=4104
It lists the 466 (366 MHz), 500 (400), and 533 (433/450) as actual commercially released products (and a 350 MHz version of the 433), but from all I can tell, the 433 was the highest rated MII to see volume production. (ie not just pre-production samples -which certainly exist for the unreleased Socket 370 Cayenne/Joshua core Cyrix III too)

Also, this site:
http://www.x86-guide.com/en/cpu/Cyrix-6x86.html
Has a lot of listings of die sizes and manufacturing processes used, but some of it seems a bit questionable . . . like rather massive die sizes relative to K6 and Pentiums of similar transistor counts, and a few cases of no change in die size between different processes. (this latter issue seems common among several other CPU types, so maybe just a common typo -mainly occurring among groups of CPUs mostly made on another process)

If that site is reasonably accurate though, it would mean the MII was being made on considerably larger/older process chips than contemporary PII/Celeron/K6 family chips. (and they were pushing pretty high clock rates for 350 nm parts, and would also explain the heat issues and difficulty in ramping up clock speeds -AMD and Intel had moved on to 250 nm much sooner -and also why voltages were so high on Cyrix chips for so much longer)
OTOH, Red Hill mentions a switch to 250 nm back with the MII 300 in April 1998.

feipoa wrote:

Hello Kool kitty and welcome to the forum. Nice analysis on the possible cause of the over-rated MII-433PR. I suppose we would need to get some 486- and 686-optimised benchmarks to determine this for ourselves. If the Cyrix Pentium rating is really entirely based off the P54C-MMX processors, I suppose we could run both the Intel and MII at 233 MHz and see how the two compare. This future comparison of all socket 7 CPUs has been proposed on page 5 of this thread, The Ultimate 486 Benchmark Comparison

If such (non-32-bit oriented) benchmarks were used, the P5 would also have appeared more favorable relative to the P6 pentiums (since 16-bit instructions are generally as fast or slightly faster on P5 chips), though there's also issues like I/O and cache bottlenecks to consider. (P6 chips with faster buses and/or larger/faster L2 caches)

So definitely a lot of variables there.

In that Cyrix 433 review, it's interesting to note that the overclocked performance increase is relatively modest in the Sandra benchmark with only around 16.5% improvement for FPU and Integer performance for a 25% clock speed boost vs the Celleron's performance increasing proportional to its overclock.
The Winbench 99 test shows a more proportional 20.6% improvement with the overclock, but still significantly less than the nominal 25% clock speed increase.

I'd guess it's an issue with the L2 Cache bottleneck of the socket 7 MII's 100 MHz board level cache, and they didn't overclock the bus, so the L2 cache stayed at 100 MHz. (if the Celeron's bus is being overclocked, that would also skew the comparison further, and I assume that's the case since Celerons 333s had the multiplier locked at 5x)

The specific Super 7 board used could also have been a factor, and the fact it had 512k cache rather than 1 MB. (or 2 MB, but that wasn't very common)
And. looking further, Red Hill's assessment of that Asus P5SB was quite poor for its use as a Super 7 board (OK as a vanilla socket 7 board, but not good for 75/83/100 MHz) So that very well could be the core problem right there.
A good VIA MVP3 based board would have been much better to benchmark on, or short of that, an ALI Aladdin 5 (like one of the good FIC or Gigabyte boards).

It would have been interesting to see the MII at 112x3.5 or 124x3, setting readily possible with an FIC VA-503 board. (let alone how the Cayenne/Joshua MIII core could have performed -or perhaps more so had it been a Super 7 chips with board level L3 cache like the K6III/K6+)

I've also seen claims that there were tweaks to later model MIIs that improved per-clock performance somewhat (aside from just the varying bus speeds).
This thread mentions that: 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.

sliderider wrote:

You should include the Pentium II and Celeron, too, as those were the current CPU's at the high end of the speed scale of the MII. A good test would be to run a 433GP against the PII and Celeron at 300mhz since that is the actual clock speed of the 433GP. Against the 300mhz PII and Celeron, the performance difference is probably not as much as one would think but those weren't the chips the 433GP was compared to, they were usually compared to chips in the 400-450mhz range against which they were usually completely outclassed. I'd throw in a 300mhz K6 and K6-2, as well, to see how they stacked up against AMD at the same speed.

Pentium II could be particularly significant to compare (relative to the Celeron 333 in that review) since the slower L2 cache could be a bottleneck for some things where the full speed 128k cache of the Celeron would be preferable (especially with an overclocked celeron's more competent FSB on top of that).
It would still be faster than the board level Socket 7 cache, but it'd be a fair bit closer.

The early Celeron 266/300 (without L2 Cache) would obviously tend to benchmark poorly for anything I/O intensive or not caching well within the limited L1 cache.

The use f the P5 as the basis of the benchmarks would seem rather odd indeed since they'd have to use hypothetical/fictional P5 counterparts for anything beyond PR-233 (or PR300 if you count the mobile P5 chips), so it would certainly be more logical to actual benchmarks against contemporary PII/III/Celeron parts. (or especially the PII and contemporary Celerons)

It's certainly possible that they DID do that and the P5 comparison thing is just a myth (or not even that since I haven't seen it as a very wide claim). But in that case, it would imply that Cyrix was using some sort of benchmarks that genuinely resulted in ALU performance in the range of PII/Celeron chips at the PR rated speeds.

And, again, there's the very serious issue of motherboard motherboard performance. (and it certainly looks like that review made a poor choice of Super 7 board)
For that reason, K6 family CPUs would be more foolproof to compare too, as the exact same boards could be used for both. (albeit sometimes certain boards are better matched to certain CPUs than others)

It probably didn't help the MII's reputation to often be used in low-end systems (including many emachines boxes) with poorer quality (or just slower) motherboards, hard drives, video, I/O cards, etc. (skewing the performance much further compared to higher end of mid-range systems -or more tactfully assembled lower end systems built by experienced dealers or DIY techies)

That said, even looking at the articles on the Red Hill site (who had a lot of experience building MII systems -and generally liked the parts), it's pretty clear that the later gen PR ratings were more generous than the early ones (where performance was generally better than the Intel counterpart -ie a 150 MHz PR-200+ was faster than a Pentium 200, in integer and I/O performance -though it should also be noted that later ratings omitted that "+" too).

