VOGONS


First post, by squelch41

User metadata
Rank Member
Rank
Member

I need some help!
The aim - to get a 933MHz Pentium 3 1.75v CPU stable.

The situation:
motherboard - msi 6119 440bx
The CPUs
- a 'proper' slot 1 450Mhz pentium 3 2.0v katami 100mhz fsb
- a celeron 1.75v [edit] 1.1GHz 100mhz fsb [edit] - SL5XU
- a pentium 3 1.75v 933MHz 133 fsb [edit] - SL5U3
- 2 slockets - slocket 'a' is a PGA slocket modified to take FC-PGAs. Has a block of jumpers to set voltage. Slocket 'b' is an FC-PGA compatible slocket without voltage jumpers

Ram:
1 stick of PC100 256mb
1 stick of PC133 256mb
1 stick of PC 133 128mb ram though it crashes the machine if the FSB speed is at 133MHz so pretty sure this is a bad stick

I can get the 450MHz P3 to do:
450MHz at 100Mhz FSB, it works fine with a 112MHz FSB and, with the L2 cache off, the machine run at 133MHz FSB. Memtest completes with the PC133 256mb ram stick without errors (and at 122 and 100). The PC133 128mb stick crashes the machine instantly (and no POST) with a 133 MHz fsb

If I use the slocket a with any of the ram sticks (only using sticks individually)
The celeron is stable up to 112MHz FSB. Instant crash if try 133MHz. The celeron is unstable if the FSB is set to 100MHz (crashes on 2nd post screen) or 103MHz
If the jumper on the slocket is set to "overclock" rather than "auto", the bios then offers up 66MHz, 75MHz and 83MHz FSBs -- it is stable on all of these.

The pentium 3 933 will not POST if the slocket jumper is set to "overclock" but if the bios is set to 83MHz and then it is put into the slocket, it's stable.
It isnt stable at 133MHz in that there are glitches in games. at 112MHz I have got it to run games but it is is prone to crashes.
memtest always fails in bulk transfer tests - instantly at 112 MHz when it reaches the bulk transfer test and at the end it is all errors at 133MHz of the bulk transfer test.
I did manage to get it through 3dbench2001 before it crashed and at 112 and 133Fsb it significantly outperformed the celeron (which is why I'd like to get it working)

If I use slocket b:
With the PC133 256mb ram:
The 933 is stable with 100MHz fsb but frequently fails to get beyond POST at 112Mhz and doesnt post at 133MHz
The PC seems to undervolt - it only feeds about 1.68v (same is true if I use the 'auto' voltage setting on slocket a). I therefore modified the slocket so that VDD1 (I think it was) is held to VCC - now the core voltage is 1.8v
However, it is still just as unstable.
The celeron is unstable at anything 100Mhz or above.
Slocket B has a 2 jumpers only - 1 that is closed witha solid link but can be opened for cyrix cpus, then another that is for "cyrix 100MHz fsb" which I have tried open and closed but makes no difference.
There were also some unpopulated pads marked "CE1" and "CE2" that looked like they were for electrolytic caps that werent populated, so in case they added stability, I added a 10uF capacitor to each but that made no difference either

I am using a farily large akasa cooler for the slockets CPUs so I dont think they are overheating

Ram timings are all set as loose as possible (to "3" and with spectral [thing] set to "off")

Thoughts:
I am pretty sure the PC133 256mb stick is good as it passes memtest etc fine with the P450 at every speed.
The 9250 AGP card and PCI cards seem fine with the overclocks given the P450 is stable at all speeds (with L2 cache off, with it on, it just crashes instantly at 133MHz)
The motherboard seems good given that the P450 works unless it is that the power delivery is marginal and the extra power needed for the P3 933 and celeron tips it over the edge? There are no obviously bulging caps on the motherboard. Slocket A has a couple of electrolytic caps, B has none except the new ones I added.

