VOGONS


Cyrix 5x86 Register Enhancements Revealed

Topic actions

Reply 60 of 80, by Montgomery

User metadata
Rank Newbie
Rank
Newbie

I may be lucky then. I got it up and running at 120Mhz only needed to decrease the RAM timing to Fast (instead of Fastest). What are you suggestions feipoa on the longevity of this setup? I believe the chip is running at 3.3/3.45 or at the most 3.6V (I don't know exactly because of the missing motherboard manual) with a standard heatsink and fan (currently without thermal compound). Would it be OK to use it for an extended period of time with these settings? It performs quite well anyway, 3DBench gives me exactly 100 points (or cannot display more) and Speedsys shows these results.

The attachment ST5x86 at 120Mhz.jpg is no longer available

Thanks debs3759 but the battery seemed to be functional (it stored 4.1V fully charged and held the CMOS data) the problem is it caused some issues with the clock so it had to go. It might be a faulty one though so I will experiment first with external battery and then maybe add a new internal one.

Reply 61 of 80, by debs3759

User metadata
Rank Oldbie
Rank
Oldbie

If you are overclocking without thermal paste, you processor is at high risk of overheating, as the heatsink won't make a perfect seal on the CPU, so won't keep it as cool as it needs. I highly recommend keeping it turned off, or at least not thrashing the system, until you are able to apply some reasonable paste.

See my graphics card database at www.gpuzoo.com
Constantly being worked on. Feel free to message me with any corrections or details of cards you would like me to research and add.

Reply 62 of 80, by Peter z80.eu

User metadata
Rank Newbie
Rank
Newbie

What about Cyrix (ST and TI built it, too) Cx486DX2 ? So far, at least this CPU (Cx486DX2-66 and Cx486DX2-80) have a lot of additional features, e.g. Write Back Cache (which was implemented for Intel 486DX CPUs later, with the 486DX4-100, also with the AMD 486DX-133). As far as I can see, some other CPU features found in the 5x86 are also implemented, so for socket 3/486 mainboard owners, even for the 5V only core voltage ones, this would be interesting/it would be interesting to enable these additional features. My own 486 mainboard does NOT offer any Cyrix CPU option in the BIOS, but at least two utilities support the Cx486DX also (ctchip34 and "wbon"/"wboff" found at https://github.com/karcherm/cx486wb). Unfortunately Peter Moss 5x86 tool are *not* working, also because it expects only a 5x86, regardless of the fact, that a lot of CPU registers/SMM storage addresses are the same.
Does anybody know an other Cx486DX compatible tool (and no, Cx486DRx tools are also NOT working) ?

Reply 63 of 80, by Rav

User metadata
Rank Member
Rank
Member

Nice document!

Thank you!

The attachment cpu.jpg is no longer available

So I got a 5x85, stepping 0, revision 5, more specifically a 100QP variant (The laptop chip, soldered on an interposer).
For all test, the chip is overclocked to 120Mhz.
The motherboard is an Acer A1GX-2 (with Ali 1429G chipset) with registry patched for fastest memory and 0 WS for the cache.

On DOS I made two config :
* All enabled except LOOP
* All enabled except BTB (some DOS application don't seam to like BTB)

On windows, I run it with everything except LOOP, RSTK, BWRT, and DTE.
You are right about stability going away (probably with heat?). But after testing I noticed that I don't need 3.85v. The chip is totally happy with 3.45V.
Once I switched to 3.45V, the stability issue went away (I think). So far the system is playing 320kbps MP3 non stop since about an hour (With 3.85V, it fail after less than 10 minutes).
I can still make it crash if I load the CPU to 100% and do a LOT of task switching (upload stuff by FTP while playing the 320k MP3).

From what I read before I did not expect that chip to be fine with 3.45V while overclocked at 120Mhz, but I suppose the QP variants might be higher bins as they are supposed to be in a (usually badly cooled) laptop.

Chip won't post at 150Mhz (at least not at 3.45v or 3.85v).
There is one issue on that motherboard with this CPU running 120Mhz : The board don't set the proper AT bus clock by default with mean one time out of two, pressing a key will <beep> in the BIOS / bootloader / dos boot menu. Also the bootloader will fail to initialize A20 and crash one time out of two.