Red Hill never managed to get any of the 100 MHz Cyrix parts (indeed it seems they never reached Australia Wholesalers at all), so the 250/83 and 263/75 MHz 333s were the last chip they actually worked with. (which, by their account, had similar business performance to a Pentium II 300 or K6-2 300 -which would imply a PR 433 would probably be closer to a P-II 400, Celeron 366, or K6-2 400)
-Actually, they don't seem to realize that the 400 and 433 were ever released in volume (though they do note the 366 being fairly common in the US).

sliderider wrote:

You should include the Pentium II and Celeron, too, as those were the current CPU's at the high end of the speed scale of the MII. A good test would be to run a 433GP against the PII and Celeron at 300mhz since that is the actual clock speed of the 433GP. Against the 300mhz PII and Celeron, the performance difference is probably not as much as one would think but those weren't the chips the 433GP was compared to, they were usually compared to chips in the 400-450mhz range against which they were usually completely outclassed. I'd throw in a 300mhz K6 and K6-2, as well, to see how they stacked up against AMD at the same speed.

I will consider adding the PII-300 and Celeron-300 to the list. The Celeron-300 came in 128 KB L2 cache and 0 KB L2 cache versions. I think the PII-300 was usually 512 KB, but there may be some mobile editions with only 256 KB. I would probably want a Celeron-300 w/out L2 cache to compare with the MII at 300 MHz.

Kool kitty, lots of useful information. Thank you for adding it here. I found this additional list in the Cyrix MII BIOS Writers Guide. It looks like Cyrix/National had grand plans for the SS7 form factor. I wonder if even a sample version of the MII-466 was ever produced?

Cyrix MII-400 95/285 3x
Cyrix MII-400 83/292 3.5x
Cyrix MII-400 75/300 4x
Cyrix MII-433 100/300 3x
Cyrix MII-433 90/315 3.5x
Cyrix MII-466 83/333 4x
Cyrix MII-466 95/333 3.5x
Cyrix MII-500 100/350 3.5x
Cyrix MII-500 90/360 4x
Cyrix MII-533 95/380 4x
Cyrix MII-550 100/400 4x

Cyrix had their own publication explaining the Pentium Rating of their MII processor. It can be found here,
http://datasheets.chipdb.org/Cyrix/M2/116ap.pdf

They used Winstone 98 for testing, which I do not have. I only have Winstone 96. If anyone has a link for it, please let me know.

I'm sure Cyrix picked a motherboard which gave them the best results. As I pointed out in another thread, the benchmarks results of the Cyrix MII-433 obtained in a Asus P5A-B motherboard with ALi chipset were about half that of the HOT-591P VIA chipset board. In the VIA motherboard, I determined that the MII-433 had the performance of an AMD K6-2-300. Cyrix's results seem to indicate that the MII-300GP (225 MHz) matches the performance of an AMD K6-300 and a Pentium II-300, and that the MII-300 clobbered a Celeron 300 (no L2 cache). Cyrix used an ASUS SP97-VX motherboard which contains the SiS 5582 chipset. I wonder if SiS boards work best with MII processors? Unfortunately, SiS socket7's didn't have an AGP slot (that I know of). My plan is to run the Socket 7 benchmark comparison on an Intel 430TX motherboard. The TX board's MII-433GP performance matched that of the VIA board's. My Intel 430TX board also supports AMD K6+ processors up to 500 MHz.

One of the main strengths of the MII is that it has 64 KB of L1 cache, as opposed to 32 KB in the Pentium MMX, PII, and Celeron. This probably played a big role in defeating the Celeron at 300 MHz.

kool kitty89 wrote:

...it would mean the MII was being made on considerably larger/older process chips than contemporary PII/Celeron/K6 family chips. (and they were pushing pretty high clock rates for 350 nm parts, and would also explain the heat issues and difficulty in ramping up clock speeds

The key is to play around with the 0.18 micron MII's.

kool kitty89 wrote:

...In that Cyrix 433 review, it's interesting to note that the overclocked performance increase is relatively modest in the Sandra benchmark with only around 16.5% improvement for FPU and Integer performance for a 25% clock speed boost vs the Celleron's performance increasing proportional to its overclock.

That review used an Asus P5SB, which is based on SiS (which has some kind of integrated AGP graphics port). As noted with the vastly different MII-motherboard speeds, I wonder if certain motherboard makers didn't really bother to test proper funcionality of the Cyrix MII?

kool kitty89 wrote:

...looking further, Red Hill's assessment of that Asus P5SB was quite poor for its use as a Super 7 board

We seem to have come to a similar conclusion.

kool kitty89 wrote:

...It would have been interesting to see the MII at 112x3.5 or 124x3

I am not aware of any Super7 board which will stably operate at 124 MHz. They all seem to have cache or TAG RAM which is too slow. Anyone find a SS7 with 6 ns TAG RAM?

I am curious why Cyrix created the MII-400GP for 285 MHz operation. Did these units really fail at 300 MHz and 2.2V? If an extra 0.1 V was all that it took to qualify at 300 MHz, why not just label the CPU as 2.3V as they did in the past? I tend to run my MII-400GP (300 MHz) at 2.3V just to be safe.

In reply to some of the comments others made in the above links, I have also had issues with the FIC VA503+. My guess is that they had some hardware issues post-production and didn't bother to recall the boards, instead servicing the ones customer's later complained about. I know FIC had quite a large stock of "refurbished" boards for sale, even as late as mid-2005). I've had new boards, but they always flake out after about a month. Some guy just paid a grip for these boards on eBay; he's in for a real surprise. http://www.ebay.ca/itm/310377156326

kool kitty89 wrote:

...it would imply that Cyrix was using some sort of benchmarks that genuinely resulted in ALU performance in the range of PII/Celeron chips at the PR rated speeds.

Perhaps Winstone 98 had some marketing deal with Cyrix? Unlikely. I think the main issue with the benchmarks is due to variations in hardware used. The socket 7 benchmark comparison will be done using identical hardware and treat ALU/FPU independently, as well as equally weighed in the overall score (as with the Ultimate 486 Benchmark Comparison).

feipoa wrote:

Kool kitty, lots of useful information. Thank you for adding it here. I found this additional list in the Cyrix MII BIOS Writers Guide. It looks like Cyrix/National had grand plans for the SS7 form factor. I wonder if even a sample version of the MII-466 was ever produced?