I have spent ages trying to work out why I cant get the pentium 3 running stably at 112 or 133 MHz fsb and why memtest keeps failing at the block transfer stage.

Can anyone shed any light on this!
Thanks

Last edited by squelch41 on 2021-11-03, 22:56. Edited 1 time in total.

V4P895P3 VLB Motherboard and AMD 486 133MHz CPU
64mb RAM, CF 4Gb HDD,
Realtek 8019 ethernet + XT-IDE bios ROM, ES1869 soundcard, VLB multi-io card Cirrus Logic GD5428 1mb VGA

440bx MSI 6119, modified slocket , celeron 1.1GHz, 256mb SD-RAM, CF 4GB HDD

Reply 1 of 48, by BitWrangler

User metadata
Rank l33t
Rank
l33t

So, tried different graphics cards? 112 is getting the PCI up over 37mhz and the AGP over 74, something might not be liking this, and thus you don't get to 133.

Unicorn herding operations are proceeding, but all the totes of hens teeth and barrels of rocking horse poop give them plenty of hiding spots.

Reply 2 of 48, by squelch41

User metadata
Rank Member
Rank
Member
BitWrangler wrote on 2021-11-03, 22:26:

So, tried different graphics cards? 112 is getting the PCI up over 37mhz and the AGP over 74, something might not be liking this, and thus you don't get to 133.

I hadnt - largely because the graphics seemed stable and with the P450 socket 1 CPU running at 133, there were no issues
I should add that I have run with no PCI cards, only the AGP graphics card but again, that made no difference

V4P895P3 VLB Motherboard and AMD 486 133MHz CPU
64mb RAM, CF 4Gb HDD,
Realtek 8019 ethernet + XT-IDE bios ROM, ES1869 soundcard, VLB multi-io card Cirrus Logic GD5428 1mb VGA

440bx MSI 6119, modified slocket , celeron 1.1GHz, 256mb SD-RAM, CF 4GB HDD

Reply 3 of 48, by BitWrangler

User metadata
Rank l33t
Rank
l33t

I see, sorry I only keep 5 or 6 things in my head at once reading through everything. Now the celeron, 1Ghz versions that ran on 1.7v were usually 66Mhz bus CPU, and those didn't clock more than one or two speed grades. Then the Katmai overclocked should be using more power than the 933 at stock speeds, however, earlier 933s had the same speed bug as the first 1G PIIIs so won't behave set much faster, but should work on 133. Slotkets are flaky when they lean askew, is it flopping around in the socket or is the retention mechanism holding it tight and straight?

Unicorn herding operations are proceeding, but all the totes of hens teeth and barrels of rocking horse poop give them plenty of hiding spots.

Reply 4 of 48, by squelch41

User metadata
Rank Member
Rank
Member
BitWrangler wrote on 2021-11-03, 22:39:

I see, sorry I only keep 5 or 6 things in my head at once reading through everything. Now the celeron, 1Ghz versions that ran on 1.7v were usually 66Mhz bus CPU, and those didn't clock more than one or two speed grades. Then the Katmai overclocked should be using more power than the 933 at stock speeds, however, earlier 933s had the same speed bug as the first 1G PIIIs so won't behave set much faster, but should work on 133. Slotkets are flaky when they lean askew, is it flopping around in the socket or is the retention mechanism holding it tight and straight?

Not at all! I appreciate the help 😀
Just tried the only other AGP card I have which is a rage 128 - used with slocket a and the P933 -- behaviour is the same as with the 9250

The celeron is a SL5XU - sorry I said it was 1GHz (I was away from my kit at the time!), it's a 1.1Ghz 1100/128/100 1.75v part
The 933 is a SL5U3

The slocket is just a loose PCB so isnt held. I'll see if I can stablize it a bit.
I've cleaned the slocket's edge connectors with IPA but that didnt help.