But once it boot, chipset registers get patched, including AT bus, the issue go away. I use fastreboot that come with QEMM so registers stay patched on reboot (if I want to go in the BIOS without having a fight with the keyboard)

Reply 64 of 80, by feipoa

User metadata
Rank l33t++
Rank
l33t++

Where did you find your CPU? It is hard to find these QFP to PGA variants.

The Vcc rail on these PCBs is kind of noisy. You could try adding some caps to reduce this some. Might help with OC, might not. I used 100 nF for the smaller caps and 10 uF for the larger ones.

Here's mine:

The attachment Cyrix_5x86-120_QFP_1.JPG is no longer available

Plan your life wisely, you'll be dead before you know it.

Reply 65 of 80, by Rav

User metadata
Rank Member
Rank
Member
feipoa wrote on 2023-06-22, 20:04:
Where did you find your CPU? It is hard to find these QFP to PGA variants. […]
Show full quote

Where did you find your CPU? It is hard to find these QFP to PGA variants.

The Vcc rail on these PCBs is kind of noisy. You could try adding some caps to reduce this some. Might help with OC, might not. I used 100 nF for the smaller caps and 10 uF for the larger ones.

Here's mine:
Cyrix_5x86-120_QFP_1.JPG

Thanks for the tip, will look for capacitors.

I got the CPU in the local online market, there is someone who regularly sell "collector / dead" CPU. for 10-15$ a pop. Also got a TI486 DX4 there.
By "dead" he mean he can't test them 😀
I look every weeks to see what "dead" chip he his selling, Usually range from 486 to P4/Athlon.

Reply 66 of 80, by BobocoCZ

User metadata
Rank Newbie
Rank
Newbie

Greetings from the Czech Republic 😀

The attachment Cyr.jpg is no longer available

I'll revive this older thread and show you my creation. I made a PGA-168 processor using a Cyrix 5x86-120qp mobile chip. I'm running it at a stable frequency of 150MHz (voltage 4.05V) using a PCChips M918 motherboard. The performance is very interesting. Below I'll write the parameters and a few tests I've done.

Configuration:
PCChips M918 motherboard (recapped, more stable power regulator for CPU added)
Cyrix 5x85-120QP processor (150MHz - 3x50MHz FSB; 4.05V and cooling with a copper heat-pipe cooler with a 4cm FAN)
1MB DualBank SRAM cache (3-1-1-1)
32MB EDO RAM
4MB PCI SVGA Matrox Mystique
MS-DOS 6.22 and W98SE

NSSI CPU Index ... 90900
NSSI FPU Index ... 29360
Quake Benchmark 320x240 ... 21,5fps
Quake Benchmark 640x480 ... 8,4fps
Doom benchmark ... 66,69fps
PCPlayer Bench 640x480 ... 13,3fps
CPU-Z CPU Speed ... 157,4
CPU-Z FPU Speed ... 551,5
SuperPi 32K ... 30,215s
SuperPi 1M ... 1803,240s (30min 3,240s)

Reply 67 of 80, by feipoa

User metadata
Rank l33t++
Rank
l33t++

Hello BobocoCZ; welcome to the forum!

Cyrix/IBM 5x86 QFP chips tend to overclock better compared to their PGA counterparts, so I am not too surprised by your preliminary success at 150 MHz. I have tested a few at 150 MHz, but didn't try them in Windows. Did you push your system hard in Windows 95/98 to see if the CPU is long term stable? Run any CPU intensive Windows apps? Which Cyrix register enhancements are you enabling?

A proper analysis of Cyrix/IBM 5x86 QFP chips at 150 MHz is something I have been meaning to do.

Which QFP208 to PGA168 interposer are you using? Do you have a photo of the underside?

Plan your life wisely, you'll be dead before you know it.

Reply 68 of 80, by bertrammatrix

User metadata
Rank Member
Rank
Member
BobocoCZ wrote on 2025-07-08, 07:53:
Greetings from the Czech Republic :) […]
Show full quote

Greetings from the Czech Republic 😀

The attachment Cyr.jpg is no longer available

I'll revive this older thread and show you my creation. I made a PGA-168 processor using a Cyrix 5x86-120qp mobile chip. I'm running it at a stable frequency of 150MHz (voltage 4.05V) using a PCChips M918 motherboard. The performance is very interesting. Below I'll write the parameters and a few tests I've done.

