Cyrix 5x86-133 eBay Extravaganza

[feipoa]

There was recently an eBay auction that ended 13 Jan 2011 for 24 Cyrix 5x86-133 cpus for $300 total. It is thought that the buyer of this auction is from this forum and that many of the chips were distributed to forum members. I have a few questions, namely,

1) Have any of you who received the chip tested it to see if it is a working Cyrix 5x86-133 with 4x multiplier? Does it cause any problems with continual operation? Or is the chip a fake? When I say fake, it could be one of those oddball Cyrix 5x86 100 or 120 with a 4x multiplier or a Cyrix 5x86-100 3x, which has been renamed as a 5x86-133, or even a non-working, or unknown chip with the Cyrix information printed on the ceramic surface.

2) If it is a working unit at 133 Mhz, are any of you willing to sell it? If so, at what cost? I have been a long-time user of the 120 Mhz version and would like to put the 133 Mhz chip into regular use.

[Neon_WA]

[CPUShack]

These chips all look legitimate, their production date fits in the timeline.

They were originally in an interposer type socket, as if they were for some sort of upgrade.

[feipoa]

Would you mind posting your cachechk and/or speedsys results when you have a chance (preferably with a cache timing of 2-1-2 or 2-1-1-1 and memory wait state of 0 -- nextgen enhancements on, if possible).

With a Biostar MB-8433UUD v3.0, Cyrix 5x86-120, 512 KB 15 ns L2 cache, and 128 MB 60 ns FPM RAM, I get,

Speedsys:
Score: 65.49
L1: 173 MB/s
L2: 68 MB/s
RAM: 48 MB/s

Cachechk 4&7:
L1: 246 MB/s
L2: 96 MB/s
RAM: 55 MB/s
RAM access time (read): 76 ns
RAM access time (write): 51 ns

I suspect your numbers would be better for a working Cyrix 5x86-133, along with long-term Windows stability under load.

[CPUShack]

I dont have a test system for these, but perhaps I could send one for testing if ya ship it back shortly

What are the markings (top and bottom) of your 120?

the 133s came in 2 different voltages

[feipoa]

The two 120 Mhz units I have are both 3.6 volts. From what I could tell visually, the eBay auction 133 Mhz units had 3.6 and 3.7 volts. This must be due to some threshold tolerances and variances in semiconductor doping, which I think should be fairly well controlled -- the deviance is probably related to significant digits.

My markings are:
5x86-120GP
026-3.6V
G5F9550C
Currently running in a Biostar MB-8433UUD v3.0 (daily use, longest uptime 5 mo. w/NT 4.0)

5x86-120GP
026-3.6V
G5F9549L
Currently running in PCchips M919 v3.4 (testbed)

The other two 120 Mhz units I have are from Gainbery/IBM w/blue anodized heatsinks. It is unclear if these are real 120 Mhz units or 100 Mhz units, although the Gainbery box that they came in claim 120 Mhz. I've run them at 120 Mhz for months without crashing, however they crashed more frequently in other motherboards (aside from the above noted). Since the Gainbery units contained a voltage regulator, I am not certain, without looking further at the voltage regular spec sheet, if setting the voltage at 3.45 volts, was at or below the threshold of the regular. If so, it would drop the regulated voltage to the cpu further, i.e. thru a transistor PN juction, resulting in the noted less frequented stability.

The markings on the back of the Gainbery, upon removal of the chip from the socket adapter reads,

COPYRIGHT USA
1995 CYRIX
IBM9314 P50065
50H6281 PQ

I'd be happy to confirm the validity of those Cyrix 5x86-133 units and compare the benchmarks against know stable results obtained from both Cyrix 5x86-100 and Cyrix 5x86-120 units. I suppose I could throw an AMD ADZ-160 comparison in there, but from my usability perspective, it is a slug. If that Cyrix is a real unit, specifically, not renamed, it will likely be the world's fastest socket 3 processor ever made. I'd be curious to see if Cyrix got branch prediction worked out on those later 133 units.

I beleive we are both situated in the pacific northwest, so travel to/fro should not be too burdensome. I'll PM you.

[CPUShack]

the 120s fit in the data I have, Late 1995 production (Gainbery as well. its 9550)

The 133s I looked through (25+ of them) all have dates ranging from 9602 to 9607

The later ones are marked 3.7V and the earlier ones tend to be 3.6V

This makes me think that there WAS some stability problems that were corrected by testing/running at a slightly higher voltage.

[feipoa]

I suppose that given tight time constraints, and if the engineers proved that a slight increase in voltage was sufficient to qualify the cpus, then what you suggest may be valid. As an engineer, I have been faced with similar rationalizations to encourage rapid development. It is also possible that the fab/growth/doping process was altered either accidentally or intentionally which required a higher core voltage on the later 3.7V units.