Given that the 433 seem to overclock well at 400 MHz and that Cyrix (at least traditionally) tended to push their chips pretty close to the edge of stability at the commercial level (hence many not overclocking well in general), it would seem odd not to at least offer MIIs up to 400 MHz at least in modest quantities. (ie both the 466 and 500 -the >400 MHz models, perhaps not)

But perhaps VIA's management changed some of those policies. (especially if faster MIIs were felt to contend with the VIA Cyrix III/CIII parts -which I assume were priced higher at the time -for the period where sales of both overlapped)

I'm sure Cyrix picked a motherboard which gave them the best results. As I pointed out in another thread, the benchmarks results of the Cyrix MII-433 obtained in a Asus P5A-B motherboard with ALi chipset were about half that of the HOT-591P VIA chipset board. In the VIA motherboard, I determined that the MII-433 had the performance of an AMD K6-2-300. Cyrix's results seem to indicate that the MII-300GP (225 MHz) matches the performance of an AMD K6-300 and a Pentium II-300, and that the MII-300 clobbered a Celeron 300 (no L2 cache). Cyrix used an ASUS SP97-VX motherboard which contains the SiS 5582 chipset. I wonder if SiS boards work best with MII processors? Unfortunately, SiS socket7's didn't have an AGP slot (that I know of). My plan is to run the Socket 7 benchmark comparison on an Intel 430TX motherboard. The TX board's MII-433GP performance matched that of the VIA board's. My Intel 430TX board also supports AMD K6+ processors up to 500 MHz.

On these motherboard issues, I can't speak from personal experience, but from what I've read:
Cyrix and IBM (at least sometimes) used different boards for benchmarks, and different boards for different grades of chips (not always just out of necessity either -ie higher bus speed or such required).

And from the CPU and motherboard guides on the Red Hill site:
-The FIC VA-502 was an outstanding match for the 6x86MX 200 (preferably the 66x2.5 version) with generally outstanding performance well beyond that chip in other boards (at least that they tested) or other chips in the same board. (within the speed ranges supported by that board)
-The IWill P55XPlus was the board used by IBM to benchmark the 83 MHz bus MII/6x86 MX 333 (3x83) and was the only board they found to work with solid reliability at 83 MHz for the 233 (2x83), 266 (2.5x83) and aforementioned 333. (which was the fastest MII they ended up selling -along with the 75x3.5 MHz 333, since they never managed to get any of the 366 models -let alone the .18 micron ones- apparently due to the lack of distribution support to Australia)
A shame they didn't get any of the faster chips since that could have given a good performance comparison. (for the 333 they pegged it comparable to the 100 MHz K6-2 300 and Pentium II 300 in general use -so not as good as Cyrix's benchmark's claims, but still competent for the time and at an incredible price)

And of course, since they never sold any of the 100 MHz bus MIIs, there aren't motherboard recommendations for those speed grades either. (the FIC VA-503+ would probably be a good bet though, at least given its overall flexibility, reliability and performance -though that's mainly based on use with K6 family chips -though the jumper set-up of that board is pretty nasty to deal with)

One of the main strengths of the MII is that it has 64 KB of L1 cache, as opposed to 32 KB in the Pentium MMX, PII, and Celeron. This probably played a big role in defeating the Celeron at 300 MHz.

Of course, the K6 (and Winchip for that matter) also had 64k L1 caches (the K-6 as dedicated 32k I/D caches iirc -avoiding flooding issues of unified caches, though also limiting flexibility for I/D ratios -albeit, the MII also had the dedicated 256 byte instruction cache/scratchpad)

The key is to play around with the 0.18 micron MII's.

Yes, though I was speaking more from a historical comparison standpoint than for modern tinkerers. 😉

And as a correction to my earlier statement, it appears that the MII 300 was indeed produced on 350 nm (as were several later chips), but they were also done in 250 nm, at least for the IBM chips (going by the x86-cpuguide lists). Again, the die size lists are a bit odd (many of the 250 nm ones are listed as 196 mm2 -same as 350 nm), but I assume the 119 mm2 size (listed for the 366) is correct for the 250 nm parts. (as is the 68 mm2 180 nm parts)
Though I'm not sure why the 250 nm parts stayed at 2.9 V. (maybe just for better yields, but it would at least make me wonder if some of those later parts would be more tolerant of undervoltage -more in the 2.2-2.5V range)

kool kitty89 wrote:

...It would have been interesting to see the MII at 112x3.5 or 124x3

I am not aware of any Super7 board which will stably operate at 124 MHz. They all seem to have cache or TAG RAM which is too slow. Anyone find a SS7 with 6 ns TAG RAM?

I was going by the listed specs on the FIC VA-503+ and 503A (at least on the Red Hill lists -I didn't bother to look up the manuals), both of which list 112 and 124 MHz speeds. And, aside from the Red Hill site, there seems to be lots of references online for 112 and 124 MHz on the 503+. (apparently the chipset and cache RAM are perfectly fine at those speeds, though DRAM and CPU are other matters)

Apart from the MVP3 based FIC boards, there's the ALI Aladdin 5 based Gigabyte GA-5AA with 105, 115, 110, and 120 MHz support.

I'm not sure how common those boards were worldwide, but they don't seem to be super hard to find. (my dad built machines with 503+ and 503A boards in the late 90s/early 2000s -still have the 503A working for that matter- and just last week I saw a 503+ at the Weirdstuff Warehouse in Sunnyvale)
If Red Hill's description is anything to go by, the 503+ was very high performance, very high quality, and lasted an extremely long time on the market (in terms of motherboard lifespans), going from a high-end/high-performance board down to the mid-range and finally entry in its twilight days. (along with the lower cost and slightly updated 503A -a 3rd SDRAM slot but no SIMM slots, a 4th PCI slot but only 1 ISA slot -rather than 3, embedded sound, and an AMR socket)

Though I've also seen mentions of early model (circa 1998) 503+ boards not supporting over 100 MHz. (I don't think the 503A has that issue)

I am curious why Cyrix created the MII-400GP for 285 MHz operation. Did these units really fail at 300 MHz and 2.2V? If an extra 0.1 V was all that it took to qualify at 300 MHz, why not just label the CPU as 2.3V as they did in the past? I tend to run my MII-400GP (300 MHz) at 2.3V just to be safe.

I wonder if those were originally created as 250/350 nm 2.9V(ish) parts (or intended to be), after all, the 3x83 MHz 366 went just over that clock speed too (291 MHz) on the old process, so maybe.

In reply to some of the comments others made in the above links, I have also had issues with the FIC VA503+. My guess is that they had some hardware issues post-production and didn't bother to recall the boards, instead servicing the ones customer's later complained about. I know FIC had quite a large stock of "refurbished" boards for sale, even as late as mid-2005). I've had new boards, but they always flake out after about a month. Some guy just paid a grip for these boards on eBay; he's in for a real surprise. http://www.ebay.ca/itm/310377156326

Huh, weird . . . I wonder how the reputation got to be so good in the case of the Red Hill dealership.