Configuration:
PCChips M918 motherboard (recapped, more stable power regulator for CPU added)
Cyrix 5x85-120QP processor (150MHz - 3x50MHz FSB; 4.05V and cooling with a copper heat-pipe cooler with a 4cm FAN)
1MB DualBank SRAM cache (3-1-1-1)
32MB EDO RAM
4MB PCI SVGA Matrox Mystique
MS-DOS 6.22 and W98SE

NSSI CPU Index ... 90900
NSSI FPU Index ... 29360
Quake Benchmark 320x240 ... 21,5fps
Quake Benchmark 640x480 ... 8,4fps
Doom benchmark ... 66,69fps
PCPlayer Bench 640x480 ... 13,3fps
CPU-Z CPU Speed ... 157,4
CPU-Z FPU Speed ... 551,5
SuperPi 32K ... 30,215s
SuperPi 1M ... 1803,240s (30min 3,240s)

Nice, and pozdrav do Cech 😀

Very impressive.

I extensively messed around with IBMs (but just PGA) and an 918 last year at 120mhz (2x60). 60mhz with this board presented some issues requiring turning off some features in bios (pci buffer?) But it seemed to do pretty good at 50 with everything on. I applaud you being able to get 3-1-1-1 with 1 mb cache and the cyrix, the best I could do with stability on fast fsb was 512kb, and even for that quality vintage sram was necessary.

Are you using EDO or FPM ram? I remember this board responding very favorably speed wise to EDO (to the point that if you ran cache slower then 3-1-1-1 it would disable itself as memory was just as fast as 3-2-2-2 cache). I do recall the necessity to place the EDO ram in inner most slot for the auto detection to work properly.

I also remember the ISA bus not installing automatically under windows 98 on this board - having to manually install "isa plug and play bus" or something after a system install to get it to pick up plug and play sound cards properly - did you also run in to this problem? Just wondering if it's inherent to the board, or if it was just a side effect of me running a 60mhz fsb

Reply 69 of 80, by BobocoCZ

User metadata
Rank Newbie
Rank
Newbie
bertrammatrix wrote on 2025-07-08, 15:53:
Nice, and pozdrav do Cech :-) […]
Show full quote

Nice, and pozdrav do Cech 😀

Very impressive.

I extensively messed around with IBMs (but just PGA) and an 918 last year at 120mhz (2x60). 60mhz with this board presented some issues requiring turning off some features in bios (pci buffer?) But it seemed to do pretty good at 50 with everything on. I applaud you being able to get 3-1-1-1 with 1 mb cache and the cyrix, the best I could do with stability on fast fsb was 512kb, and even for that quality vintage sram was necessary.

Are you using EDO or FPM ram? I remember this board responding very favorably speed wise to EDO (to the point that if you ran cache slower then 3-1-1-1 it would disable itself as memory was just as fast as 3-2-2-2 cache). I do recall the necessity to place the EDO ram in inner most slot for the auto detection to work properly.

I also remember the ISA bus not installing automatically under windows 98 on this board - having to manually install "isa plug and play bus" or something after a system install to get it to pick up plug and play sound cards properly - did you also run in to this problem? Just wondering if it's inherent to the board, or if it was just a side effect of me running a 60mhz fsb

Greetings 😀 Thank you very much for the response, I can see right away that I am among the experts 😉

- M918 is a great board for me, even though in its time it was the cheapest shit often with fake cache
- I couldn't set 60MHz FSB on this board. Do you have a jumper combination for 60MHz FSB somewhere? It would be a great benefit for AMD 5x86. But Cyrix, I think it's happy even with the stable 50MHz FSB 😁
- Don't even talk to me about the cache 😁 It was a lot of hours of searching for the right chips before I achieved stability. And I have to blow the chips gently with a fan. As the temperature rises above 40°C, they start to become unstable 😁 But the most important is the TAG chip. I probably only have 2 or 3 chips that can be used as TAG with Cyrix with a 3-1-1-1 setting

I can even do some tests with the best cache settings (2-1-1-1 0WS). But it is of course not stable. Quake then has 22.4fps (or even 22.6 if both read and RAM are set to "fastest") or e.g. 14.4fps with PCPlayer 640x480. I also attach photos of Speedsys (read and RAM set to "faster") and CacheCHK (RAM set to "fastest"), which is brutal, because the EDO RAM is then completely "cached"

The attachment 20250704_082648.jpg is no longer available

Cache 2-1-1-1 0WS - 22.4fps when I set Read for RAM to "faster" or even 22.6fps when I set Read to "fastest"

The attachment 20250703_084559.jpg is no longer available

Cache 2-1-1-1 0WS and read for RAM set to "faster"

The attachment 20250703_111703.jpg is no longer available

Cache 2-1-1-1 0WS and RAM set to "fastest"), which is brutal, because the EDO RAM is then completely "cached"

I agree that once the cache is set to 3-2-2-2 with EDO memory, then the EDO memory is faster than the cache 😁 EDO support is very good for me. But with such an extreme setting, only one module should be used for stability.

