VOGONS


Reply 120 of 219, by Tetrium

User metadata
Rank l33t++
Rank
l33t++
udam_u wrote:

@Tetrium thanks - I will be waiting for info from you. I have to admit you are very patient. If I were you I would request shipping immediately. ^^
I wonder if this cpu is good overclocker - I would expect overclocking to150MHz without need to increase voltage above 4V. (;

I doubt these CPU's are any good at overclocking though. What I've gathered is that these 5x86-133 came in 2 batches. The 1st batch was rated for 3.6v and the 2nd batch for 3.7v. Yup, 0.1v higher. Apparently they were having problems getting it to run stable at it's default speed.

And anyway, I have no problems with testing hardware, even very old hardware, that I have spares of. But since this CPU is so rare, it'll be impossible to find a replacement so I might not even put it in a board.
I'd rather try overclocking one of their 100 ones (as I don't even have the 120 one either!).
I'm sure you'll understand my position. And yes, I am very patient, but since I saved up quite a bit of chips, I'll be asking for shipment soon.
Just want to make sure some particular chips are included in the package, thats all 😉

Whats missing in your collections?
My retro rigs (old topic)
Interesting Vogons threads (links to Vogonswiki)
Report spammers here!

Reply 121 of 219, by feipoa

User metadata
Rank l33t++
Rank
l33t++

@udam_u
I used an IBM 5x86-100HF with the factory blue anodized aluminum heatsink, but I modified it to accomodate a 1.5", 12V cpu cooler. The heatsink always feels cool to the touch. I think these IBM 100HF units are actually Cyrix 120 Mhz parts because,

1) On the ID1 register, it lists Stepping 0, Revision 5, which is the same as that listed for my two known Cyrix 5x86-120 chips. My known Cyrix 5x86-100 chip lists Stepping 1, Revision 3.

2) The same registry enhancements work at 120 Mhz for both the IBM and Cyrix 5x886 cpus, but not the 100 Mhz Cyrix part. Branch prediction only seems to work on the Cyrix 5x86-100 (FYI, you actually have to turn off BWRT, LOOP_EN, and RSTK_EN to get BTB working stably in W9X/NT.).

3) The Cyrix 5x86-100 will crash at 120 Mhz, but the IBM 100HF is just as stable as the known Cyrix 5x86-120 chips.

I do all my recent testing on an M919 since my Biostar 486 is for everyday use, e.g. being used now to type this. I would like to find another Biostar MB-8433UUD v3.0 or 3.1 for testing, but I can only seem to find v2.0 units for a price nobody would never pay. Anyone have luck flashing a 2.0 unit to the latest May 20, 1996 BIOS (UUD960520s)? I killed a v2.0 board's BIOS after attempting this many years back. A v3.1 is preferred to a v3.0 as it has a socketed RTC. It was a scary exercise a few months ago to unsoldier all 24 RTC pins and soldier on a new Dallas RTC (ODIN is gone). If anyone has any spare 3.0 or 3.1's, please PM me.

@Tetrium
As I eluded to in the CPU-World forum, I am not entirely convinced that the 3.7V marking is for stability reasons at a higher voltage, although it is probably the case. We'd need to test both 3.6V and 3.7V varieties at various voltages to really determine what's up. All the Cyrix manuals for the 5x86 refer to 3.6V as being the maximum voltage, but perhaps the manual was printed way before the batching. The Cyrix 5x86-120's I have, which have 3.6V printed on the ceramic surface, run fine at 3.45V, which is what the manual states is the nominal operating voltage. Tolerance is +-0.15V. I've already worked out (on paper) a voltage regulation modification circuit to adjust Vcc to whatever I want for whenever fate drops some of these rare beasts in my lap. I might just try this mod anyway on one of my flakey MB's. EDIT: Successful. I can now run both the M919 and Biostar at 3.45, 3.5, 3.6. 3.65, or 3.7V using the 4.0V regulator jumper (or whatever voltage for that matter).

EDIT: The IBM-based Cyrix 5x86-120's that are on a socket3 adapter board were removed from the board for testing. The voltage regulation circuit on the adpater board regulates Vcc to 3.45 V, not 3.6V, as measured by my voltmeter. All the Cyrix 5x86-120 chips I've seen have 3.6V on the ceramic surface. If 3.6V was to be the nominal operating voltage, then you'd think the upgrade socket3 adpater's voltage regulator would be set for 3.6V instead of 3.45V. However, if the same upgrade company (Gainberry, in this case) wanted to use the same socket adapter for both Cyrix 5x86-100 (at 3.45V) and Cyrix 5x86-120 (at 3.6V) then they may have just left the 3.45V regulation in place.