I think our 503+ may have died too (though after several years of use at least), and that 503A was a direct replacement. (which is still working after more than 10 years -albeit mostly sitting idle for the past 5+ years -I checked a few weeks ago, and it still boots up fine)

I wonder if the 503A is actually more reliable, or if it's more random than that. (or just a general tend with the later gen 503+ and A boards -which would also explain the impressions on Red Hill -since the vast majority would have likely been later models sold after pricing hit midrange levels -the foreign/international market perspective may also be a factor -perhaps FIC ended up distributing more of the flawed models in the US)

If the Gigabyte GA-5AA is easier to find reliable examples of, that should be a good choice too. (as well as, obviously, boards explicitly cited for Cyrix benchmarks)
The Pentium board used would also matter, of course, but I'd assume they'd have used pretty standard higher-end Intel-favored boards for that. (playing fair for chipsets/boards optimized for both systems -within limitations at the times the testing was done)

Oh, and some more general comments on the 6x86/M2:
Reading some of the stuff on this site:
http://www.realworldtech.com/altcpu/welcome.html

http://www.realworldtech.com/altcpu/subpages/ … faq/faq5of7.htm
You can see there's also a .44 micron die shrink listed (in-between .5 and .35 micron chip production), as well as an overview of some 6x86 chip revisions.

This article:
http://www.realworldtech.com/altcpu/subpages/ … q/mxfaq1of7.htm
among several other 6x86MD related pages on that site, mentions that the 6x86MX/MII design included significant improvements in 32-bit integer performance (unlike the heavier 16-bit performance emphasis of the original 6x86), so the bias on later benchmarks shouldn't be too bad. (though the overall 16 and 32-bit performance flexibility is also very good, so benchmarks making heavy use of mixed 32/16-bit code -as was fairly common in win9x stuff- would represent that rather well -and especially show advantages over the original 6x86 and P5 pentium -as well as possibly exposing the P6 pentium's weaker 16-bit performance)

Granted, aside from actual instruction/code distribution, there's the general issue of Pentium-optimized compilers being used (let alone assembly language optimizations), and that would certainly skew things against non Intel parts in general (and more so for designs further away from the pentium designs).
That and the issue of optimization on the motherboard/support hardware ends too. (but that's already been pretty well covered)

There's also an interesting section here:
http://www.realworldtech.com/altcpu/subpages/ … d/ibm6x86mx.htm
(under "The Name") that seems to imply that the 6x86 had improvements to its FPU relatively early on for the original 6x86, though perhaps there were issues on very early models with really poor FPU performance opposed to the decent (but more modest -by P5 standards) performance of later models. (the6x86MX faq page also mentions a 5% per-clock improvement in FPU performance over the older 6x86 on top of the integer performance boost -there may also have been small tweaks later on too)

http://www.realworldtech.com/altcpu/subpages/ … faq/faq3of7.htm
(under How poor is the Cyrix 6x86 FPU performance?)
Gives a good description of the trade-offs of the Cyrix FPU and overall performance.

It's probably good to note that on a clock for clock basis, the P5 FPU doesn't benchmark all that much higher than the 6x86s (let alone the MX) at around 90% the performance), but it obviously compared more dramatically by the PR numbers (since the clock speeds would be significantly lower).
And even so, applications with moderate use of FPU-intensive operations mixed with a majority of ALU operations could still favor the 6x86 (especially since the architecture allowed simultaneous integer and floating point execution -though not multiple simultaneous FLOPs as on the Pentium)

The P6 chips would obviously be more dramatically biased in FPU performance (K6 would be generally closer to P5), but in the case of Quake (and a few similarly designed games), there's much more to the issue than just floating point intensive operations but general heavy optimization for the Pentium architecture as a whole. (which is a big part of why Quake also runs slower on a K6 than a P classic/MMX of the same clock speed).

And for a while, Quake was the only game in existence with that sort of bias. (not all that common later on either -aside from more generalized FPU usage- and, of course, games relying on standard APIs -directX, openGL, etc- would be at the mercy of the driver used for those APIs -and you could potentially have versions or patches that catered to different CPUs of the time or totally substituted Floating point with fixed point operations -or heavily curtailed FPU usage- but I have no idea what sort of support there actually was for that -3DNow! and SSE obviously became very important too around '98/99)

kool kitty89 wrote:

-Actually, they don't seem to realize that the 400 and 433 were ever released in volume (though they do note the 366 being fairly common in the US).

The fastest I have ever seen in any quantity is the 366. 400 shows up for sale once in a rare while but the 433 seems to be practically non-existent. I've seen Engineering Samples of other chips with larger production runs than the MII 433. I have yet to see one being sold anywhere. I think your best odds for getting one of the faster ones is to find a Cyrix based system for sale where the seller knows the CPU is a Cyrix but is unaware of what model it actually is. I found a guy on another vintage computing forum a while back who acquired an AMD K5 system and it turned out to be a PR200, so there are probably still a few of those hidden gems out there.

Of the SS7 boards I own, the...

Asus P5A-B (ALi) has jumper settings for 60, 66, 75, 83, 95, 100, 105, 110, 115, 120 MHz operation.

HOT-591P (MVP3) has jumper settings for 60, 66, 75, 83, 90, 100 MHz operation, though they have left off one jumper configuration. I should measure it (it is probably 50 or 110 MHz)

FIC PA-2013 (MVP3) has jumper settings for 66, 68, 75, 83, 95, 100, 112, 124 MHz operation.

Tyan S1598C2 (MVP3) has jumper settings for 60, 66, 75, 83, 95, 100, though they have left off 2 possible jumper configurations. Again, I should measure these with the scope.

For more on the ALi vs. MVP3 topic, you can refer to this recent thread, Re: Super Socket 7: VIA MVP3 vs. ALi Aladdin V
After my experience with the ASUS ALi Aladdin V and Cyrix MII combination, I'm a bit soured on ALi boards and am wondering if any ALi board properly supports the MII (see above link). I'm also stearing clear of the VA-503+, but have high hopes for his ATX big brother, the PA-2013. I am not at all surprised that your VA-503+ died. I've not tested a GA-5AA or GA-5AX (ALi) and am a little reluctant to source one given my bad experience with ASUS/ALi. This is a decent website to begin hunting for a S7/SS7 board, http://web.inter.nl.net/hcc/J.Steunebrink/k6plus.htm

Does Weirdstuff Warehouse have any decent PCI socket 3 boards? That's where the money's at these days. SS7 needs another 5-10 years.

I wonder if those were originally created as 250/350 nm 2.9V(ish) parts (or intended to be), after all, the 3x83 MHz 366 went just over that clock speed too (291 MHz) on the old process, so maybe.