- And yes, I encountered a problem with ISA in Windows and I didn't solve it 😁 I suspected that the problem would be somewhere in poor PnP support. CT3670 works perfectly for me in DOS, but in Windows only as SB16, it doesn't recognize AWE32 😀 The reason will probably be exactly what you mention.

Reply 70 of 80, by BobocoCZ

User metadata
Rank Newbie
Rank
Newbie
feipoa wrote on 2025-07-08, 10:58:
Hello BobocoCZ; welcome to the forum! […]
Show full quote

Hello BobocoCZ; welcome to the forum!

Cyrix/IBM 5x86 QFP chips tend to overclock better compared to their PGA counterparts, so I am not too surprised by your preliminary success at 150 MHz. I have tested a few at 150 MHz, but didn't try them in Windows. Did you push your system hard in Windows 95/98 to see if the CPU is long term stable? Run any CPU intensive Windows apps? Which Cyrix register enhancements are you enabling?

A proper analysis of Cyrix/IBM 5x86 QFP chips at 150 MHz is something I have been meaning to do.

Which QFP208 to PGA168 interposer are you using? Do you have a photo of the underside?

I salute you 😀

I dare say it is stable. I have currently had 1M SuperPi calculated several times and alternated it with playing 320kbit MP3 in WinAmp. But today I changed the power regulator on the board to its final form, so tomorrow I will let the assembly run all day if possible, so that I can test it properly. The key to stability for me was the power supply (the regulator on the board was not sufficient), the CPU cooling and then the cache. Without air flow around the SRAM chips, Windows becomes unstable. I have a thermal camera, so I know that it is necessary to keep the SRAM chips below 40°C. As soon as the temperature is higher, the 3-1-1-1 0WS settings start to become unstable.

The interposer was designed and made by my colleague from the Czech Republic, Rayer. He is active here on Vogons, so I hope he doesn't mind a little advertising 😉
https://rayer.g6.cz/hardware/retropc2.htm#486DX5_INTERPOSER

I performed the test with the settings:
LSSER=OFF BTB_EN=ON LOOP_EN=OFF (after switching on, great instability even in DOS) FP_FAST=ON DTE_EN=ON

I am still testing the remaining registers:
BWRT=ON Probably stable in DOS, but immediately BSOD in Windows
MEM_BYP=ON In DOS, for example, I saw a small performance increase in NSSI, but SuperPi in Windows gives slightly worse results than MEM_BYP=OFF
LINBRST=ON has no effect. I also tried turning off L1, activating LINBRST and then turning L1 back on and also with no effect (In DOS. In Windows it probably caused a system crash)
RSTK_EN=ON I think it slightly increases FPU performance

I'm still trying 😀 A few days ago I was still struggling with cache instability, so I left some registers inactive to be sure to exclude their influence. I wonder if the system will be fully stable tomorrow when running Windows. It was stable in DOS a few weeks ago. But installing Windows was the real challenge 😁 In addition, I have more SRAM chips ordered, I'm still dreaming of a stable 2-1-1-1 0WS setting at 50 MHz 😜

Reply 71 of 80, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I doubt you will get 2-1-1-1 stable on this motherboard, but I hope I am wrong. You also need to swap about the EDO sticks to check for 2-1-1-1 cache stability. From my experience, the best sticks tend to have TSOP chips rather than SOJ.

Can you show your speedsys and cachechk screenshot for 3-1-1-1?

Is 3-1-1-1 still stable?

Attached are my results from 2011 at 150 MHz.