V4P895P3 VLB Motherboard and AMD 486 133MHz CPU
64mb RAM, CF 4Gb HDD,
Realtek 8019 ethernet + XT-IDE bios ROM, ES1869 soundcard, VLB multi-io card Cirrus Logic GD5428 1mb VGA

440bx MSI 6119, modified slocket , celeron 1.1GHz, 256mb SD-RAM, CF 4GB HDD

Reply 5 of 48, by BitWrangler

User metadata
Rank l33t
Rank
l33t

Some ideas here.. Slotket (slot1 > s370): In need of finding retention brackets/stilts for slotket

Ah right, 1.1 celly coppermine, 100mhz stock, still a bit near the raggedy edge for what coppermine can clock to though, so don't expect to see much out of it.

Unicorn herding operations are proceeding, but all the totes of hens teeth and barrels of rocking horse poop give them plenty of hiding spots.

Reply 6 of 48, by squelch41

User metadata
Rank Member
Rank
Member
BitWrangler wrote on 2021-11-03, 23:02:

Some ideas here.. Slotket (slot1 > s370): In need of finding retention brackets/stilts for slotket

Ah right, 1.1 celly coppermine, 100mhz stock, still a bit near the raggedy edge for what coppermine can clock to though, so don't expect to see much out of it.

Yeah, the celeron crashing out at 133MHz is what I'd expect 😀 that is certainly the CPU just hitting it's limit. It does seem odd that it is unstable at the stock 100MHz speed, but given that it's fine at 112Mhz, it doesnt really matter.
It's the P933 that is annoying me!
It's running fine at 112MHz at the moment in the slocket A (just run3dmark2000 and 2001SE ok) but just instability at 133 and get a load of ram errors if I run memtest in win98

V4P895P3 VLB Motherboard and AMD 486 133MHz CPU
64mb RAM, CF 4Gb HDD,
Realtek 8019 ethernet + XT-IDE bios ROM, ES1869 soundcard, VLB multi-io card Cirrus Logic GD5428 1mb VGA

440bx MSI 6119, modified slocket , celeron 1.1GHz, 256mb SD-RAM, CF 4GB HDD

Reply 7 of 48, by PARKE

User metadata
Rank Oldbie
Rank
Oldbie

Can you post the number that is on the clock generator at the left side of the AGP slot ?

EDIT
Your clock generator is likely a Cypress W48S111-14 and if that is the case then the PCI is running at 44.4 MHz with CPU @ 133MHz. This may explain some of the symptoms in your report. See page 6 in the attached spec sheet.

Filename
W48S111-14.pdf
File size
140.26 KiB
Downloads
10 downloads
File license
Fair use/fair dealing exception
Last edited by PARKE on 2021-11-06, 15:21. Edited 1 time in total.

Reply 9 of 48, by squelch41

User metadata
Rank Member
Rank
Member
PARKE wrote on 2021-11-06, 13:33:

Can you post the number that is on the clock generator at the left side of the AGP slot ?

EDIT
Your clock generator is likely a Cypress W48S111-14 and if that is the case then the PCI is running at 44.4 MHz with CPU @ 133MHz. This may explain some of the symptoms in your report. See page 6 in the attached spec sheet.W48S111-14.pdf

Yes, that's the model of chip just next to the AGP slot on the left.
However, the fact that it is still unstable without PCI cards present makes that seem unlikely? And also, I get RAM errors on memory testing - would that fit with an unstable pci bus?
Also, it looks like at 85MHz FSB the PCI is clocked at not far off the same as it is at 133MHz (41MHz vs 44MHz) and the system is stable at that speed

(these are genuine questions! am no way an expert in any of this!)

Last edited by squelch41 on 2021-11-06, 16:24. Edited 1 time in total.

V4P895P3 VLB Motherboard and AMD 486 133MHz CPU
64mb RAM, CF 4Gb HDD,
Realtek 8019 ethernet + XT-IDE bios ROM, ES1869 soundcard, VLB multi-io card Cirrus Logic GD5428 1mb VGA

440bx MSI 6119, modified slocket , celeron 1.1GHz, 256mb SD-RAM, CF 4GB HDD

Reply 10 of 48, by squelch41

User metadata
Rank Member
Rank
Member
SSTV2 wrote on 2021-11-06, 15:18:

Post high ress pictures of bare slotkets from both sides, electric FC-PGA CPU design flaws should be apparent.

