That is a standard size for a fan. I'd replace it with a high airflow fan, though that might make it sound 'jet enginey'.

Something like a noctua.

https://www.amazon.com/Noctua-NF-A8-PWM-Premi … /dp/B00NEMG62M/

(It does not have to be noctua, that's just a well known brand of high-airflow 80mm fan. It can be any high-airflow 80mm fan. They will all drop-in replace that failing one.)

edit, 'how to remove'
That looks like a 'press fit' fitting. I would make/use a jig that grabs the frame of the fan, and has 4 metal pins that press against the plastic lugs, with a 'screw' that applies force, to slowly pull the fan off the pressfit lugs safely.

Think 'reverse C-Clamp', or 'screw-jack'.

^This

Same ones are used on my dual IBM IntelliStation M Pro - the fans are held on by the 4 chamfered corner pins. Just ease the fan off slowly with a thin bladed screwdriver, going round each corner alternatively.

PC Hoarder Patrol wrote on 2026-11-13, 04:52:

^This

Same ones are used on my dual IBM IntelliStation M Pro - the fans are held on by the 4 chamfered corner pins. Just ease the fan off slowly with a thin bladed screwdriver, going round each corner alternatively.

That's exactly the one, thank you!

Any thoughts on the fin assembly falling away so easily when the plastic clip is removed? I figure at the least the thermal paste needs replacing - but was that actually meant to be adhesive? Or is normal paste enough? I was wondering about putting a little solder into where the two ends of the fins meet, to hold it together, and in place, permanently. Overkill?

Is it also worth removing the heatsink altogether to re-paste the CPU? I've never tried that with Slot1 before ...

justin1985 wrote on 2026-11-13, 13:58:

That's exactly the one, thank you! […]
Show full quote

PC Hoarder Patrol wrote on 2026-11-13, 04:52:

^This

Same ones are used on my dual IBM IntelliStation M Pro - the fans are held on by the 4 chamfered corner pins. Just ease the fan off slowly with a thin bladed screwdriver, going round each corner alternatively.

That's exactly the one, thank you!

Any thoughts on the fin assembly falling away so easily when the plastic clip is removed? I figure at the least the thermal paste needs replacing - but was that actually meant to be adhesive? Or is normal paste enough? I was wondering about putting a little solder into where the two ends of the fins meet, to hold it together, and in place, permanently. Overkill?

Is it also worth removing the heatsink altogether to re-paste the CPU? I've never tried that with Slot1 before ...

Never had one fully disassembled so can't say re. the fin > heatsink core thermal paste type, though would have thought regular rather than adhesive was more likely (don't overdo any reapplication). Not sure I'd bother with solder - your choice. Far as re-pasting the cpu core goes, can't hurt after 25 years - the rear spring clip is easy enough to remove with sufficient sideways pressure, though be careful the heatsink hasn't partially bonded to the cpu core due to repeated heat cycles (may need gentle twisting / rocking to break any bond if it has - probably best done when the assembly is already warm).

FYI, these heatsinks (slot 1 / skt 370) were made by Silent Systems, a division of Molex they purchased in the late 90's. Not sure they were available outwith the larger OEM buyers.

That's a pretty unique little system you have there. Fujitsu model names can be pretty opaque, but the motherboard is definitely a CPO32130, and I'm pulling up blanks on that one.

Nothing like it on TheRetroWeb

Looks like a very, very good loadout for its small size, almost regardless of which video chip is under that heatsink. The video memory chips are 16Mb SDRAM with 1Mx16 layout. That means it looks like 64MB 64Mb = 8MB with a 64b wide bus. That's pretty impressive for a 2000-era SFF system; I'd suspect it's a GeForce2MX given the combo of large size but 'only' 64b bus.

As for the i820 chipset, it was all the talk of 1999 and 2000. Commercially it was a disaster, due in part to the excessively high price of RDRAM - but just as much due to Intel making some extremely poor (marketing-driven) technical decisions around its release and promised features; the final i820 with 2 RIMM slots like you have here is rock solid and performs very well (on-par with i440BX if paired with PC800 RIMMs, but with AGP 4x support and ATA-66 or maybe 100 depending on the ICH (=southbridge)), but earlier attemts were a disaster. Read up (ideally on Tom's Hardware Guide, back when it was one of the leading tech journalism sites) on Caminogate in 1999 about the 3rd RIMM slot leading to a recall of all motherboards, and the MTH in 2000 - trying to make i820 work with SDRAM. It was slow and unstable.

That 'blanking card' you referred to is a C-RIMM (C stands for continuity), and it's required if only one RIMM is in use. If you happen to come across more i820 stuff, don't confuse it with the CT-RIMM (T stands for termination), which was required in the third RIMM slots of those early boards supplied with them, as well as a C-RIMM in either of the other two slots if only one RIMM were in use. The CT-RIMM has a load of SMD resistors on it, the C-RIMM just has some traces.

PC Hoarder Patrol wrote on Yesterday, 08:58:

Never had one fully disassembled so can't say re. the fin > heatsink core thermal paste type, though would have thought regular rather than adhesive was more likely (don't overdo any reapplication). Not sure I'd bother with solder - your choice. Far as re-pasting the cpu core goes, can't hurt after 25 years - the rear spring clip is easy enough to remove with sufficient sideways pressure, though be careful the heatsink hasn't partially bonded to the cpu core due to repeated heat cycles (may need gentle twisting / rocking to break any bond if it has - probably best done when the assembly is already warm).