The attachment IBM-5x86c-Thinkpad_TP365E_Mod_1.jpg is no longer available
The attachment IBM-5x86c-150-Speedsys_1.png is no longer available

I don't recall what the L2 was set at. I was using FPM. Although ugly, these Thinkpad interposers allow for a very clean Vcc rail. Since then, I have about 10 interposers from China, but with only 6 locations for capacitors, their outputs aren't as clean. Something like 35 mV noise on the Thinkpad vs. 80 mV on the Chinese interposers.

Plan your life wisely, you'll be dead before you know it.

Reply 72 of 80, by bertrammatrix

User metadata
Rank Member
Rank
Member
BobocoCZ wrote on 2025-07-08, 19:23:
Greetings :) Thank you very much for the response, I can see right away that I am among the experts ;) […]
Show full quote
bertrammatrix wrote on 2025-07-08, 15:53:
Nice, and pozdrav do Cech :-) […]
Show full quote

Nice, and pozdrav do Cech 😀

Very impressive.

I extensively messed around with IBMs (but just PGA) and an 918 last year at 120mhz (2x60). 60mhz with this board presented some issues requiring turning off some features in bios (pci buffer?) But it seemed to do pretty good at 50 with everything on. I applaud you being able to get 3-1-1-1 with 1 mb cache and the cyrix, the best I could do with stability on fast fsb was 512kb, and even for that quality vintage sram was necessary.

Are you using EDO or FPM ram? I remember this board responding very favorably speed wise to EDO (to the point that if you ran cache slower then 3-1-1-1 it would disable itself as memory was just as fast as 3-2-2-2 cache). I do recall the necessity to place the EDO ram in inner most slot for the auto detection to work properly.

I also remember the ISA bus not installing automatically under windows 98 on this board - having to manually install "isa plug and play bus" or something after a system install to get it to pick up plug and play sound cards properly - did you also run in to this problem? Just wondering if it's inherent to the board, or if it was just a side effect of me running a 60mhz fsb

Greetings 😀 Thank you very much for the response, I can see right away that I am among the experts 😉

- M918 is a great board for me, even though in its time it was the cheapest shit often with fake cache
- I couldn't set 60MHz FSB on this board. Do you have a jumper combination for 60MHz FSB somewhere? It would be a great benefit for AMD 5x86. But Cyrix, I think it's happy even with the stable 50MHz FSB 😁
- Don't even talk to me about the cache 😁 It was a lot of hours of searching for the right chips before I achieved stability. And I have to blow the chips gently with a fan. As the temperature rises above 40°C, they start to become unstable 😁 But the most important is the TAG chip. I probably only have 2 or 3 chips that can be used as TAG with Cyrix with a 3-1-1-1 setting

I can even do some tests with the best cache settings (2-1-1-1 0WS). But it is of course not stable. Quake then has 22.4fps (or even 22.6 if both read and RAM are set to "fastest") or e.g. 14.4fps with PCPlayer 640x480. I also attach photos of Speedsys (read and RAM set to "faster") and CacheCHK (RAM set to "fastest"), which is brutal, because the EDO RAM is then completely "cached"

The attachment 20250704_082648.jpg is no longer available

Cache 2-1-1-1 0WS - 22.4fps when I set Read for RAM to "faster" or even 22.6fps when I set Read to "fastest"

The attachment 20250703_084559.jpg is no longer available

Cache 2-1-1-1 0WS and read for RAM set to "faster"

The attachment 20250703_111703.jpg is no longer available

Cache 2-1-1-1 0WS and RAM set to "fastest"), which is brutal, because the EDO RAM is then completely "cached"

I agree that once the cache is set to 3-2-2-2 with EDO memory, then the EDO memory is faster than the cache 😁 EDO support is very good for me. But with such an extreme setting, only one module should be used for stability.

- And yes, I encountered a problem with ISA in Windows and I didn't solve it 😁 I suspected that the problem would be somewhere in poor PnP support. CT3670 works perfectly for me in DOS, but in Windows only as SB16, it doesn't recognize AWE32 😀 The reason will probably be exactly what you mention.

Once you install the ISA bus manually the sound cards should work normally (it should rediscover the hardware on the bus). There may be something else, like PnP bios (failsafe) or something it needs installed too, try and see what it does. The board was a real pain for me in windows before I figured that out.

I'm really impressed by your results, I do remember thinking that this board had potential. Thinking I will revisit it.

Sure, here's a picture of the jumper settings for the FSB I managed to figure out.