Images attached, if they arent clear enough let me know and I'll try again.

Slocket A has the PGA->FC-PGA wire bodge
Slocket B has the voltage pin bodge and I added the capacitors to the unpopulated CE1 and CE2 pags - they are 10uF capacitors. They made no difference to stability

Attachments

  • Sloket b bottom.jpg
    Filename
    Sloket b bottom.jpg
    File size
    1.17 MiB
    Views
    760 views
    File comment
    Slocket A (NOT B) bottom
    File license
    Public domain
  • Sloket A top.jpg
    Filename
    Sloket A top.jpg
    File size
    1.81 MiB
    Views
    760 views
    File comment
    Slocket A top
    File license
    Public domain
  • Slockt B bottom.jpg
    Filename
    Slockt B bottom.jpg
    File size
    1.87 MiB
    Views
    760 views
    File comment
    Slocket B bottom
    File license
    Public domain
  • Slocket b top.jpg
    Filename
    Slocket b top.jpg
    File size
    1.19 MiB
    Views
    760 views
    File comment
    Slocket B top
    File license
    Public domain
Last edited by squelch41 on 2021-11-07, 23:20. Edited 1 time in total.

V4P895P3 VLB Motherboard and AMD 486 133MHz CPU
64mb RAM, CF 4Gb HDD,
Realtek 8019 ethernet + XT-IDE bios ROM, ES1869 soundcard, VLB multi-io card Cirrus Logic GD5428 1mb VGA

440bx MSI 6119, modified slocket , celeron 1.1GHz, 256mb SD-RAM, CF 4GB HDD

Reply 11 of 48, by PARKE

User metadata
Rank Oldbie
Rank
Oldbie
squelch41 wrote on 2021-11-06, 16:05:

Yes, that's the model of chip just next to the AGP slot on the left.
However, the fact that it is still unstable without PCI cards present makes that seem unlikely? And also, I get RAM errors on memory testing - would that fit with an unstable pci bus?

I don't know.
There is a similar chip in the upper right corner next to a jumper which is refered to as "DIMM clock buffer" in the manual. Maybe this chip acts in the same way to regulate RAM frequency and causes similar problems. But that is just guesswork on my behalf.
One point that remains is that this board was meant for Pentium II and early Pentium III. Maybe you are asking too much of the old girl.

Reply 12 of 48, by PARKE

User metadata
Rank Oldbie
Rank
Oldbie
squelch41 wrote on 2021-11-06, 16:15:

Slocket A has the PGA->FC-PGA wire bodge
Slocket B has the voltage pin bodge and I added the capacitors to the unpopulated CE1 and CE2 pags - they are 10uF capacitors. They made no difference to stability

A is supposedly a Fastfame product and B is a PCchips. Not the most desirable products for many.

Reply 13 of 48, by squelch41

User metadata
Rank Member
Rank
Member
PARKE wrote on 2021-11-06, 16:25:
squelch41 wrote on 2021-11-06, 16:05:

Yes, that's the model of chip just next to the AGP slot on the left.
However, the fact that it is still unstable without PCI cards present makes that seem unlikely? And also, I get RAM errors on memory testing - would that fit with an unstable pci bus?

One point that remains is that this board was meant for Pentium II and early Pentium III. Maybe you are asking too much of the old girl.

