Intel D945GTP VRM Cooling mod

[c_hegge]

While I have generally had pretty decent luck with C2D and newer Intel boards, some of the P4 era boards, like my D945GTP, had problems with hot-running VRMs - especially when using 125W Pentium D CPUs, like the one in this board. Standing the FETs up certanly doesn't help with this, but - it does make it kind of easy to fix.

I found an old PSU heatsink in my stash and broke off some of the fingers. I then used some thermal epoxy to stick them to the FETs. I then used some RAM heatsinks (which came with a GPU cooler, but didn't get used) on the coils - which also get quite toasty on this board. The next step is to do something about the 12V side FETs. Those are soldered to the board, but also get hot. There are blank spaces next to them all (which are in parallel). I'm thinking that, If I can find some identical ones, I might solder a few more in parallel (and maybe stick some RAM heatsinks to them as well)

[Majestyk]

Re: Intel D945GTP VRM Cooling mod

One of the reasons why the additional cooling is necessary might be the absence of the original Intel "box-CPU-cooler". The original Intel coolers produce an airstream at their bottom reaching - and cooling - all the components around the CPU-socket while the passive cooler on the board does nothing like that.

So the thermal design of the board relies on the Intel cooler-specifications. If the Intel-specified cooler is missing, the whole thermal management is at stake.

[c_hegge]

Re: Intel D945GTP VRM Cooling mod

You'd think so, but it doesn't help that much. I've seen similar boards with cooked VRMs even when the stock cooler is used. Those Pentium D CPUs get so hot with the stock cooler that the air coming off it doesn't actually cool the VRM that much.

[Majestyk]

Re: Intel D945GTP VRM Cooling mod

The tremendous heat loss of those CPU is obviously the culprit.

If this is 100W and more the fan must "cool" the VRMs with air that is 50 - 60° C hot already. So either the fan would have to spin at 10000 rpm or the vrms get boiled.

[stj]

Re: Intel D945GTP VRM Cooling mod

but the problem is the bios by default runs the fan real slow.
i'v got a few and you need to change the bios settings to fan-full-on all the time.

[Majestyk]

Re: Intel D945GTP VRM Cooling mod

That´s surprising since Intel are always aiming at the enterprise / workstation market where (long time) stability is more important than silent computing.
I would have expected more conservative defaults here.

They provided a rocksolid cooling for the MCH and obviously neglected the VRMs.

And indeed - the way they soldered the MOSFETs standing up with one leg bent 90° is quite unconventional and not state of the art.

[mockingbird]

Re: Intel D945GTP VRM Cooling mod

Those Fujitsu FPCAP come in both 2,000 hour and 5,000 hour versions. I still haven't found any method of differentiating between the two merely by the markings on them.

[c_hegge]

Re: Intel D945GTP VRM Cooling mod

The tremendous heat loss of those CPU is obviously the culprit.

If this is 100W and more the fan must "cool" the VRMs with air that is 50 - 60° C hot already. So either the fan would have to spin at 10000 rpm or the vrms get boiled.

but the problem is the bios by default runs the fan real slow.
i'v got a few and you need to change the bios settings to fan-full-on all the time.

That does all contribute, but the root cause of the problem lies with the VRM itself IMO. Most other motherboards have no problems with the VRM cooking itself - regardless of the CPU used, the cooler or the fan settings.

[Majestyk]

Re: Intel D945GTP VRM Cooling mod

...the root cause of the problem lies with the VRM itself IMO. Most other motherboards have no problems with the VRM cooking itself - regardless of the CPU used, the cooler or the fan settings.

Everyone else has their MOSFETs soldered flat to the PCB and allows efficient cooling by the copper-layers.

When unsoldering the MOSFETs or even capacitors from larger copper-layers we get an idea about how efficient this cooling can be.

Intel has chosen a different - inferior - way here and solderd the MOSFETs upright without any cooling except for the airflow.
I would consider this a thermal-design fault.

[Th3_uN1Qu3]

Re: Intel D945GTP VRM Cooling mod

I've seen many other boards done like that - Asus and Biostar are two examples that come to mind. I agree that it is poor thermal design. But hey, they're using MOSFETs rated for operation at 175C, so it should work fine, shouldn't it? At least it made it past warranty.

[Pentium4]

Re: Intel D945GTP VRM Cooling mod

Awesome mod, c_hegge. These boards seem to last a long time with high TDP CPU's as it is. So even without adding additional FETs, those heatsinks should make a significant difference.

[c_hegge]

Re: Intel D945GTP VRM Cooling mod

The blank spaces I was referring to are on the VRM high side (see the picture). It was the low side that got the heatsinks.

I ended up scavenging some FETs from another dead board and replacing them, as well as adding the extra ones. I forget which parts I used but they have a slightly lower RDS-on than the originals, and they seem to run cooler.

[c_hegge]

Re: Intel D945GTP VRM Cooling mod

And as a final mod, that lone 1200uF Nichicon HD on the VRM high side got replaced by three 470uF 16V OSCON SEPC polies. Pics coming soon...

[TELVM]

Re: Intel D945GTP VRM Cooling mod

While you're at it I'd also sink the drivers, they get even hotter than the FETs :