The attachment 20250708_192526.jpg is no longer available

I have 2 of these, my first is now a parts board. I ran it extensively at 1:1 fsb/pci at 60mhz with a sis6326...the speed was great but eventually the board started to be unreliable, until it became basically useless- I'm not sure if running it that way may have helped kill it, or if it simply died for another reason. The replacement board ended up being able to run a bit faster (3-1-1-1 instead of 3-2-2-2 with the same sram) so perhaps there was some problem

The attachment 20250708_192805.jpg is no longer available

Reply 73 of 80, by BobocoCZ

User metadata
Rank Newbie
Rank
Newbie
bertrammatrix wrote on Yesterday, 03:06:
Once you install the ISA bus manually the sound cards should work normally (it should rediscover the hardware on the bus). There […]
Show full quote

Once you install the ISA bus manually the sound cards should work normally (it should rediscover the hardware on the bus). There may be something else, like PnP bios (failsafe) or something it needs installed too, try and see what it does. The board was a real pain for me in windows before I figured that out.

I'm really impressed by your results, I do remember thinking that this board had potential. Thinking I will revisit it.

Sure, here's a picture of the jumper settings for the FSB I managed to figure out.

The attachment 20250708_192526.jpg is no longer available

I have 2 of these, my first is now a parts board. I ran it extensively at 1:1 fsb/pci at 60mhz with a sis6326...the speed was great but eventually the board started to be unreliable, until it became basically useless- I'm not sure if running it that way may have helped kill it, or if it simply died for another reason. The replacement board ended up being able to run a bit faster (3-1-1-1 instead of 3-2-2-2 with the same sram) so perhaps there was some problem

The attachment 20250708_192805.jpg is no longer available

Thank you very much, for AMD at 180 or 200 MHz it could be useful 😁

And one more important thing, what BIOS do you have? Have you managed to successfully use the 4x multiplier (for AMD) with a fully functional L2 cache? There are 3 BIOSes on RetroWeb and the 4x multiplier does not work properly with any of them. So it works, but at that moment the L2 cache stops working.

By the way, I successfully installed ISA PnP, then other HW including AWE32 PnP installed itself perfectly smoothly but from that moment on I had significantly lower performance in SuperPi 😁 By 5-10%. So I uninstalled ISA PnP again in safe mode and the higher performance is back. Strange...
I think it would be more sensible to use an older sound card without PnP.

Reply 74 of 80, by BobocoCZ

User metadata
Rank Newbie
Rank
Newbie
feipoa wrote on 2025-07-08, 22:01:
I doubt you will get 2-1-1-1 stable on this motherboard, but I hope I am wrong. You also need to swap about the EDO sticks to ch […]
Show full quote

I doubt you will get 2-1-1-1 stable on this motherboard, but I hope I am wrong. You also need to swap about the EDO sticks to check for 2-1-1-1 cache stability. From my experience, the best sticks tend to have TSOP chips rather than SOJ.

Can you show your speedsys and cachechk screenshot for 3-1-1-1?

Is 3-1-1-1 still stable?

Attached are my results from 2011 at 150 MHz.

The attachment IBM-5x86c-Thinkpad_TP365E_Mod_1.jpg is no longer available
The attachment IBM-5x86c-150-Speedsys_1.png is no longer available

I don't recall what the L2 was set at. I was using FPM. Although ugly, these Thinkpad interposers allow for a very clean Vcc rail. Since then, I have about 10 interposers from China, but with only 6 locations for capacitors, their outputs aren't as clean. Something like 35 mV noise on the Thinkpad vs. 80 mV on the Chinese interposers.

Today my system ran for 5 hours straight, in Windows 98, later in DOS, and all without any problems.

So yes, it seems stable. I was also modifying the power regulator, so in the end I increased the voltage to approximately 4.10-4.15V. The previous regulator was probably better, but it was too big 😁

Just to be sure. These results are on a stable system, 3x50MHz FSB, 1MB 3-1-1-1 0WS cache and 32MB EDO RAM with the second best setting in the BIOS ("Faster" for reading and "Fastest" for writing)

I also attach a short video where you can see the active register settings. CPU-Z incorrectly reports the cache size 😁
https://youtu.be/-9pe5x4bg8Y

The attachment 20250709_111453.jpg is no longer available
The attachment 20250708_151209.jpg is no longer available
The attachment 20250709_095903.jpg is no longer available
The attachment 20250709_095817.jpg is no longer available
The attachment 20250709_095809.jpg is no longer available

