Still Crazy

980810 - Cyrix 5x86

Note: This page has not yet been verified!

Note: Please check and read 021005-1, 021005-2, 021005-3 for OS/2 Warp 3.

Continued from #980809 Windows 98 and PS/55.

There are few variations in Cyrix 5x86 CPU.

  1. 5x86 100GP 3x
  2. 5x86 100GP 4x (25MHz x4)
  3. 5x86 120GP 3x (basically 40MHz x3. Pre-production sample of 30Mhz x4 version rated 120GP were once sold in the market)
  4. 5x86 133 4x (33MHz x4)
  5. IBM 5x86-100 3x
  6. and others made by TI (Texas Instruments) and ST (SGS Thomson).

It's relatively easy to get 1), 3) and 5) but it's hard to get 2) and 4). I was very lucky that I got these 4x models. Among above 5 models, 4x models rated 100MHz seemed be a different beast. It even can tolerate 133MHz operation if it get cooled well (at least for a few hours. I haven't ever run it in 133MHz for a day or more). When I got 4x 133MHz model I really expected that it would be the best CPU among all of 486/586 class CPUs. But it seemed that 133MHz 4x model is a bit different from 100MHz 4x model. In the same system and with same cache setting, 4x 100MHz version and 4x 133MHz version scored different result. 100MHz rated model gives higher performance result in general as if there are major difference in core design.

Sorry I'm not an expert about this issue so I can't tell definitely yes or no)

Added on 2002/10/06: Sysbench under OS/2 Warp connect, performance result were almost same as if they are same products. Really strange...

Benchmark programs usually reports that Cyrix 5x86s are slower than AMD AMD 5x86-133 but if you run business software you'll get better performance with Cyrix 5x86 rather than with AMD 5x86. If you use an appropriate cache program to activate L1 WB function of Cyrix CPU, system performance will be boosted. One of such program is et586.exe which contained in et9060.exe provided by Evergreen. et586.exe should be placed in your config.sys with appropriate option switches in HEX value to activate registers. In order to determine HEX values you can use a nice EXL file which was written by Mr. F355, a guy who reported overclocking method of Reply Model 70 PowerBoard. To use this program, you have to install analyzing tool using TOOL > ADDIN > Analyzing Tool in MS Excel.

5551-R: EIDE without SPOCK SCSI
5551-N: SCSI, IML

Parameter setting for ET586.EXE ; Install c:\ ETI\ET586.EXE /CCR2=56 /CCR4=1D /WBE

CPU HDBENCH ETDIAG Notes
Model Clock Float Integer Dhrystone
5551-R + Win98
Cyrix 5x86-100 4x 133 MHz 3877 6585 92592 Enabled
Cyrix 5x86-100 4x 133 MHz 3762 6102 45872 NOT loaded
Cyrix 5x86-120 3x 99 MHz 3178 4970 65789 Enabled
Cyrix 5x85-120 3x 99 MHz 3093 4635 45871 NOT loaded
AMD 5x86-133 4x 133 MHz 3583 5759 65789 *3
DX4 ODP100 35 x3 105 MHz 2691 5373
PODP5V83 33 x2.5 88 MHz 2565 4962
5551-N + Win95
Cyrix 5x86-120 3x 99 MHz 3130 4690 Not recorded
Cyrix 5x86-120 3x 126 MHz 3910 5929 Not recorded
AMD 5x86-133 42 x4 168 MHz 3883 7028
5551-N + Win98
Cyrix 5x86-120 *1 120 MHz 3716 5575 53763 NOT loaded
Cyrix 5x86-120 *1 120 MHz 3704 5561 56818 Enabled
Cyrix 5x86-100 4x 133 MHz 3791 6162 Not recorded
AMD 5x86-133 4x 166 MHz 4200 6776 51020 *2

*1 No remarkable improvement observed on 5551-N with or without ET586.exe.

*2 As already reported in 980809, AMD 5x86 and IBM SPOCK will run in MS-DOS compatibility mode under Win98.

*3 Write-Back caching enabled with PL586 interposer.

et9603.exe Contains DOS, Win NT and OS/2 drivers and associated programs run the program with /d switch (or -d whatever you like) for directory sorting.

5x86rege.zip EXL file for calculating HEX values of et586 cache driver. Written by Mr. F355.

SGS4191.pdf ST5x86 Data sheet by SGS-Thomson.

5x86.exe PeterMoss's 5x86 Utility


Update. 2002.10.05

Quoted comments regarding Cyrix CPU in the newsgroup.

Tim Clark wrote to the News Group on 2002/09/01:
   IMHO, the Cyrix/IBM 5x86 is the better bet, as it has notionally better performance than even a PODP83. Especially if you can find a 4x chip for 33MHz core. This would approximate to a P100.

> It's been my experience that the POD 'feels' faster with graphical
> operating systems. YMMV.
>
> > Also, could this thing run OS/2 Warp 4? Would you get good performance?
>
> It should, although the gurus will probably tell you that Warp 3 is a
> better match for this machine.

Quite possibly, the "page-tuning" of the Warp 3 "core", although probably lost to some extent by the significant no. of FixPaks issued since, is very likely to give better memory performance on systems limited to 64MB max.

Jim Sherney replied:
   I will agree that the performance of the IBM-branded Cyrix 5x86 chips is somewhat astonishing. Dunno about the Cyrix-branded ones, I've heard rumors that IBM tweaked the microcode in their version. Meaningless benchmarks on my P70 page show that the IBM 5x86-100 compares very favorably to the AMD 133, despite the lower core speed. It resides in my P75 now. I'd love to get my hands on the 4x version.

Tim replied:
   It's a difficult question to determine at what point the Cyrix and IBM-labelled chips became the same. Initially, the M9/M1SC chip design was outsourced to various fabrication companies, IBM being only one (TI and ST also being in there). In a later deal, IBM took on the majority of the fabrication, with one of those "we'd like to brand x% for our own use" deals. At this point there was some further "production engineering" and development work done to increase yield and, I believe, shrink the die. TI and ST continued to lease the design with the older spec. well after the IBM fabrication ended, as their core voltage requirements tended to be ~3.6V rather than the Cyrix/IBM 3.45V. I may have some of the details wrong, but that's mostly correct, I think. Certainly, any Cyrix 5x86-nnn/4X will be a late IBM fab.

Content created and/or collected by:
Louis Ohland, Peter Wendt, William Walsh, David Beem, Tatsuo Sunagawa, Jim Shorney, Tim Clarke, Kevin Bowling, Tomáš Slavotínek, and many others.

Ardent Tool of Capitalism - MAD Edition! is maintained by Tomáš Slavotínek.
Last update: 06 Oct 2021 - Changes & Credits | Legal Info & Contact