It would be a shame to store that Cyrix 5x86-133 without testing it. If it is destined for storage and not use, it wouldn't really matter if it worked or not. In which case, you just as well test it. I've been long suspicious of whether these 5x86-133's actually work because I've never seen anyone report on their benchmarks. Cyrix may have had stability issues with them, which is why they never went to market. They may have tried to push off their stability issues onto the upgrade kit companies?

Reply 122 of 219, by retro games 100

User metadata
Rank l33t
Rank
l33t
feipoa wrote:

A v3.1 is preferred to a v3.0 as it has a socketed RTC. It was a scary exercise a few months ago to unsoldier all 24 RTC pins and soldier on a new Dallas RTC (ODIN is gone). If anyone has any spare 3.0 or 3.1's, please PM me.

On page 1 of this thread, Vogons user Markk says that he has hacked an RTC unit, in order to add an accessible coin/button battery to it. Also, if you google for: hack dallas RTC, you can read some details about this operation. For example.

Reply 123 of 219, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I do remember reading up on hack jobs for the RTC, but I do not think it is worth the risk -- other features of the RTC may be inoperative over time (i.e. square wave output, IRQ output, etc). It is a much safer route to replace the entire RTC with a new one.

You can get a new Dallas RTC for about $4 each (that's what I paid), and it will last a minimum of 10 years according to the specification sheet. Considering my Biostar was 13 years old before the RTC gave way, I consider this timeframe very realisitc for a soldier job. It took me about a hour to desoldier and resoldier a new one on, but next time (10 years from now), I'll soldier on the 24-pin socket extension piece for quick replacement, as is on your Biostar v3.1 board.

Reply 124 of 219, by feipoa

User metadata
Rank l33t++
Rank
l33t++

To shed some further light on the question I posed earlier about how to determine which PCI revision exists on your 486-era motherboard, the programs PCI104VK and PC-Config both say, "shows PCI-BIOS version 2.10" on my M919. This seems to be in agreement with Speedsys's claim for PCI 2.10 as well.

One other interesting tid-bit I found is that Speedsys 4.70 gives slightly lower scores than Speedsys 4.78. I had posted results earlier with 4.70. 4.78 also correctly displays the Cyrix 5x86 cpu speed, video/memory bandwidth, as opposed to 4.70.

Has anyone run Bytemark (bytecpu.zip) in Win98SE or WinNT? I find that it hangs once it passes the Huffman tests This is very unfortunate because you need to pass this stage to get to the more important results (Interger and floating point performance). The program says that it needs the 32-bit NT mode to run, so it won't run in DOS. Anyone find the DOS equivalent, or had any luck with the 32-bit version in Win98/NT? If you've had luck, which processor were you using?

For anyone else who's a fan and big user of the Cyrix 5x86 cpus, enabling the enhancements in Roy Longbottom's benchmarks yields a,
1% increase in the VAX MIPS rating - DHRY10D (Drystones, score=124)
20% increase in Rolled Double Precision - LINPCOD (6.74 Mflops)
33% increase in (N1) Floating point - WHETCOD (Whetstones, N1 FPscore = 15.0)
12% increase in MWIPS (53.86 MFLOPS)

Reply 125 of 219, by udam_u

User metadata
Rank Member
Rank
Member

@feipoa
AFAIK steppings and revisions refer to improvements acquired during manufacturing process to fix functional bugs and sometimes improve manufacturing process. For me this mean only that your Cx5x86100MHz is newer than your IBM100MHz and cx5x86120MHz processor. However your observation can be very important - maybe changes added in later cx5x86 steppings and revisions increased chip complexity, that cause problems when it comes to work at higher CPU frequency but allow to use branch prediction. As far as I remember my two cx5x86100MHz also have stepping 1 and revision 3. If other people could check and publish steppings and revisions their cx5x86 processors this would be very interesting and instructive for us. (;

@Tetrium

I doubt these CPU's are any good at overclocking though. What I've gathered is that these 5x86-133 came in 2 batches. The 1st batch was rated for 3.6v and the 2nd batch for 3.7v. Yup, 0.1v higher. Apparently they were having problems getting it to run stable at it's default speed.

Hmm literally this mean that 3.7V CPU should work at 4V because I don't know any 486 motherboard which allow to set 3.7V.

But since this CPU is so rare, it'll be impossible to find a replacement so I might not even put it in a board.

I understand this cpu is rare as hell but if I were you I would test it even at nominal settings. This will allow you to check whether it works at all and additionally publish its stepping and revision?