Reply 75 of 80, by BobocoCZ

User metadata
Rank Newbie
Rank
Newbie
feipoa wrote on 2025-07-08, 22:01:
I doubt you will get 2-1-1-1 stable on this motherboard, but I hope I am wrong. You also need to swap about the EDO sticks to ch […]
Show full quote

I doubt you will get 2-1-1-1 stable on this motherboard, but I hope I am wrong. You also need to swap about the EDO sticks to check for 2-1-1-1 cache stability. From my experience, the best sticks tend to have TSOP chips rather than SOJ.

Can you show your speedsys and cachechk screenshot for 3-1-1-1?

Is 3-1-1-1 still stable?

Attached are my results from 2011 at 150 MHz.

The attachment IBM-5x86c-Thinkpad_TP365E_Mod_1.jpg is no longer available
The attachment IBM-5x86c-150-Speedsys_1.png is no longer available

I don't recall what the L2 was set at. I was using FPM. Although ugly, these Thinkpad interposers allow for a very clean Vcc rail. Since then, I have about 10 interposers from China, but with only 6 locations for capacitors, their outputs aren't as clean. Something like 35 mV noise on the Thinkpad vs. 80 mV on the Chinese interposers.

I chose ram and cache for a really long time and carefully (fortunately I have quite a large stock), but still I couldn't find EDO memory that would handle the fastest settings with Cyrix at 50MHz FSB. I also have a few modules sorted out that can at least boot into DOS, but they are not stable by any means. Yes, there are also a few modules with a more modern TSOP package. Paradoxically, I also have 50ns EDO RAM, several different modules, some even 45ns and they are not stable at all.

I took a few pictures with the best settings for cache and RAM. Quake unfortunately won't finish, but it should have about 22.6fps. I'm afraid that's an unattainable goal.

I'll add that the board originally had a fake cache, so I put the sockets there myself and changed all the old electrolytic capacitors to be safe. And then the made-up regulator of course 😁

The attachment 20250709_143918.jpg is no longer available
The attachment 20250709_144017.jpg is no longer available

1.2fps increase compared to stable cache and ram settings:

The attachment 20250709_144436(0).jpg is no longer available

Reply 76 of 80, by douglar

User metadata
Rank l33t
Rank
l33t
BobocoCZ wrote on Yesterday, 14:27:

I chose ram and cache for a really long time and carefully (fortunately I have quite a large stock), but still I couldn't find EDO memory that would handle the fastest settings with Cyrix at 50MHz FSB. I also have a few modules sorted out that can at least boot into DOS, but they are not stable by any means. Yes, there are also a few modules with a more modern TSOP package. Paradoxically, I also have 50ns EDO RAM, several different modules, some even 45ns and they are not stable at all.

I find this all fascinating.

Reading the manual, looks like you have to populate ram bank 0 first and you have to use 36 bit parity simms. Is that correct?

Do the three empty banks cause signal reflections? If that's the case, could there be anything like a "terminating simm" that can be installed to make a cleaner signal without creating additional loads?

p.s. I have a M919 v3.4 where the fake cache isn't event attached, but at least it has a socket for proprietary COAST modules.

Reply 77 of 80, by bertrammatrix

User metadata
Rank Member
Rank
Member
BobocoCZ wrote on Yesterday, 14:01:
Thank you very much, for AMD at 180 or 200 MHz it could be useful :D […]
Show full quote
bertrammatrix wrote on Yesterday, 03:06:
Once you install the ISA bus manually the sound cards should work normally (it should rediscover the hardware on the bus). There […]
Show full quote

Once you install the ISA bus manually the sound cards should work normally (it should rediscover the hardware on the bus). There may be something else, like PnP bios (failsafe) or something it needs installed too, try and see what it does. The board was a real pain for me in windows before I figured that out.

I'm really impressed by your results, I do remember thinking that this board had potential. Thinking I will revisit it.

Sure, here's a picture of the jumper settings for the FSB I managed to figure out.

