Is there anyone here who could try to benchmark and upload a WTX file wih their cyrix cpus at 133mhz with wintune97?

What’s the objective?

I am collecting benchmarked systems to my wintune97 database.

Post your Wintune97 scores here!

I have a really big ever growing database with systems. I have a couple of cyrixes, but no GP 133.

Wintune97 is awesome for comparing systems against other systems. Check out the thread above.

Why not use Wintune98? You could pull the results from just about every x86 CPU from the data table in the PDF of the Ultimate 686 benchmark comparison.

feipoa wrote on 2023-06-04, 18:11:

Why not use Wintune98? You could pull the results from just about every x86 CPU from the data table in the PDF of the Ultimate 686 benchmark comparison.

Well because I have been collecting systems since the program was released. My database is big. It contains hundreds of benchmarked systems.

I really like my wintune97 database. I am sharing the database with everyone as well.

Edit: I have attached the latest database with this post. It is NOT rared. Just delete the .rar extension and the file will work.

How do you determine stability of Cyrix 5x86 system? I found my Cyrix 5x86 is very sensitive at thermal condition. I’m now building Cyrix 5x86 120MHz system working at 133MHz (4x, 33MHz). Case is old type with no consideration of air flow and I used several fans to improve airflow. I mainly run GL QUAKE to test stability in Windows 95 and Latest result is around 16.5 hours of running (finally GL QUALE stopped with error but could normally shut down Windows 95 and restart was successful without any problem). Current Cyrix enhancements resistor is set as normally recommended one (/BTB_EN=off /LOOP_EN=on /LSSER=off /RTSK_EN=on /FP_FAST=on /MEM_BYP=on /DTE_EN=on /BWRT=on /LINBRST=on)
I will run another test with /RTSK=off and /DTE_EN=off with others are the same with above but I wonder how to judge stability. Now I have one question arises that what is judgement criteria of stability. Technically it can be said unstable but 16+hours is not normal condition in my case (I may not be able to run GL QUAKE 16+hours continuously) and also even though it passes 24+ hours it cannot be guaranteed to run 30+hours, 48+hours and so on.
I wonder normally how CPU makers quality control stability of CPU for normal consumers (not for server user) and what is its criteria. And also I wonder retail Cyrix 5x86 133MHz CPU can pass 24+ hours with the same enhancement registers conditions above. For reference my Cyrix 5x86 CPU is normal 120GP S0R5 and mainboard is PCCHIPS M919.

All Cyrix branded 120s capable of 4X?

I suppose a simplified description of stability might be that the system doesn't crash for your needs or desires. For my Cyrix systems, I run several dozen apps, from 3D Windows games, to Photoshop, IE6 with website loadings, mp3 loop playback while doing other tasks, benchmarks and whatever else I have on the HDD.

I have found that when running IBM/Cyrix 100/120 MHz chips at 133 MHz, that tests during the hot summer months are more telling, especially inside cases. I have an IBM 5x86C-100HF QFP chip that I marked as a "pass" when testing on the desktop, but after I put it in a case and ran it in the summer, I started to get random errors after a hour or two of uptime. By contrast, two other -100HF QFP chips did fine.

I have tested three Cyrix 5x86-120/4x chips at 133 MHz and they all eventually failed my testing criteria. I'd be surprised if your chip is a pass at 133 MHz. With regard to running GLQuake in a loop for 16 hours - that may be a good start, but find apps that will crash it faster, e.g. run GLQuake in loop for 2 hours, then close GLQuake and start opening other Glide games, while simultaneously opening Photoshop or IE6, for example. I find random user tasks quicker to crash a system than running the same Quake loop over and over again.

What voltage are you running your Cyrix at? I'd suggest 3.75 V as a good starting point, then taper down if possible.

dirkmirk: From my experience, only the Cyrix 5x86-120 chips which have a "4x" printed on the surface can do 4x.

feipoa wrote on 2024-10-09, 04:53:

I suppose a simplified description of stability might be that the system doesn't crash for your needs or desires. For my Cyrix […]
Show full quote

I suppose a simplified description of stability might be that the system doesn't crash for your needs or desires. For my Cyrix systems, I run several dozen apps, from 3D Windows games, to Photoshop, IE6 with website loadings, mp3 loop playback while doing other tasks, benchmarks and whatever else I have on the HDD.