Regards! (:

Reply 126 of 219, by Tetrium

User metadata
Rank l33t++
Rank
l33t++
feipoa wrote:

@Tetrium
We'd need to test both 3.6V and 3.7V varieties at various voltages to really determine what's up.[/b]

This may be a problem, good luck finding both these chips 😒

Anyway, I might test it when I get it, but I'm in no hurry.
I might test it along with a whole batch of other 486 CPU's I have acquired in the last year or so, all in 1 go.

I don't think the 3.6v and 3.7v parts are any different from eachother, since their production dates are so similar (as read on cpu-world).

What programs should I use to ID it? Any download links? Or better, upload them here so I can find the required programs with ease when I actually need them. If I download them now, chances are I can't find it anymore when I need it 🤣!

Whats missing in your collections?
My retro rigs (old topic)
Interesting Vogons threads (links to Vogonswiki)
Report spammers here!

Reply 127 of 219, by feipoa

User metadata
Rank l33t++
Rank
l33t++

@udam_u

Cyrix 5x86-100
Stepping 1
Revision 3
Fabbed 44th week of 1995

Cyrix 5x86-120
Stepping 0
Revision 5
Fabbed 50th week of 1995

It could be that the stepping/revisions numbers do not cross over between different speeds of the 5x86, that is to say that maybe stepping 0, rev5 of the 120 part contains improvements that the 100 unit doesn't.

For the 120 part to pseudo-work with branch prediction, LOOP and RSTK must be off, but BWRT can be on, unlike for the 100 part that needs all three of these registers off. The 120 eventually crashes with branch prediction on, unlike the 100 unit. But at 100 Mhz, with BTB (branch prediction) on, it is not faster than the 120 Mhz unit with all over enhancements on (except BTB).

I beleive that may be part of the reason why Cyrix allegidly sent all their 5x86-133 products to upgrade unit companies, that is, so that the upgrade companies would be responsible for up regulating the voltage to 3.7V. My motherboards all have 3.7V settings now :)

@Tetrium

Agreed for the most part, but they did have a 7 week run of the 133 Mhz cpu based on what cpushack mentioned on cpuworld. From what I've read, Cyrix's CEO pushed their engineers quite hard, so I wouldn't be terribly surprised if a slight modification as made. We've definately made such last minute component mods in less time where I work -- but sending stuff to/from China is cheap these days, maybe not so in the US back in 1995/6? The revision numbers of the chips may shed some light onto this.

Without getting into the register moding programs, an easy way for you to check the revision number of your Cyrix 5x86-133 is likely with a program that you already have. Its called ckcpu16, or sometimes chkcpu. Let me know if you don't have it. I've noticed people in this forum generally use it to check cpu, bus, and CPUID information. The Cyrix 5x86's I have do not have CPUID enabled (apparently some do?), but chkcpu still lists the registers that mention the model/stepping, you just need to decode it a little.

Example 1.1
For a Cyrix 5x86-120 chkcpu reports,
DIR0=2Dh
DIR1=5h

The 'h' just means that the number is in HEX. The 2D in DIR0 means that the cpu is intended to run with a 3X multiplier. If you have a real 4X Cyrix 5x86, DIR0 should be 2E, or 2C. FYI, 28 and 2A are for 1x, 29 or 2B are for 2x, 2D or 2F are for 3x.

For DIR1, convert the number to binary and associate the first 4 bits with the revision, and the last 4 with the stepping. For example, 5 (HEX) = 00000101 (BIN). The first 4 bits are 0101 (BIN), which is 5 in DECIMAL. This is the Revision. The next 4 bits, 0000 is still 0 in DECIMAL. This is the Stepping.

Example 1.2
For a Cyrix 5x86-100 chkcpu reports,
DIR0=2Dh
DIR1=13h

DIR0, still 3x.
DIR1, 13 (HEX) = 00010011 (BIN). 0011 (BIN) = 3 (DEC). Revision 3. 0001 (BIN) = 1 (DEC). Stepping 1.

There is a small possibility that your 4X unit is in 3X mode that you will need to enable it to 4X via the registers. Cyrix may have left this up to the upgrade companies? I'm speculating. By setting the PMR register, you can set your Cyrix to 2x, 1x, 3x, but the 4x command was ignored on 100/120 Mhz cpus. I tried this already! I was working on a PLL circuit to force an output waveform of 133 Mhz to my 120 Mhz cpu on 3x, but after trying my 120 at 150 Mhz, I determined that these cpus do not overclock well at all, and that this mod may be a waste of my free time. 66Mhz x 2 = 133 Mhz also failed.