The attachment 20250708_192526.jpg is no longer available

I have 2 of these, my first is now a parts board. I ran it extensively at 1:1 fsb/pci at 60mhz with a sis6326...the speed was great but eventually the board started to be unreliable, until it became basically useless- I'm not sure if running it that way may have helped kill it, or if it simply died for another reason. The replacement board ended up being able to run a bit faster (3-1-1-1 instead of 3-2-2-2 with the same sram) so perhaps there was some problem

The attachment 20250708_192805.jpg is no longer available

Thank you very much, for AMD at 180 or 200 MHz it could be useful 😁

And one more important thing, what BIOS do you have? Have you managed to successfully use the 4x multiplier (for AMD) with a fully functional L2 cache? There are 3 BIOSes on RetroWeb and the 4x multiplier does not work properly with any of them. So it works, but at that moment the L2 cache stops working.

By the way, I successfully installed ISA PnP, then other HW including AWE32 PnP installed itself perfectly smoothly but from that moment on I had significantly lower performance in SuperPi 😁 By 5-10%. So I uninstalled ISA PnP again in safe mode and the higher performance is back. Strange...
I think it would be more sensible to use an older sound card without PnP.

I will have to check the bios revision to be certain, but, I never updated either of these boards so it should be whatever came with them originally. I don't remember an issue with 3x/4x, however I mainly used the board with Cyrix so if it had an issue with AMD I may have not noticed it. Could it be a jumper configuration issue?

Yes, CPU-Z also displayed only half of the cache I had installed.

One more thing I remember- I think the jumper directly below the multiplier jumper did something strange with cyrix CPUs, like enable a fractional multiplier of 1.5 or something. It wasn't useful to me at the time, but definitely interesting.

Is that voltage regulator you used an adjustable DC-DC "buck converter"?

Reply 78 of 80, by bertrammatrix

User metadata
Rank Member
Rank
Member
douglar wrote on Yesterday, 15:09:
I find this all fascinating. […]
Show full quote
BobocoCZ wrote on Yesterday, 14:27:

I chose ram and cache for a really long time and carefully (fortunately I have quite a large stock), but still I couldn't find EDO memory that would handle the fastest settings with Cyrix at 50MHz FSB. I also have a few modules sorted out that can at least boot into DOS, but they are not stable by any means. Yes, there are also a few modules with a more modern TSOP package. Paradoxically, I also have 50ns EDO RAM, several different modules, some even 45ns and they are not stable at all.

I find this all fascinating.

Reading the manual, looks like you have to populate ram bank 0 first and you have to use 36 bit parity simms. Is that correct?

Do the three empty banks cause signal reflections? If that's the case, could there be anything like a "terminating simm" that can be installed to make a cleaner signal without creating additional loads?

p.s. I have a M919 v3.4 where the fake cache isn't event attached, but at least it has a socket for proprietary COAST modules.

I've never used parity Simms on the 918 board and never had a problem in that regard.

I often as well wonder about signal reflections when these boards are getting used at around double the FSB they were designed for, I do think it comes in to play. Perhaps that's why often one particular slot works better then others, and at the same time perhaps why a double sided simm may be at times better then a single sided one that has a bunch of unconnected traces on the back side

Reply 79 of 80, by RayeR

User metadata
Rank Oldbie
Rank
Oldbie

BTW I recently designed a new interposer v2 with onboard VRM with fast response DC-DC switcher RT5789AGJ8F at bottom side. I hope it could improve Vcore clarity and maybe allows to lower Vcore a bit while keeping the stability. I'll use it for Am5x86 that is much less power hungry that this Cyrix 0,6um fat boy 😀
I measured Vcore on my Am5x86 under load on Octek Hippo 10 board (MB's linear regulator) and it's not so bad:
http://rayer.g6.cz/hardware/retropc2/hipo10d.png
http://rayer.g6.cz/hardware/retropc2/hipo10e.png
So I don't expect any miracle will happen with new VRM, maybe I could go from 3,7-3,8V to 3,5-3,6V @40MHz, 50MHz seems to be no go on that VLB MB anyway...

>bertrammatrix
for SIMM reflections - I think it should be enough to install SIMM to the last slot (physically at end of the bus) to not leaving any stubs but I'm not sure if it's significant at that speeds. Terminating DIMM was used on RAMBUS system that I barely remembered... A taky zdravim z/do Cech 😀

Gigabyte GA-P67-DS3-B3, Core i7-2600K @4,5GHz, 8GB DDR3, 128GB SSD, GTX970(GF7900GT), SB Audigy + YMF724F + DreamBlaster combo + LPC2ISA