I sorta doubt it. That's a lot of voltage to drop for not changing the fab process. The fastest marked MII at 2.9V seems to be the engineering sample MII-350 at 90*3 (270 MHz). Do you have a photo of a 2.9V piece marked for 83*3.5 (292 MHz)? Since there were MII-366GP's marked as 2.2V, it would seem highly irregular to continue downmarking the voltages on even later chips (MII-400GP). I have an MII-366GP marked as 2.2V, but unfortunately it was DOA. I do have two working MII-400GP's though. The MII-366GP 2.9V chip I have does not overclock well at all; 300 MHz was definately out of the question when I tested it.

Back onto the question of the MII-400GP. I am still puzzled by its 95 MHz FSB. I wonder if the limiting factor was not in running the chip at 300 MHz, but rather the gain-bandwidth product of the onboard PLL frequency multiplier? From my tests with 486 boards, the input clock signal can sometimes have a rather poor swing, not even reaching TTL levels, so I figured that, in addtion to multiplying the frequency, the CPU also did some amplification and comparison to bring the CPU clock to digital TTL levels. It could be that the PLL/amplifier combination used on the MII-400GP wasn't up to stuff inasmuch as the gain-bandwidth product is concerned (among other things). An interesting test would be to try running the chip at 83x4 (333 MHz) vs. 112x3 (336 MHz). Provided that the motherboard is proven stable enough at 112 MHz (on say an AMD K6-2 CPU), if 112 fails on the MII, but passes on the 83, the issue is likely related to the CPU's ability to multiply/condition higher frequency buses.

Thank you for bringing Real World Tech to our attention. I've read their Cyrix 6x86MX links.
"the 6x86MX is not specifically optimized for Windows NT. It is optimized for both 16-bit and 32-bit code that is primarily found in Windows 95"
Strange, I thought Cyrix CPUs were targeting the business application world, which is where NT was used mostly. Maybe that targeting wasn't until the MX's. Real World Tech also seems to recommend SiS chipsetted boards for the Cyrix 6x86MX (perhaps because SS7's weren't out yet when the article was written?). It was interesting to read that Cyrix revision 2.6 and earlier CPUs had some bug issues with NT4.0 which Cyrix could not replicate in the lab. I wonder if the NT service packs ever fixed this?

This was interesting to read as well, "...their (SGS Thomson) manufacturing process was so poor that Cyrix themselves didn't even buy the 6x86s that came off their fab lines." So it seems like the IBM branded pieces were the best, followed by Cyrix, then SGS.

Another interesting bit, "The Cyrix supplied heatsink/fan rotates three times faster than most standard heatsink/fans to help reduce heat." Some said it sounds like a jet engine. I've never had an official Cyrix S7 heatsink/fan. Perhaps they were sold to the end-users?

My main interest with Cyrix started with the 5x86, because on the socket 3 platform, Cyrix was King if you could configure your system properly. There are a few IBM 5x86c CPUs out there which apparently work ok at 150 MHz. The few I have will only work at 133 MHz reliably, but I have a few coming in that I'm hopeful for. This post has a Speedsys Image of one such Cyrix 5x86-150, though the board used there didn't have its memory optimised.
Re: Hypothetical 5x86

Now that I'm pretty much OD'd on the S3 platform, I'm slowly moving towards the fastest Cyrix S7's - MII at 300 MHz.

There's the general issue of Pentium-optimized compilers being used

Quake seemed to be heavily optimised for the Pentiums, as you pointed out. I enjoyed the read on the FPU; if Cyrix was somewhat resource strapped, I think they did the right thing to focus on ALU applications given this statement, "One reason Cyrix did this is because the most commonly used games and applications in Windows don't use FPU. These applications depend on integer operations."

sliderider wrote:

I've seen Engineering Samples of other chips with larger production runs than the MII 433. I have yet to see one being sold anywhere.

Which particular MII's above 433 have you seen and under what conditions did you see them? Do you have photos?

sliderider wrote:

The fastest I have ever seen in any quantity is the 366. 400 shows up for sale once in a rare while but the 433 seems to be practically non-existent. I've seen Engineering Samples of other chips with larger production runs than the MII 433. I have yet to see one being sold anywhere. I think your best odds for getting one of the faster ones is to find a Cyrix based system for sale where the seller knows the CPU is a Cyrix but is unaware of what model it actually is. I found a guy on another vintage computing forum a while back who acquired an AMD K5 system and it turned out to be a PR200, so there are probably still a few of those hidden gems out there.

Yes,it seems VIA only briefly continued manufacturing of those faster grades. (though the 2.2V versions of the 366 seem a bit more common, but still rather rare -not to mention the mobile chips)

The 2.9V 366 (100x2.5) seems to be quite common though, at least given the number I've seen pass through ebay recently. (and the number of references I've seen online -it seems emachines used quite a lot of them)

DO you know if all the 2.2V parts are 180 nm, or are some 250 nm too? (I can't seem to find solid info on the 250 nm parts, but it would seem a bit odd to continue the 2.9V of the 350 nm chips on the newer process -AMD and Intel weren't doing that, and in the extreme cases they usually topped at 2.4V for some early release up-clocked versions)

feipoa wrote:

For more on the ALi vs. MVP3 topic, you can refer to this recent thread, Re: Super Socket 7: VIA MVP3 vs. ALi Aladdin V
After my experience with the ASUS ALi Aladdin V and Cyrix MII combination, I'm a bit soured on ALi boards and am wondering if any ALi board properly supports the MII (see above link). I'm also stearing clear of the VA-503+, but have high hopes for his ATX big brother, the PA-2013. I am not at all surprised that your VA-503+ died. I've not tested a GA-5AA or GA-5AX (ALi) and am a little reluctant to source one given my bad experience with ASUS/ALi. This is a decent website to begin hunting for a S7/SS7 board, http://web.inter.nl.net/hcc/J.Steunebrink/k6plus.htm

I'm honestly not really sure what happened with the 503+ we had, but by my dad's recollection (who built the system) it did fail and got replaced by the 503A. (not sure how long it lasted before that though, but it was running a K6-2 300 at the time)

If my old board is anything to go by, the 503A is pretty reliable, but a sampling of 1 doesn't mean much statistically. 😉 (and as you say, there are other good MVP3 options)

Does Weirdstuff Warehouse have any decent PCI socket 3 boards? That's where the money's at these days. SS7 needs another 5-10 years.

I didn't get a change to check the back warehouse section (with the less organized bulk storage), but the main area didn't have much in the way of 486 stuff (though it did have some odd single-board computer designs without normal expansion slots). Not much socket 7 stuff in general beyond that FIC-503 on that particular day at least (there was an odd single board computer card with a P200 classic though).