Trace impurities of Silicon may have been of a higher density in a second batch of material. For example, although a second batch of raw material may have been in spec, there are still much better/worse samples in the batch that are still in spec. Having worked at a semiconductor growth facility, I can say this is a fairly common occurance. National may have run out of the better grade material for this particular client and resorted to implementing lower quality material (higher impurities) at the request from Cyrix to meet production deadlines. There may have also been several slight variances in the growth process (i.e. temperature, humidity, clean room air particles levels) which had not yet been optimized for this batch of CPUs. LOTS OF SPECULATION HERE

It is curious that you report dates ranging from what I deduce to be Feb '96 to July '96. I had thought they had ceased production long before this, which amplifies my suspicion as to whether or not these are veritable cpus.

Could you provide me with your source for decrypting the serial numbers?
From,
9550C, I can see 1995, maybe the month of May, but 0th day? Or perhaps the 50th unit of 5x86-120's in 1995, where the 'C' is for a Christmas run?! How about the IBM decryption?

The Cyrix 5x86 product manual mentions a 3.45 V operation. My Cyrix 5x86-120 chips have printed on them 3.6 V, however they run stable at 3.45 V. This fact beckons your previous thought. If those later 5x86-133's truely required 3.70 V to operate, there would not have been a motherboard out there which could output to that precision and, thus regulation would be in the hands of the upgrade module companies. It is possible to regulate to this precision off-board given the proper selection of resistances, capacitences, and a DC/DC linear regulator. If you have one of the roomered Cyrix 5x86-133 upgrade kits (perhaps from Gainbery, Evergreen, or ComputerNerd RA4), could you let me know what the numbers are on the voltage regulator?

The only authoritative source for these answers may be from Jerry Rogers himself.

[Neon_WA]

[feipoa]

From the information Neon just relayed to me, those Cyrix 5x86-133 chips were fabbed Jan 8 - Feb 16, which is closer to what I had expected.

From what I could deduce from the Cyrix product manual, the voltage printed on the surface of the chip seems to coorespond with the maximum supply voltage. The medium of min (3.3) to max (3.6) being 3.45 V, and not necessarily the ideal operating voltage.

I leaning back towards my original theory of significant digits and slight deviations in min and max operating voltages. If we knew the min and max we could derive a new nominal operating voltage. If we treat things linearly, it would be 3.55 V for those marked 3.7 V.

It may still be that the 120 Mhz unit is actually the 133 Mhz unit with bits 0 and bits 1 of the PMR register set to a 4x multiplier (10) as opposed to a 3x (11). The manual specifies that the 4x multiplier is not available on production parts, so Cyrix had another fab run to either burn in the 4x mode, or removed their logic that restricted setting the 4x mode.

These 2x, 3x, 4x multiplier rates come from the motherboard's BIOS and write to the cpu thru the CLKMUL pin. Since no 486 era board has a 4x frequency multiplier output, the Cyrix will either need to have logic that senses 2x and doubles it again, as with the Amd X5, or it is left to the user to manually enable it using a registry bit enabler.

Or if overclocking a 120 to 133, as I brought up in another forum, it would be possible to set the FSB at 33.33 Mhz, have the motherboard report 3x thru CLKMUL, and add a fractional frequency multiplier chip through a socketed interposer card. 33.33 x 1.33 x 3 = 133 Mhz. There may be off-the-shelf frequency modifiers that will work, but I haven't explored these much yet. Digikey is a good start.

[CPUShack]

It is very common to up the voltage of a higher speed grade to get higher yields.

Intel, IBM, AMD, etc have been doing so for years.

CMOS designs typically respond well to higher voltage as far as speed is concerned (limited by how fast the voltage can switch)
One of the first designs to use this fact was the RCA 1802 which responded almost linearly to voltage 1V/MHz


All of the Cyrix (and ST) 586s were fab'd by IBM, some ended up with IBM markings on the back, some with Cyrix.

[feipoa]

It may not be that the 3.6V units were earlier and the 3.7V units were later runs.

After much effort, I was able to obtain this Cyrix 5x86-133 3.7V chip with bottom marking, G5D9549A. From above, its production would be from the 49th week of 1995, which falls between my Cyrix 5x86-100 (44th week) and my Cyrix 5x86-120 (50th week).

The bottom pins contain coloured ink markings which are similar to what I've found on IBM-Cyrix 5x86-120 chips removed from upgrade socketed adapters.

[feipoa]

This Cyrix 5x86-133 ended up having the same revision as the 5x86-120's I have. Stepping 0, Revision 5.

Can anyone with one of the early 1996-era Cyrix 5x86-133's report the revision?