Reply 128 of 219, by udam_u

User metadata
Rank Member
Rank
Member
Cyrix 5x86-100 Stepping 1 Revision 3 Fabbed 44th week of 1995 […]
Show full quote

Cyrix 5x86-100
Stepping 1
Revision 3
Fabbed 44th week of 1995

Cyrix 5x86-120
Stepping 0
Revision 5
Fabbed 50th week of 1995

It could be that the stepping/revisions numbers do not cross over between different speeds of the 5x86, that is to say that maybe stepping 0, rev5 of the 120 part contains improvements that the 100 unit doesn't.

Hmmm I'm a little bit puzzled. I always thought that stepping refer to major CPU changes and revision to minor CPU changes. In this case cx5x86 100MHz should be never than cx5x86 120MHz, but your stepping 0 revision 5 120MHz processor is newer than 100MHz stepping 1 revision 3 what suggests I was wrong...

For the 120 part to pseudo-work with branch prediction, LOOP and RSTK must be off, but BWRT can be on, unlike for the 100 part that needs all three of these registers off. The 120 eventually crashes with branch prediction on, unlike the 100 unit. But at 100 Mhz, with BTB (branch prediction) on, it is not faster than the 120 Mhz unit with all over enhancements on (except BTB).

It is all true - only branch prediction worked correctly on my cx5x86 100MHz...what a pity I haven't got any cx5x86 stepping 0 revision 5 CPU! I really like cx5x86 CPU family and I would like to test all its hidden enhancement features! Have you got any spare Cx5x86 120MHz..? (;

I beleive that may be part of the reason why Cyrix allegidly sent all their 5x86-133 products to upgrade unit companies, that is, so that the upgrade companies would be responsible for up regulating the voltage to 3.7V. 😀

He he he I agree with you. I read somewhere on Internet that Cyrix 5x86 was nicknamed the "Chili" because it was very hot when compared to other CPUs from its period.

My motherboards all have 3.7V settings now

Have you changed resistor on motherboard?

Regards! (:

Reply 130 of 219, by Tetrium

User metadata
Rank l33t++
Rank
l33t++

I have a MediaGX motherboard and 2 CPU's but I can't get it to post, alas.

Whats missing in your collections?
My retro rigs (old topic)
Interesting Vogons threads (links to Vogonswiki)
Report spammers here!

Reply 131 of 219, by feipoa

User metadata
Rank l33t++
Rank
l33t++

@udam_u
The stepping/revision thing confused me as well. If I were to speculate, Cyrix probably realised that with branch prediction working, other enhancements did not (BWRT, LOOP, RSTK). An executive decision was probably made to not include this stepping rev. in the 120 Mhz cpus due to uncertainty in how it would perform, and perhaps due to the timeline to market. Risk assessment is a huge deciding factor. I am also not entirely sure if we can cross stepping/revisions between 100 Mhz and 120 Mhz versions like this.

Yes, to change the regulated voltage, i.e. for the 4.0V jumper, you need to find the surface mount resistor that corresponds to Vadj. of the regulator (for that jumper setting only). Vreg., for example, can be measured on PGA pins C4 and B4.

Extra Cyrix 5x86-120's... If this wasn't my daily machine, I'd say yes, but alas I need the backup proc. for this system in case of failure. The exception to this would be if you wanted to trade a working Cyrix 5x86-133 for a 5x86-120. :) In the past few years, most webpages have been going the route of Flash, which has made web browsing slow or impossible on this machine -- always good to have a 'modern' PIII on your KVM for those troublesome webpages.

For playing around with register enhancements, DOS is generally most forgiving, followed closely by Win9X, then further down the line, NT4/W2K.

@swaaye
I've always been intrigued by the MediaGX, but am afraid it would open a can of worms with my Cyrix 5x86 obsession. It could end up being a money pit of doom. I'm also not too keen on being forced to use MediaGX specific motherboards.

Reply 132 of 219, by sliderider

User metadata
Rank l33t++
Rank
l33t++
retro games 100 wrote:

I've just found out something useful. When I ran those Win98 tests above, I was using 2 sticks of EDO RAM. Each stick was 32MB, total 64MB. I'm doing some more DOS testing now, and so I set the BIOS settings back to their original very aggressive settings. I noticed this caused a problem, which did not occur before. I can't boot up DOS any more. What's the problem?