There was 1 socket 8 board and a lot of, Slot 1, 370, and Socket A stuff.

But yeah, nothing really interesting beyond that, at least at the time. (their website doesn't catalog any of the warehouse stuff either -just has listings for current online auctions and such, so that's not a great resource either)

It varies a lot though and a lot of stuff passes through there, so it's hard to say. (I have a friend who goes there more often, but that's the first time I'd gone there in a long time -just started getting into the retro computer stuff more seriously a year or so ago, and my dad kept a ton of our old hardware so there hasn't been that much need to look elsewhere -though it's certainly fun to just poke around at all the stuff they had there 😉)

I sorta doubt it. That's a lot of voltage to drop for not changing the fab process. The fastest marked MII at 2.9V seems to be the engineering sample MII-350 at 90*3 (270 MHz). Do you have a photo of a 2.9V piece marked for 83*3.5 (292 MHz)? Since there were MII-366GP's marked as 2.2V, it would seem highly irregular to continue downmarking the voltages on even later chips (MII-400GP). I have an MII-366GP marked as 2.2V, but unfortunately it was DOA. I do have two working MII-400GP's though. The MII-366GP 2.9V chip I have does not overclock well at all; 300 MHz was definately out of the question when I tested it.

OK, I was probably mistaken on that assumtion of the 2.9V 292 MHz model. (I can't seem to find any references to that -and rather little even on the 2.2V version)

It looks like the fastest 2.9 MII in volume production was the 75x3.5 333 at 262.5 MHz.

Back onto the question of the MII-400GP. I am still puzzled by its 95 MHz FSB. I wonder if the limiting factor was not in running the chip at 300 MHz, but rather the gain-bandwidth product of the onboard PLL frequency multiplier? From my tests with 486 boards, the input clock signal can sometimes have a rather poor swing, not even reaching TTL levels, so I figured that, in addtion to multiplying the frequency, the CPU also did some amplification and comparison to bring the CPU clock to digital TTL levels. It could be that the PLL/amplifier combination used on the MII-400GP wasn't up to stuff inasmuch as the gain-bandwidth product is concerned (among other things). An interesting test would be to try running the chip at 83x4 (333 MHz) vs. 112x3 (336 MHz). Provided that the motherboard is proven stable enough at 112 MHz (on say an AMD K6-2 CPU), if 112 fails on the MII, but passes on the 83, the issue is likely related to the CPU's ability to multiply/condition higher frequency buses.

AMD had similar problems with the early model K6-2 300, right? (having to down-bin some of those to 66x4.5 rather than 100x3)
Though the 95 MHz bus is a rare example of that. (95 MHz bus chips -in as far as the K6-2- were clocked as such due to core speed/yield limits rather than multiplier issues)

And in such a case (limited bus freq tolerance, but adequate core speed tolerance), you'd also think Cyrix would have commonly offered 66x4.5, 75x4, and 83x3.5 MHz parts at the same time. (which doesn't seem to be the case at all)

On a similar note, I'd assume those potential bus speed/multiplier related issues would also have been a limiting factor on Cyrix not getting a 100 MHz bus part out sooner than they did. (ie no 100x2 MHz part -which could have been very significant in early/mid 1998)

I wonder if any of the older (non 100 MHz) 6x86s/MIIs will actually well at 100 MHz bus. (and multipliers to match similar or lower clock rates to their rated speed -so avoiding heat/core stability problems)

Thank you for bringing Real World Tech to our attention. I've read their Cyrix 6x86MX links.
"the 6x86MX is not specifically optimized for Windows NT. It is optimized for both 16-bit and 32-bit code that is primarily found in Windows 95"
Strange, I thought Cyrix CPUs were targeting the business application world, which is where NT was used mostly. Maybe that targeting wasn't until the MX's. Real World Tech also seems to recommend SiS chipsetted boards for the Cyrix 6x86MX (perhaps because SS7's weren't out yet when the article was written?). It was interesting to read that Cyrix revision 2.6 and earlier CPUs had some bug issues with NT4.0 which Cyrix could not replicate in the lab. I wonder if the NT service packs ever fixed this?

I'd gotten more of an impression that Cyrix generally targeted mainstream integer performance for general desktop home/business computing, and the "for business" connotation was applied more after the fact due to games/multimedia stuff shifting to a heavy pentium/FPU bias toward the end of the 90s (the K6's stronger FPU and later 3DNow! solidified that further).
Plus, even if they HAD optimized the MX for WinNT/OS/2 style software, they'd still be building onto the excellent 16-bit integer performance of the M1 core. (so have inherently better balanced 16/32-bit performance than the P6 designs)

And on the games/multimedia issue, prior to Quake, the 6x86 should have been among the very top performing CPUs for games/multimedia in general (due to the greater emphasis on 486 coding -or even 386/486SX compatibility- ) and when Quake did arrive it was still one of only a very few pentium-specific mainstream applications available at the time. (though that pentium and/or floating-point bias grew and AMD's newer chips catered much better to that -and added 3DNow!- while Cyrix started to lag in general around that time -apparently in large part due to National Semiconductor's management decisions to de-emphasize development of high performance dektop CPUs in favor of focusing on embedded SoC type designs -stemming from the Media GX)

This was interesting to read as well, "...their (SGS Thomson) manufacturing process was so poor that Cyrix themselves didn't even buy the 6x86s that came off their fab lines." So it seems like the IBM branded pieces were the best, followed by Cyrix, then SGS.

Cyrix and IBM chips should have generally come from similar IBM facilities, with the main difference being packaging and IBM's more stringent quality control for speed rating. (and sometimes slightly lower retail pricing than Cyrix, oddly enough -at least according to that faq)

Another interesting bit, "The Cyrix supplied heatsink/fan rotates three times faster than most standard heatsink/fans to help reduce heat." Some said it sounds like a jet engine. I've never had an official Cyrix S7 heatsink/fan. Perhaps they were sold to the end-users?

According to that site, Cyrix later stopped packaging CPUs with fans, but instead included documentation of the requirements/specifications of the fans to be used.

My main interest with Cyrix started with the 5x86, because on the socket 3 platform, Cyrix was King if you could configure your system properly. There are a few IBM 5x86c CPUs out there which apparently work ok at 150 MHz. The few I have will only work at 133 MHz reliably, but I have a few coming in that I'm hopeful for. This post has a Speedsys Image of one such Cyrix 5x86-150, though the board used there didn't have its memory optimised.