I have found that when running IBM/Cyrix 100/120 MHz chips at 133 MHz, that tests during the hot summer months are more telling, especially inside cases. I have an IBM 5x86C-100HF QFP chip that I marked as a "pass" when testing on the desktop, but after I put it in a case and ran it in the summer, I started to get random errors after a hour or two of uptime. By contrast, two other -100HF QFP chips did fine.

I have tested three Cyrix 5x86-120/4x chips at 133 MHz and they all eventually failed my testing criteria. I'd be surprised if your chip is a pass at 133 MHz. With regard to running GLQuake in a loop for 16 hours - that may be a good start, but find apps that will crash it faster, e.g. run GLQuake in loop for 2 hours, then close GLQuake and start opening other Glide games, while simultaneously opening Photoshop or IE6, for example. I find random user tasks quicker to crash a system than running the same Quake loop over and over again.

What voltage are you running your Cyrix at? I'd suggest 3.75 V as a good starting point, then taper down if possible.

dirkmirk: From my experience, only the Cyrix 5x86-120 chips which have a "4x" printed on the surface can do 4x.

Thanks for your comment.
I didn’t mod and stock voltage is supplied to CPU. As you know, PCCHIPS M919 only supply 3.3V, 4V and 5V. At first I worried 3.3V may be too low for Cyrix 5x86 but I found real voltage supplied is 3.5x V (I remember it was 3.52V with IBM 5x86 CPU running). Strangely 4V and 5V is nearly exact (I remember 4.0-4.1V is supplied for 4V setup and 5.0-5.1V is supplied for 5V setup).I guess PCCHIPS might intentionally set supply voltage higher than 3.3V nearly 3.5V or 3.45V target for CPUs requiring 3.45V or more)

And my 5x86 CPU is with no 4X mark and produced in 2nd week of 1996. Please refer to the attachments.

As for stability, I understand your comment that find my purpose and confirm it is fit for my purpose or not. Then this may be enough for my case. I think I may run some games for several hours maximum with this PC….^^. Let me some test such as running GL Quake several hours and then load word or excel and so on. But I cannot imagine I will run photoshops or IE and so on with this PC. I wonder how retail CPU producers set stability criteria such as Intel or AMD, and Cyrix in this case.

I suggest you modify the M919's VRM to output variable voltage. 3.5 V is normally insufficient for a Cyrix 5x86 at 133 MHz. 4 V is too high.

Most boards I've tested, which specify a 3.3V setting, have measured between 3.4 V and 3.5 V.

Most Cyrix PGA chips with a 1996 datecode can do 4x.

I have also wondered what testing scenario Cyrix used for their 5x86. In fact, I haven't heard about any test jig used by any of the main CPU manufacturers from the 90's.

feipoa wrote on 2024-10-09, 06:32:

I suggest you modify the M919's VRM to output variable voltage. 3.5 V is normally insufficient for a Cyrix 5x86 at 133 MHz. 4 V […]
Show full quote

I suggest you modify the M919's VRM to output variable voltage. 3.5 V is normally insufficient for a Cyrix 5x86 at 133 MHz. 4 V is too high.

Most boards I've tested, which specify a 3.3V setting, have measured between 3.4 V and 3.5 V.

Most Cyrix PGA chips with a 1996 datecode can do 4x.

I have also wondered what testing scenario Cyrix used for their 5x86. In fact, I haven't heard about any test jig used by any of the main CPU manufacturers from the 90's.

Let me consider mod VRM. Before it I should test several tests you recommended, running Quake then several other apps only or simultaneously. And also I wonder disable RTSK_EN and DTE_EN affect stability or not. I will run Quake only test also with these two registers disabled.

BTW, I’m building this for all rounder 486 PC covering 386(or fast 286)-late 486 (or early Pentium) performance. I configured turbo button to choose FSB 33MHz or 25MHz with 4x fixed multiplier. Using SETMUL 6 different clock (25MHz to 133MHz) can be selected. Including L1 cache enable/disable and Cyrix Enhancements enable/disable with different clock selection, this will cover 386-late 486 PC performance. In fact choosing 25MHz to 100MHz clock is enough for my purpose and 133MHz is considered to be bonus to me. Anyway the more, the better. If 133MHz is more stable it would be better.