Oh, I definately am, that's the fun 😀
Just want to see if there are any avenues to push it to 133MHz -- it's just the challenge I enjoy with these old machines

V4P895P3 VLB Motherboard and AMD 486 133MHz CPU
64mb RAM, CF 4Gb HDD,
Realtek 8019 ethernet + XT-IDE bios ROM, ES1869 soundcard, VLB multi-io card Cirrus Logic GD5428 1mb VGA

440bx MSI 6119, modified slocket , celeron 1.1GHz, 256mb SD-RAM, CF 4GB HDD

Reply 14 of 48, by PARKE

User metadata
Rank Oldbie
Rank
Oldbie
squelch41 wrote on 2021-11-06, 16:32:

Oh, I definately am, that's the fun 😀
Just want to see if there are any avenues to push it to 133MHz -- it's just the challenge I enjoy with these old machines

What is the number of the chip in the upper right corner ?

Reply 15 of 48, by squelch41

User metadata
Rank Member
Rank
Member
PARKE wrote on 2021-11-06, 16:39:
squelch41 wrote on 2021-11-06, 16:32:

Oh, I definately am, that's the fun 😀
Just want to see if there are any avenues to push it to 133MHz -- it's just the challenge I enjoy with these old machines

What is the number of the chip in the upper right corner ?

It's W40S11-23G on the top line, then B311 1910TD
Looks like it accepts 133MHz input
"Input Frequency:............................................... 0 to 133 MHz" on the datasheet https://www.digchip.com/datasheets/parts/data … S11-23G-pdf.php

Last edited by squelch41 on 2021-11-06, 16:52. Edited 1 time in total.

V4P895P3 VLB Motherboard and AMD 486 133MHz CPU
64mb RAM, CF 4Gb HDD,
Realtek 8019 ethernet + XT-IDE bios ROM, ES1869 soundcard, VLB multi-io card Cirrus Logic GD5428 1mb VGA

440bx MSI 6119, modified slocket , celeron 1.1GHz, 256mb SD-RAM, CF 4GB HDD

Reply 16 of 48, by squelch41

User metadata
Rank Member
Rank
Member

One of the other angles of attack I was looking at was with SoftFSB as this hints (as does that data sheet) that there is a way of setting the PCI clock divider to CPU/4
However, I cant work out the right settings in SoftFSB to do that -- does anyone have any ideas on that?

Attachments

  • SoftFSB.jpg
    Filename
    SoftFSB.jpg
    File size
    69.54 KiB
    Views
    716 views
    File license
    Public domain

V4P895P3 VLB Motherboard and AMD 486 133MHz CPU
64mb RAM, CF 4Gb HDD,
Realtek 8019 ethernet + XT-IDE bios ROM, ES1869 soundcard, VLB multi-io card Cirrus Logic GD5428 1mb VGA

440bx MSI 6119, modified slocket , celeron 1.1GHz, 256mb SD-RAM, CF 4GB HDD

Reply 17 of 48, by PARKE

User metadata
Rank Oldbie
Rank
Oldbie
squelch41 wrote on 2021-11-06, 16:48:

It's W40S11-23G on the top line, then B311 1910TD
Looks like it accepts 133MHz input
"Input Frequency:............................................... 0 to 133 MHz" on the datasheet https://www.digchip.com/datasheets/parts/data … S11-23G-pdf.php

Yes, but it may be dependant on other signals to deliver the right output.
There is a photo of a later version of this board on-line and there is a jumper installed on the jumper block next to the W40S11 plus there is a switch block added at the bottom of the RAM slots next to the chip that is positioned there.
My take is that your board is not really, or really not, up for the task that you are asking of it an it is doubtful that you can overcome that via software solutions. But, I may well be wrong and when being perplexed by the behaviour of hardware is fun for you then you are on the right track 😀.

Reply 18 of 48, by SSTV2

User metadata
Rank Oldbie
Rank
Oldbie

Slotket A is not fit for the FC-PGA CPU support modification, it's lacks all of the V_CMOS pull up resistors for the CMOS signal group (A20M#, FLUSH#, IGNNE#, INIT#, LINT0/INTR, LINT1/NMI, PREQ#, SLP#, SMI#, STPCLK# etc.) and most likely doesn't even have a trace which connects all of the VTT pins. Don't bother with it.