At 133 MHz, a 5x86 would probably even be significantly faster than a Pentium overdrive of AM5x86 at 160 MHz (barring the unreliable 4x50 MHz overclock).
Hell, even for Quake it might have been better than an 83 MHz overdrive. (the Cyrix FPU would be running at 133 MHz after all -was the 5x86 FPU similar performing to the 6x86?)
Let alone a 150 MHz (50 MHz bus) 5x86. (though that would probably be about as hard to achieve as on an AM5x86 -or tougher given the touchier nature of the 5x86 -as you say, limited to very specific revisions)

I kind of wonder how well the 5x86 core would have done as a socket 5/7 part. The per-clock ALU performance was obviously worse, but the yields seemed to be considerably higher than 6x86 parts of the same time (given the high clock speeds) and die size also was much smaller. (so cheaper too -and cooler running, at least at similar clock rates)

If nothing else, it probably would have been a great entry level part to complement the higher end 6x86/MII (more like the Winchip -except considerably better performing, especially assuming all the bugs were fixed and those added features were enabled).
In hindsight at least, that probably would have been a much better investment than moving the 5x86 onto the Media GX project. (not to mention the fateful impact of National Semiconductor taking interest in the company because of that design)

Quake seemed to be heavily optimised for the Pentiums, as you pointed out. I enjoyed the read on the FPU; if Cyrix was somewhat resource strapped, I think they did the right thing to focus on ALU applications given this statement, "One reason Cyrix did this is because the most commonly used games and applications in Windows don't use FPU. These applications depend on integer operations."

Yes, prior to Quake, even all 3D games catered heavily to pure ALU performance (a few had optional pentium-specific settings -iirc Descent allowed that).
Another issue with quake is that you had very limited control over detail/quality settings (just resolution), so you couldn't reduce or disable perspective correction (one of the most FPU intensive tasks in the game). For a Pentium that made perfect sense (as disabling it would had little to no performance advantage due to the way its programmed) but for other chips it was a major disadvantage. (if not for that, Quake probably would have run decently well on fast 486 systems and perhaps marginally playable even on slower systems)

Tomb Raider is a good contemporary example for quake, actually using a more complex/advanced game engine (generally more complex levels/terrain/animation/lighting/etc), but still catered to non pentium systems and allowed perspective correction to be disabled entirely. (as well as variable resolutions and screen sizes)
The original Tomb Raider is probably one of the most intensive games that can actually run on a stock 386 (in terms of actually booting and working) . . . and actually might be semi-playable on a good 386DX40 system. (with cache, fast RAM, fast VESA VGA card, etc)

It would be interesting to get more details on Direct3D/Glide/OpenGL/etc driver support for fixed-point libraries, but as it is I can't really do much more than speculate. (but it definitely should have been possible to offer drivers catered to the K5 or 6x86's integer performance -let alone for 486 based systems for that matter -or the Winchip; so not 6x86 specific optimizations per se -to the same extent as the Pentium started getting- but more generalized support to cater to a broad spectrum inclusive of the 6x86's strengths -as well as pretty much every other non-Pentium chip on the market)

And back on the topic of Cyrix's CPU design emphasis: at the time of the MII/MX being designed, floating point performance was still relatively niche too, so that really wouldn't have been a major consideration either.

It appears the big problem came with Cyrix's developments after the National Semiconductor merger and the resulting shift in management.
No only did the shift in priorities limit things like updating the MII's FPU or adding 3DNow! (or improving MMX performance for that matter, and addressing the few missing pentium-specific instructions), but general development of faster conventional (ALU-oriented) MII models as well.

As it was, under NS, Cyrix engineers did still end up developing the next generation Cayenne core (pipelined FPU, 256k L2 cache, 3DNow!), but the development pace was much slower than it could have been and there was a general lack of interim architecture updates on top of that.
Aside from problems with ramping up clock speeds (and related late switch to denser/lower power smaller process tech), there was the late entry of a 100 MHz bus part, the FPU/multimedia instruction updates (namely 3DNow!), and the less significant problem of lacking complete P5 ISA compatibility. (it would have been rather like AMD leaving the K6 on older/high-voltage tech -actually worse since the K6 was on 300 rather than 350 nm, lacking full P5 compatibility, introducing the 100 MHz bus versions a year late and without 3DNow!, and lagging a year late with K6-III development while not designing the K7 at all)

Losing the IBM manufacturing connection probably didn't help either. (given IBM's excellent manufacturing facilities)

Just a comment about FPU and Cyrix / AMD / Intel and Winstone 98: I made the mistake of relying on that benchmark to make my purchasing decision when it came time to replace my P166. I could have gotten a pentium 233 MMX, but instead I opted for the K6-233, based on reviews that used Winstone/Winbench 98, Tom's Hardware being prominent among them. We ran that demo in the store I worked at all the time, as it had a loop function. The problem was, and I didn't really understand this until later, that AMD's floating point performance suuuuuuuucked compared to the intel chip. So why did the Ziff Davis benchmarks show the AMD chip as faster than the Intel chip? Because none of the programs used in the Winstone 98 benchmarks used floating point arithmetic. They are all integer-based programs!

So needless to say, when I got home and wanted to see how awesome Quake 2 would be, instead it sucked about the same or perhaps worse than on the 166. I never for a moment ever considered owning a Cyrix, knowing full well from being forced to build systems around them and sell them to people how badly they sucked. We often paired them with "Opti Viper" motherboards which made things even worse. They were cheap, and you got what you paid for.

And in such a case (limited bus freq tolerance, but adequate core speed tolerance), you'd also think Cyrix would have commonly offered 66x4.5, 75x4, and 83x3.5 MHz parts at the same time. (which doesn't seem to be the case at all)... On a similar note, I'd assume those potential bus speed/multiplier related issues would also have been a limiting factor on Cyrix not getting a 100 MHz bus part out sooner than they did. (ie no 100x2 MHz part -which could have been very significant in early/mid 1998)

As noted above, Cyrix did at least list out,
Cyrix MII-400 95/285 3x,
Cyrix MII-400 83/292 3.5x, and
Cyrix MII-400 75/300 4x
in the BIOS Writer's Guide, but I have never seen a MII-400GP with anything less than 95 MHz stamped on it. This, however, may have just been a marketing ploy; a CPU running with a 75 MHz bus is less attractive than one running at nearly 100 MHz (95 MHz, in this case). It is curious how the MII-333 came in 66, 75, and 83 MHz variants though. Either they were targeting the upgrade market whereby the motherboard only went up to 66/75 MHz, or they had some PLL/amplifier frequency issues at higher frequencies. It would have been nice if Cyrix stamped all possible bus combinations on the CPU. Considering that all the MII-366 pieces have a 100 MHz bus stamped on them leads to marketing strategies. I still don't have a clear answer on the MII-400GP's bus weirdness. This will take some extensive testing to determine. If I can find a repeatable means to crash the MII-400 at 3x100 and not at 4x75, the issue may be bus PLL multiplication/amplification related with a dependance on internal frequency and/or core voltage. I know I've said this before, but it would be awefully nice to track down a former Cyrix engineer to get some straight answers.

