Question Heatsink upgrade for Gemini Lake mini-ITX board?

bit_user

Polypheme
Ambassador
I recently picked up an ASRock J5040-ITX mini-ITX board with a Pentium Silver J5040 Gemini Lake Refresh SoC on it. While testing it on an open bench (I literally have the motherboard sitting on a table, with 1/2 inch feet under it), I was pretty shocked to see it rather easily hit 84 C (ambient was probably 24 C; no perceivable air movement). I had planned to put this in a fanless case, but that's not going to happen without some significant throttling.

My first thought was to pull off the heatsink and replace the thermal compound with something higher-performance than I'll bet they used on such a budget board. However, I thought "why stop there?" and I'm wondering if there might be other heatsinks I can put on it. The attachment points of the current heatsink are approximately 44.5 mm x 56 mm (2.7 mm diameter). Is this any kind of standard, or just specific to what ASRock came up with?

My other thought is to just buy a copper heatsink that's the biggest I can find (within 87 mm x 51.5 mm) and let gravity hold it in place, since the case is horizontal.

Update: I pulled off the heatsink and found a second set of holes. There's both the outline for the original heatsink (87.5 mm x 55 mm) and an outline for the second heatsink (45.5 mm x 44.5 mm). In this case, there are only 2 holes at a diagonal arrangement. Rather than being at the corners, the mounting holes are in 10 mm from each corner and set about 4 mm outside of the footprint.

Basically, the second heatsink outline looks a lot like what you'd see mounted on motherboard chipset.
 
Last edited:

Lutfij

Titan
Moderator
I've built an itx system in an Antec ISK110 with an Asus J3355I-C and by design the processor doesn't need an active cooling solution. In fact if you're going to cool the board, you could do with a couple of Noctua 40mm fans blowing across the existing heatsink but it's not necessary.

Further reading. There's also a reddit on that board of yours.

"why stop there?"
I've been there, managed to claw myself out of that rabbit hole into sanity...heh.
 
  • Like
Reactions: bit_user

bit_user

Polypheme
Ambassador
Its FCBGA-1090 package is just 25x24 mm and I've estimated the die size at about 9.3 x 9.9 mm (~92 mm^2), based on photos. Based on what I've found, the 4-core Skylake die is 122.3 mm^2. Considering the iGPU in the J5040 is 18 EU and 24 EU of the same Gen graphics takes about half of the Skylake die, I'd say it's plausible. Especially when you factor in the other SoC functions of a South Bridge, that have been integrated into the die. The reason I'm focused on the die size is that this package lacks an IHS.

So, thermal transfer of ~ 10 W from a 92 mm^2 wouldn't seem to be a bottleneck. However, that cheap aluminum heatsink they used on the board might be. I'm definitely looking at copper alternatives that have a nice, thick base.
 

bit_user

Polypheme
Ambassador
I've built an itx system in an Antec ISK110 with an Asus J3355I-C and by design the processor doesn't need an active cooling solution.
That's one of just a couple other case options that seem viable. I thought about buying one, but I can't seem to find it with the vertical stand - it seems they now sell it only with the VESA mounting option.

Your CPU is 2-core with a 12 EU iGPU, whereas mine is 4-core with 18 EU iGPU. However, they're both rated at 10 W. Gemini Lake is also a bigger core and clocks higher. I don't know if mine is being driven outside its TDP envelope, for it to be getting so hot. I want to try and measure this, at some point.

As for whether it requires active cooling, I've been having some self-doubt about that. I think just because the board has a passive heatsink doesn't mean they expect you to use it in a case without a case fan! That said, I have noticed the T_Junction on these SoCs is 105° C. However, I'm uncomfortable with the idea of running it so hot - even if it throttles somewhere in the low 80's, as seems to be the case, that still feels wrong. Again, especially considering this is on an open bench.

After my little die size analysis, I'm pretty confident that simply using a beefy copper heatsink and quality TIM should give me the thermal headroom I'm looking for - at least, in the open bench setup. For use within my fanless case, I'll probably still have to either restrict its clockspeed or just let it throttle.

In fact, the heatsink on yours looks like it's copper - similar to what I want.

"why stop there?"
I've been there, managed to claw myself out of that rabbit hole into sanity...heh.
I already resent the amount of time taken by the DC-converter issue and RAM compatibility problems. At this point, fiddling with the heatsink won't make too much difference in overall time & investment, yet it could help me get a lot more out of it.

BTW, another rabbit hole I went down was moving the storage external to the case, due to how hot this gets, which had me looking at 2.5" eSATA enclosures...
 

bit_user

Polypheme
Ambassador
I bought a heatsink! This will fit the foot print, and still leave enough room for me to rig a hold-down bracket to secure it from the top:


The base of the heatsink is THICK copper, which I think is almost perfect for direct-die cooling. I think distributing the heat out to the edges of the heatsink and all the fins is likely a bottleneck in the system.

The only thing better than thick copper would be a vapor chamber. Sadly, I did not find a vapor chamber cooler that would fit the footprint. Dynatron even makes heatsinks like this with a vapor chamber base, but I think they didn't start doing that until about 5 years after this one was made.

For TIM, I'm about to order some Artic Cooling MX-6, which they claim is well-suited to direct-die cooling.
 
Last edited:

bit_user

Polypheme
Ambassador
So, I got the heatsink. I'll follow up with details, but I've been experimenting on an Apollo Lake board and just wanted to mention that I measured its die at about 8.7 x 10.7 mm.

Also, I tried replacing the TIM under its heatsink with Arctic MX-4, but was still able to reach about 92 C after about 15 minutes of continuous load.

I noticed the rubber spacers that help keep the heatsink level are a bit too tall. They're compressible, but take more force to bottom out than the little springs would provide. So, this probably leaves a gap of like 0.2 mm between the die and heatsink. Therefore, I'm going to cut them down to about 1/3rd the area, so the heatsink makes better contact.