Thanks for your comment again.

RSTK and DTE don't do much for performance, so disabling them should be OK. About 15 years ago, I ran a Cyrix 5x86-133/4x in a cased system that was on 24/7. It ran my HTTP and FTP servers in NT4 and was also used for very light web browsing. I ran it at 3.7 V. Every few days, I'd get a BSOD. It took a lot of time to troubleshoot, but on this particular CPU, disabling MEM_BYP resolved the issue with BSODs. Since 133 MHz is nearing the ceiling of what these CPUs can be run at, it wouldn't surprise me if any of these register settings can be causing issues at the ceiling. I also ran into a case whereby a 120 MHz CPU ran fine at 120 MHz, but at 133 MHz, I had to enable LSSER (optimal is disabled). As such, it might be best to start with all the enhancements disabled, then enable them one-by-one.

Then it would be best to just enable FP_FAST and LESSER (which are the most important top 2 registers for performance according to your test results) and with all other registors disabled if stability is required. In fact I’m thinking that these parameters are factory disabled properties and it means Cyrix does not gurantee stability with these registers enabled. So this should be determined according to users specific purpose as you pointed out and it would be no surprise any of these parameters affect stability.

I chose RTSK and DTE to exclude according to your test results and other users comment about stability. I hope this improves stability.

feipoa wrote on 2024-10-09, 07:12:

RSTK and DTE don't do much for performance, so disabling them should be OK. About 15 years ago, I ran a Cyrix 5x86-133/4x in a cased system that was on 24/7. It ran my HTTP and FTP servers in NT4 and was also used for very light web browsing. I ran it at 3.7 V. Every few days, I'd get a BSOD. It took a lot of time to troubleshoot, but on this particular CPU, disabling MEM_BYP resolved the issue with BSODs. Since 133 MHz is nearing the ceiling of what these CPUs can be run at, it wouldn't surprise me if any of these register settings can be causing issues at the ceiling. I also ran into a case whereby a 120 MHz CPU ran fine at 120 MHz, but at 133 MHz, I had to enable LSSER (optimal is disabled). As such, it might be best to start with all the enhancements disabled, then enable them one-by-one.

BTW, I feel 3DMark99 Max seems very unstable. Typically Voodoo2 3DMark99 test is done without problem but Matrox G200 PCI always hang up during test for my experience. I read you also used G200 PCI in your Cyrix 5x86 system and I wonder 3DMark99 test using G200 was OK to you or not. And also I wonder how you think 3DMark99 for app to test stability .

feipoa wrote on 2024-10-09, 07:12:

RSTK and DTE don't do much for performance, so disabling them should be OK. About 15 years ago, I ran a Cyrix 5x86-133/4x in a cased system that was on 24/7. It ran my HTTP and FTP servers in NT4 and was also used for very light web browsing. I ran it at 3.7 V. Every few days, I'd get a BSOD. It took a lot of time to troubleshoot, but on this particular CPU, disabling MEM_BYP resolved the issue with BSODs. Since 133 MHz is nearing the ceiling of what these CPUs can be run at, it wouldn't surprise me if any of these register settings can be causing issues at the ceiling. I also ran into a case whereby a 120 MHz CPU ran fine at 120 MHz, but at 133 MHz, I had to enable LSSER (optimal is disabled). As such, it might be best to start with all the enhancements disabled, then enable them one-by-one.

I recall 3dmark99 and 99max having issues with Cyrix 5x86 and 6x86 CPUs. The results were abnormally low. I think mkarcher made a fix for this, somewhere in here: Re: 3dmark99 MegaThread
I haven't tested it yet. Nonetheless, 3dmark99 shouldn't hang on Cyrix 5x86 CPUs. The first thing you should do is increase your voltage by adding a trim pot in place of the resistor that corresponds to the 4 V setting on the M919.

feipoa wrote on 2024-10-09, 08:38:

I recall 3dmark99 and 99max having issues with Cyrix 5x86 and 6x86 CPUs. The results were abnormally low. I think mkarcher made a fix for this, somewhere in here: Re: 3dmark99 MegaThread
I haven't tested it yet. Nonetheless, 3dmark99 shouldn't hang on Cyrix 5x86 CPUs. The first thing you should do is increase your voltage by adding a trim pot in place of the resistor that corresponds to the 4 V setting on the M919.