Slotket B looks more like it, but for some reason it uses 75 Ohm VTT pull up resistors instead of recommended 56 Ohm to terminate host signals from CPU, plus few resistor packs weren't even soldered. You can see that there are at least 8 resistors in the socket's cavity, which should be pulling up at least that many CMOS signals to V_CMOS, but there should be more resistors than that.

To do list for Slotket B:

*check whether all VTT pins are connected together, if not connect at least few of them to V_1.5, in order to improve current flow;
*check whether G35 and G37 pins are connected, if not, short them. If 1st requirement in the list isn't met, connect these pins to V_1.5 also;
*figure out which signals the missing resistor packs were supposed to pull up and add them if necessary;
*check whether V_CMOS (AB36) is shorted to V_1.5 (AD36) or V_2.5 (Z36) on the slotket, if it's shorted to V_2.5, remove the short (unlikely case for B, but A might have this flaw);
*make sure that V_1.5 voltage is at the right level;
*map out all of the CMOS signals that are left floating and pull them up to V_CMOS (consult intel design guides);
*improve current flow for the CPU core voltage in order to rectify undervolts. Bridge some of the Vcc and GND slotket cartrige contacts directly to the CPU socket;
*make sure that CPU uCode is up to date.

Properly modified slotket might still not guarantee you stable operation @ 133MHz FSB with PIII, due to possible MB design flaws.

Reply 19 of 48, by squelch41

User metadata
Rank Member
Rank
Member
SSTV2 wrote on 2021-11-06, 18:35:
Slotket A is not fit for the FC-PGA CPU support modification, it's lacks all of the V_CMOS pull up resistors for the CMOS signal […]
Show full quote

Slotket A is not fit for the FC-PGA CPU support modification, it's lacks all of the V_CMOS pull up resistors for the CMOS signal group (A20M#, FLUSH#, IGNNE#, INIT#, LINT0/INTR, LINT1/NMI, PREQ#, SLP#, SMI#, STPCLK# etc.) and most likely doesn't even have a trace which connects all of the VTT pins. Don't bother with it.

Slotket B looks more like it, but for some reason it uses 75 Ohm VTT pull up resistors instead of recommended 56 Ohm to terminate host signals from CPU, plus few resistor packs weren't even soldered. You can see that there are at least 8 resistors in the socket's cavity, which should be pulling up at least that many CMOS signals to V_CMOS, but there should be more resistors than that.

To do list for Slotket B:

*check whether all VTT pins are connected together, if not connect at least few of them to V_1.5, in order to improve current flow;
*check whether G35 and G37 pins are connected, if not, short them. If 1st requirement in the list isn't met, connect these pins to V_1.5 also;
*figure out which signals the missing resistor packs were supposed to pull up and add them if necessary;
*check whether V_CMOS (AB36) is shorted to V_1.5 (AD36) or V_2.5 (Z36) on the slotket, if it's shorted to V_2.5, remove the short (unlikely case for B, but A might have this flaw);
*make sure that V_1.5 voltage is at the right level;
*map out all of the CMOS signals that are left floating and pull them up to V_CMOS (consult intel design guides);
*improve current flow for the CPU core voltage in order to rectify undervolts. Bridge some of the Vcc and GND slotket cartrige contacts directly to the CPU socket;
*make sure that CPU uCode is up to date.

Properly modified slotket might still not guarantee you stable operation @ 133MHz FSB with PIII, due to possible MB design flaws.

Wow, thanks!
Wierldy, at least at 112Mhz, slocket A is the better behaved.
I shall get probing!

V4P895P3 VLB Motherboard and AMD 486 133MHz CPU
64mb RAM, CF 4Gb HDD,
Realtek 8019 ethernet + XT-IDE bios ROM, ES1869 soundcard, VLB multi-io card Cirrus Logic GD5428 1mb VGA

440bx MSI 6119, modified slocket , celeron 1.1GHz, 256mb SD-RAM, CF 4GB HDD