Well, to solve the problem, I removed 1 SIMM stick. I reduced the amount of RAM from 64 to 32 MB. Now, DOS will boot. Sure, I could get DOS to boot OK with 64 MB of RAM, but the crucial thing is that I needed to make the BIOS timings less aggressive. So, if you want to max out your BIOS settings, and make them as fast as possible, consider removing some memory from your 486 system.

Edit update: I just tried a Dell Nitro S3 GX 4MB PCI video card, and I get 37 points for the Shiny "CPU/Video Gamers Performance Test v1.4" benchmark program! This download link can be found in @udam_u's post above.

Going backwards for a second, could it be possible that it was the memory stick you removed that was not responding well to the BIOS settings? Did you try any others to get the memory back up to 64 megs and bootable in DOS with the fast memory timings?

Reply 133 of 219, by feipoa

User metadata
Rank l33t++
Rank
l33t++

Have you tried running Memtest version 4 from a boot floppy and letting it finish the run?
http://www.memtest86.com/

This case sounds similar to a case I had which was tough to troubleshoot. I ran this system with the fastest ram and cache settings; everything was set to fast and the system ran well for about 2 years. As time went, the system would only sometimes boot on the fastest setting. Months later I had to slow down the cache setting to obtain stability. As time went on further, even the slowest cache setting was yielding hit-or-miss boots.

This case occured using a 486 motherboard with 512 KB cache (single bank), in fact it was a Biostar MB-8433UUD. No errors came up with Memtest, which I found odd. I replaced the cache modules with a new set of 256 KB cache (double banked) and everything worked great on the fastest setting. I no longer run with the fastest cache setting after witnessing this, even though the system appeared stable after 7 days of uptime.

Plausible causes:
1) Cache TAG RAM gone bad
2) Cache gone bad
3) Malfunctioning Northbridge (memory controller), such that it can no longer cache 512 KB cache.
4) Malfunctioning Northbridge, such that it can only work with interleving cache (double banked)

Dislocation ghost strikes again?

Suggestions for rg100:
1) Run Memtest for 4 hours. If it fails, switch out the RAM modules until you find the faulty module. If all RAM modules fail, increase your cache/RAM wait states and retest.
2) Swap out your cache.
3) I've noticed that sometimes the WT cache setting is more problematic than the WB setting, oddly enough.

Reply 134 of 219, by retro games 100

User metadata
Rank l33t
Rank
l33t

I'll try all of these interesting ideas when I next test the Biostar mobo. ATM, it's somewhere - not exactly sure where. I'm sorry but right ATM, I've got a box of untested stuff that I need to plough through...

Reply 135 of 219, by sliderider

User metadata
Rank l33t++
Rank
l33t++

Relative to Pentium, how fast is a 200mhz 486? About a Pentium 100? If that's as fast as it goes, wouldn't a Pentium Overdrive 83 have more overclock potential on this motherboard since it's almost to 100mhz to start with?

Reply 136 of 219, by swaaye

User metadata
Rank l33t++
Rank
l33t++
feipoa wrote:

@swaaye
I've always been intrigued by the MediaGX, but am afraid it would open a can of worms with my Cyrix 5x86 obsession. It could end up being a money pit of doom.

😁 Oh I know how that goes. One thing leads to another and soon there are 10 packages on the way.

sliderider wrote:

Relative to Pentium, how fast is a 200mhz 486? About a Pentium 100? If that's as fast as it goes, wouldn't a Pentium Overdrive 83 have more overclock potential on this motherboard since it's almost to 100mhz to start with?

Yeah a Pentium tends to be about twice as fast at the same clock rate unless you are using the FPU a lot or have an app with significant Pentium optimizations. In some cases Pentium may be about 3x faster per clock because of optimizations in addition to its vastly faster platform.

Reply 137 of 219, by retro games 100

User metadata
Rank l33t
Rank
l33t
swaaye wrote:
feipoa wrote:

@swaaye
I've always been intrigued by the MediaGX, but am afraid it would open a can of worms with my Cyrix 5x86 obsession. It could end up being a money pit of doom.

😁 Oh I know how that goes. One thing leads to another and soon there are 10 packages on the way.

I can relate to that. 1 new thing, and 9 back ups. 😉

Reply 139 of 219, by retro games 100

User metadata
Rank l33t
Rank
l33t
sliderider wrote:

So has anyone tested a Pentium OD with these boards? What clock speeds, FSB and multipliers are doable? I'm guessing even 2.5 x 40 or 3 x 33 is going to be pushing it.

You participated in this thread, here. In it, I mention that the POD works very well on this mobo. The POD overclocks to 100 MHz, but Quake is not stable at that speed.