Yes, that patch is for all processors that do not support RDTSC instruction. It will fix low scores on those systems. Nothing else has been changed.

dirkmirk wrote on 2024-10-09, 04:40:

All Cyrix branded 120s capable of 4X?

Every cyrix branded 5x86 I’ve ever owned was 2x/3x. I don’t think the 4x enabled ones are nearly as common. Most of the 1996 models are 3x/4x , but I have an oddball from 1996 that’s still 2x/3x. I believe it’s even made by ST, not IBM.

The ST branded Cyrix 5x86 chips from 1996 are harder to determine. I have one that is 3x/4x and another that is 2x/3x. I have a Cyrix branded QFP chip from 1996 that is 2x/3x.

One interesting thing about Cyrix 5x86 CPU is that informations on its surface varies.
According to below site, there are many variation of information on its surface.

https://www.cpushack.com/cyrix-486-cpus/

Marority of Cyrix 5x86 133MHz seems to have speed rating, multiplier and voltage and so on on its surface. One scenario I can think of is that These CPUs were supplied to other producers who made CPU modules such as Evergreen, IO-DATA and so on. Informations on its surface may be those requested by CPU module producers who should set voltage of its CPU module. I cannot imagine that Cyrix printed these informations for normal consumers who cannot set or modify supply voltage and also I don't know which mainboard can supply specific voltage of 3.6V or 3.7V. If this scenario is right, we may know these CPU were made for which customers. For example Cyrix 5x86 with only speed ratings may be for normal users.

And also, I could find different markings existed in the same production lot.
For example, G5LB602J sold with two different markings, one is with no speed rating and the other is 133/4X.
Another example is G5FB603B. This lot was sold as two speed rating, one is 133/4X and the other is 120/4X. One scenario that I can imange is that Cyrix did test on every single CPU of production and marked speed rating according to its results. This means Cyrix 5x86 CPU production was not stable and there were variations even affecting speed rating. I suspect this scenario in that testing every single CPU requires a lot of testing units, persons and testing time. It will affect Cyrix production schedule also and will reduce income of Cyrix. I think it is more possible scenario that Cyrix tested some samples of the lot. The other scenario is that Cyrix got specific volume of orders of each CPU and according to production schedule marked different speed ratings. In this case they should have confirmed some or all of its products can meet their QC criteria of 133MHz. There should be no CPUs not meeting 133MHz requirements in the lot, otherwise they cannot guarantee 133MHz capable ones supplied to customers who ordered 133MHz CPUs. I wonder how they controlled CPU ratings in the same production lines, I guess each production line may be the same for certain range of CPU. Anyway these are all my imagination. If there are someone who worked for Cyrix and know how Cyrix controlled production and quality control, it would be pleasure to reveal some information to us.

If the information above site is right, there should be many scenarios why Cyrix did print different markings.
During building Cyrix 5x86 system, I could find many interesting points.

Could you let me know which resistor should be changed? There are several resistors near voltage setting jumpers.(Mine is M919 3.4F) I guess it may be one of them (R50, R51, R52) is related to 4V setting.
And 5kohm trim pot is OK for this mod? For example below one.
https://www.amazon.com/Projects-Variable-Resi … r/dp/B08MDJ4JD7

It seems all of Cyrix 5x86 133MHz have voltage information on them and it is 3.6V or 3.7V. I guess this information is for other producers to set voltage. I suspect Cyrix 5x86 133MHz inherently has very narrow operation window of voltage therefore Cyrix might sold them to mainly CPU module producers who can set voltage. Cyrix might not be able to sell 133MHz version to commercial normal users if its voltage operation windows is too narrow. If this assumption is right, 3.6V-3.7V could be sweet spot for 133MHz operation according to experiments results of Cyrix.

feipoa wrote on 2024-10-09, 08:38:

I recall 3dmark99 and 99max having issues with Cyrix 5x86 and 6x86 CPUs. The results were abnormally low. I think mkarcher made a fix for this, somewhere in here: Re: 3dmark99 MegaThread
I haven't tested it yet. Nonetheless, 3dmark99 shouldn't hang on Cyrix 5x86 CPUs. The first thing you should do is increase your voltage by adding a trim pot in place of the resistor that corresponds to the 4 V setting on the M919.