FYI, these heatsinks (slot 1 / skt 370) were made by Silent Systems, a division of Molex they purchased in the late 90's. Not sure they were available outwith the larger OEM buyers.

Thanks so much for this. In the end it was relatively painless. Easing with a thin plastic spudger one corner at a time (it was too tight to get a screwdriver in) did the job getting it off, although I had to actually drill out the holes in the replacement 50mm fan (a random Aliexpress I had in my spares box, I think) for the clips to go in. I used a fair amount of basic electronics heatsink compound for the fins, and held them together with thin masking tape when clipping the fan clip back over.

Getting to the CPU die was much less painless than I'd feared too, and the original thermal compound was crusty as anything, so glad I replaced it (with a decent thermaltake compound).

While I was at it, I also replaced the 60mm fan in the PSU (with a quite likely fake Delta), and cleaned out and lubricated the floppy drive, which got it working 😀

dionb wrote on Yesterday, 13:18:

That's a pretty unique little system you have there. Fujitsu model names can be pretty opaque, but the motherboard is definitely […]
Show full quote

That's a pretty unique little system you have there. Fujitsu model names can be pretty opaque, but the motherboard is definitely a CPO32130, and I'm pulling up blanks on that one.

Nothing like it on TheRetroWeb

Looks like a very, very good loadout for its small size, almost regardless of which video chip is under that heatsink. The video memory chips are 16Mb SDRAM with 1Mx16 layout. That means it looks like 64MB with a 64b wide bus. That's pretty impressive for a 2000-era SFF system; I'd suspect it's a GeForce2MX given the combo of large size but 'only' 64b bus.

As for the i820 chipset, it was all the talk of 1999 and 2000. Commercially it was a disaster, due in part to the excessively high price of RDRAM - but just as much due to Intel making some extremely poor (marketing-driven) technical decisions around its release and promised features; the final i820 with 2 RIMM slots like you have here is rock solid and performs very well (on-par with i440BX if paired with PC800 RIMMs, but with AGP 4x support and ATA-66 or maybe 100 depending on the ICH (=southbridge)), but earlier attemts were a disaster. Read up (ideally on Tom's Hardware Guide, back when it was one of the leading tech journalism sites) on Caminogate in 1999 about the 3rd RIMM slot leading to a recall of all motherboards, and the MTH in 2000 - trying to make i820 work with SDRAM. It was slow and unstable.

That 'blanking card' you referred to is a C-RIMM (C stands for continuity), and it's required if only one RIMM is in use. If you happen to come across more i820 stuff, don't confuse it with the CT-RIMM (T stands for termination), which was required in the third RIMM slots of those early boards supplied with them, as well as a C-RIMM in either of the other two slots if only one RIMM were in use. The CT-RIMM has a load of SMD resistors on it, the C-RIMM just has some traces.

Thanks - yes it is super interesting isn't it? And thank you for the info on the i820 - intriguing!

Alas I still seem to be getting odd behaviour with the RAM - from a cold boot, it always seems to give the RAM error beep code. Re-seating the RIMM (+ C-RIMM) in the same slots, gently cleaning their contacts, and squirting with contact cleaner doesn't seem to make any difference. Only swapping them between slots seems to get it to boot. Once off for a while, turning it back on, I have to do the same again. Very weird!

Considering how cheap RAMBUS RAM now seems to be, buying a replacement matched pair of RIMMs seems like a good next step. Anyone have any other ideas?

dionb wrote on Yesterday, 13:18:

Looks like a very, very good loadout for its small size, almost regardless of which video chip is under that heatsink. The video memory chips are 16Mb SDRAM with 1Mx16 layout. That means it looks like 64MB with a 64b wide bus. That's pretty impressive for a 2000-era SFF system; I'd suspect it's a GeForce2MX given the combo of large size but 'only' 64b bus.

I finally identified the GPU once I'd swapped the original hard drive with the locked down copy of XP, which was full of bad sectors when I tried to format it, for an SD card adapter (interesting it recognised and let me format a 32Gb card, but wouldn't boot from it, but a 16Gb card was fine).

Alas it isn't anything as exotic as a GeForce 2MX with 64Mb - it's an ATI Rage 128 8Mb, running at AGPx4 - at least according to NSSI.

Besides i820 there was the workstation/server targeted i840 as well, sporting RDRAM in dual channel and support for dual P3s ... that was the oomph!

justin1985 wrote on Yesterday, 20:09:

[...]

Considering how cheap RAMBUS RAM now seems to be, buying a replacement matched pair of RIMMs seems like a good next step. Anyone have any other ideas?

No need for matched pairs, the i820 has a single-channel controller, so any two RIMMs will do. Do make sure they're the same speed & latency though. You definitely want PC800, preferably 40ns, but 45ns is also acceptable.

As for your memory problems, they could also be power related. This is typically what you might expect if some of the capacitors on the line supplying 2.5V to the memory subsystem were failing. Looking at your photograph I don't see any obviously bad caps, but it's worth checking and if you have the equipment (an oscilloscope would be useful) check if the Vrimm is in-spec and stable, particularly while booting.

justin1985 wrote on Yesterday, 22:33:

[...]

Alas it isn't anything as exotic as a GeForce 2MX with 64Mb - it's an ATI Rage 128 8Mb, running at AGPx4 - at least according to NSSI.

Bugger, I mixed up my bits and bytes (just as you're doing with your letters: capitalization matters): it's 64Mb, not 64MB, and 64Mb=8MB. Which rules out the Gf2MX or similar - so then a Rage128 isn't so surprising. Shame, as that reduces its utility as a DOS system - the Rage chips have poor VESA support.