I wonder if any of the older (non 100 MHz) 6x86s/MIIs will actually well at 100 MHz bus. (and multipliers to match similar or lower clock rates to their rated speed -so avoiding heat/core stability problems)

A great question. I should test for this. Considering that the Cyrix 5x86 seems to run OK at 2x66 MHz implies there is some wiggle room with the buses. The MII-400 might be wiggled out though. It is clear to me that they were gunning for a 100 MHz bus. Perhaps the MII-400's are just nice performing MII-366 2.2V pieces and some new growth/yield strategies were implemented for the MII-433? Sometimes little things like humidity and temperature combinations during wafer growth can make a break your yields.

At 133 MHz, a 5x86 would probably even be significantly faster than a Pentium overdrive of AM5x86 at 160 MHz (barring the unreliable 4x50 MHz overclock). Hell, even for Quake it might have been better than an 83 MHz overdrive.

I think you would enjoy reading the Ultimate 486 Benchmark Comparison and Cyrix 5x86 Register Enhancements Revealed (links are on my signature); it will answer some of your questions. There are easier to read PDF files in those links which make for great bedtime stories. Enabling the register feature FP_FAST on the Cyrix 5x86 boosted FPU performance by an average of 18%, and 39% in some tests. The average FPU performance of a 5x86-133 bests the POD83 by about 10 Pentium ratings (so the differance between a P90 and a P100). The exception seems to be with Quake, whereby a properly configured Cyrix 5x86-133 scored 18.4 FPS and the POD83 scored 20.8 FPS. If the Cyrix 100/120/133 data is linearly extrapolated to 150 MHz, a Cyrix 5x86-150 would marginally beat the POD83 in Quake. When overclocking a Cyrix 5x86, it is important to keep in mind that your mileage will vary; not all Cyrix 5x86 next generation features overclock well, however my tests have shown that the important ones seem to overclock.

The Cyrix FPU would be running at 133 MHz after all -was the 5x86 FPU similar performing to the 6x86?

This is what the Ultimate 586/686 Benchmark Comparison will tell us. I plan on including the Cyrix 5x86's in this comparison since they are a sorta 486/686 hybrid. I'm pretty sure the 6x86's will win clock-for-clock for ALU performance since they included two, as opposed to one, integer unit. I also plan on running the 5x86-133 at 2x66 MHz to be a fairer comparison with the 6x86-133 MHz. I'll use the same sticks of RAM, graphics card, etc to be as consistant as possible.

...kind of wonder how well the 5x86 core would have done as a socket 5/7 part. The per-clock ALU performance was obviously worse, but the yields seemed to be considerably higher than 6x86 parts of the same time (given the high clock speeds) and die size also was much smaller. (so cheaper too -and cooler running, at least at similar clock rates)

Cyrix was targeting super low power consumption and reduced transistor count with their 5x86 series, which is probably why the die size is so small and the yields so high. I wonder if Cyrix did IR inspection of the wafer's before cutting them? Or did they just cut and ignore any possible wafer impurities? At one of the semiconductor growth facilities I used to work at, we would always mark the good areas on the wafers so that the impurities were not included in the final cuts. These were thick wafers though, and we couldn't really afford to cut, lap, and polish a wafer which wasn't previously IR inspected.

If nothing else, it probably would have been a great entry level part to complement the higher end 6x86/MII (more like the Winchip -except considerably better performing, especially assuming all the bugs were fixed and those added features were enabled.

The only bug I see with the 5x86 was the high frequency dependance of specific register features. Branch prediction seems to run better at 100 MHz than at 133 MHz, although I have a 2x66 5x86 Cyrix running with all features enabled (including branch prediction) in Windows 98SE without issue, even on step0rev5 pieces. The trick is that other not-so-important features must be disabled, particularly LOOP and RSTK, and to a lesser extent, DET_E and BWRT. The ones you want enabled (BTB, LSSER, FP_FAST, LINBRST, and MEM_BYP) can still be enabled.

I'd have liked to see Cyrix continue with the 5x86 chips like AMD did. The AMD X5-133 is what kept AMD profitable while developing the K5/6. If Cyrix had the resources to simultaneously develop the 5x86 into 150, 160, and 200 MHz pieces, perhaps their fate would have been different. As the literature mentioned, Cyrix had yield issues with their 6x86's and couldn't keep up the pace. Had there been 160-200 MHz Cyrix 5x86's in Q1/2 of 1996, I don't think many people would have bothered buying Pentiums until the PII's came out.

Indeed, it would have been more ideal for Cyrix if game makers targeted the MII's strengths, but if I was a game maker, I'd still probably want to appease the big buy, Intel.

Losing the IBM manufacturing connection probably didn't help either. (given IBM's excellent manufacturing facilities)

That begs a question in my mind - Does IBM still do much production fabbing these days? It seems like they took a turn into project-specific consulting for big businesses.

McMick wrote:

So why did the Ziff Davis benchmarks show the AMD chip as faster than the Intel chip? Because none of the programs used in the Winstone 98 benchmarks used floating point arithmetic. They are all integer-based programs!

Well that certainly explains Cyrix's decision to bench their CPU with Winstone 98. The glory of now is that you can get a Cyrix MII-400 for next to nothing, along with a suitable motherboard. For the obscurity factor of building a retro box, I'd be content if the Cyrix MII @ 300 MHz had the FPU of a P233MMX.

sliderider wrote:

DonutKing wrote:

OK, I will. I have a CRT here and a shoebox full of common 486 CPUs. I'll leave the CPU there for a day or two and test it again. My money is on a perfectly functional CPU afterwards.

Don't forget to turn the monitor on.

There wouldn't be much point to the exercise without turning it on would there? 😉

Anyway, I used 2 CPUs, just old 486DX-33s. One directly in front of the monitor at its base and one at the rear on top of the casing.
I left them there all weekend and left the monitor running - I played a few games on it and when I went out I left it on a screensaver. So that's over 40 hours exposure to the radiation/magnetic field.

As expected both CPU's are perfectly functional. They both boot and I can run 3dbench and speedsys and doom.

I know there is no fancy shielding or anything in this old CRT since I've had it open before to replace the VGA cable. I even pulled out a compass and it goes haywire in these two spots.

So I guess that's myth busted then 😀

Sorry for the derail 😀