It took quite a bit of prying - removed the screws and pry out one of the small green plastics outwards…
I’ve removed the green tab to make it easier to remove that panel. Whole case still fits very nicely…
It took quite a bit of prying - removed the screws and pry out one of the small green plastics outwards…
I’ve removed the green tab to make it easier to remove that panel. Whole case still fits very nicely…
It sure looks pretty, as long as you aren’t averse to green, but there are a few issues:
I could cut away the block and install the fan, or even keep they block and use a copper shim, and I guess I could grind down a NVME heatsink at the right angle for it to press against the bottom plate (the drive sits at an angle, so the socket end is about 1mm closer to the plate) … but really?
I had expected a better design given the previous cases from Radxa.
For NVMe a 2mm thermal pad works.
That the block doesn’t touch the SOC is worrisome. I guess I’m getting another 0.5mm thermal pad there.
And the whole “force the base in” thing, yikes.
Haven’t gotten mine from Ameridroid yet. Patience … patience.
I ordered a 3mm hope that also fits and used thermal paste for the SOC.
Hmm, perhaps this case mod is in my future. Putting it together the first time was quite difficult, I am loathing the next time i need to take it apart!
Got my metal case today, isy=talled the SoC into the case.
NVMe and eMMC fit perfectly.
Also I’ve applied some thermal paste:
With thermal paste the CPU idle temp reduced from ~43°C to ~36-37°C in the same environment.
alex@rock-5b:~$ sensors
gpu_thermal-virtual-0
Adapter: Virtual device
temp1: +36.1 C
littlecore_thermal-virtual-0
Adapter: Virtual device
temp1: +37.0 C
bigcore0_thermal-virtual-0
Adapter: Virtual device
temp1: +36.1 C
tcpm_source_psy_4_0022-i2c-4-22
Adapter: rk3x-i2c
in0: 12.00 V (min = +12.00 V, max = +12.00 V)
curr1: 2.25 A (max = +2.25 A)
npu_thermal-virtual-0
Adapter: Virtual device
temp1: +36.1 C
center_thermal-virtual-0
Adapter: Virtual device
temp1: +36.1 C
bigcore1_thermal-virtual-0
Adapter: Virtual device
temp1: +37.0 C
soc_thermal-virtual-0
Adapter: Virtual device
temp1: +36.1 C (crit = +115.0 C)
As for me, the main minus is a hidden GPIO including UART pins.
UPD:
The average temperature under 100% CPUs load (Building the kernel):
Every 2.0s: sensors rock-5b: Sun Dec 4 07:55:24 2022
gpu_thermal-virtual-0
Adapter: Virtual device
temp1: +51.8 C
littlecore_thermal-virtual-0
Adapter: Virtual device
temp1: +56.4 C
bigcore0_thermal-virtual-0
Adapter: Virtual device
temp1: +58.2 C
tcpm_source_psy_4_0022-i2c-4-22
Adapter: rk3x-i2c
in0: 12.00 V (min = +12.00 V, max = +12.00 V)
curr1: 2.25 A (max = +2.25 A)
npu_thermal-virtual-0
Adapter: Virtual device
temp1: +53.6 C
center_thermal-virtual-0
Adapter: Virtual device
temp1: +52.7 C
bigcore1_thermal-virtual-0
Adapter: Virtual device
temp1: +58.2 C
soc_thermal-virtual-0
Adapter: Virtual device
temp1: +56.4 C (crit = +115.0 C)
What size/type of screw do i use to attach the rock 5b to the metal cases backplate?
Or did you all get the needed screws included with the metal case? Mine didn’t include any. Just the 4 tiny ones that hold the case and backplate together.
They came in a small bag with the case … in fact there were 5 in mine!
They’re M3 … I’d guess at 6-8mm length (from memory)
It must be really hit or miss on the mounting screws-I didn’t get any either-luckily I have a couple of bags of screws and other mounting hardware from years of computer building so I’ll find something!
The idea of using goop to fill the gap between the case-post and the SOC didn’t sit well with me, so I’ve completely solved the issue by using 20mm x 20mm x 0.3mm Copper Shims
I actually bought 0.3mm and 0.5mm, but the 0.3mm lot arrived first!
I used two shims, and a thin smear of the (recycled) heatsink paste on each surface.
Case + fan + Rock5B now operate as I would expect, and certainly a HUGE improvement on what it was. Previously I had the fan minimum set to 45C in order to have a “mostly silent” experience, but that’s now reduced to 40C, with ‘idle’ temps at 35C (idle-ish running a whole heap of docker stacks ). So now I’ve got the silence of a passive case, and the reassurance of (some+adjustable) active cooling under serious loads, even with a less-than-ideal position for the fan.
No, I didn’t remove the anodising on the case-post, and I probably could have smoothed+polished the shims to prove a point, but extremely pleasing outcome regardless .
I have a more involved solution planned for the NVME drive when the parts are available/assembled … happy to share if it works.
UPDATE:
After ~100 minutes of
stress -c 8
at an ambient 26°C the BigCore temp was 57.3°C (and the entire case, of course!).
Previous temps with only the fan (I mean, the HSF) using the same test gave BigCore temp of 69.4°C (in the ‘acrylic’ case).
Once the stress test was cancelled, the BigCore temp reduced to 44.4°C in under 2 minutes, and the fan started turning off after ~5 minutes.
So, it seems to be a big reduction in SOC temp (assuming the whole case would NOT reach ~70°C over a significant period of time), but you end up with a big + hot brick under a sustained load . I probably should have run it overnight/all-day to ensure that the temp stabilised, but I ran out of time.
I actually bought 0.3mm and 0.5mm, but the 0.3mm lot arrived first!
I used two shims, and a thin smear of the (recycled) heatsink paste on each surface.
I would say if you can manage to have only one shim instead of 2 would be better, as you might loose a bit of conductivity between each layer.
Would the 0.5mm shim alone with thermal paste on both faces be enough ?
Wonder what is best between this solution and just a thermal pad
I reckon it would be, but as I mentioned the 0.3mm arrived first, and a single 0.3mm shim didn’t look right (although I may have rushed that step and not fitted the board properly) … I probably should have added that the 0.5mm have NOT yet arrived.
I could possibly test the 0.5mm when they arrive, just out of interest, but it’ll depend on the never-ending battle between time and priorities . Given the combination of cost, benefit and time I figured that one or both of the shims would probably provide the best result … I’m not sure I need perfection, but I’m sure as hell going to do better than what was provided with the case! There’s probably something to be said for the extra pressure that a larger shim, or combination of shims, might provide … but then we’re stressing the board … hehehe … let’s not go down this rabbit hole!
Check out the difference between the thermal conductivity of copper vs a thermal pad!
The numbers are 398 W/m•K for copper vs 10-14 ish (IIRC) … and the 10-14 is the GOOD stuff!
ie: the copper conducts heat, and a “thermal pad” is an insulator (by comparison).
For that matter, the “thermal paste” is also a thermal insulator (compared to the copper), but should only be the thinnest smear: it’s purpose is more to replace the air that would otherwise be between the imperfect mating of the two surfaces.
That’s what I suspected regarding thermal pads being less efficient than pure copper for heat conductivity.
Best of course would have been direct contact with case heatsink and just thermal paste on top of SOC but it seems they let too much gap in between.
If I have a chance to get that case, I’ll probably use thermal pad first, as I have some to hand and see if I need to improve especially during summer.
I have used a simple 0.5mm thermal pad (Arctic). I have tried to install 2 thermal pad layers (=1mm) but the PCB started to bend a little bit. One 0.5mm pad is just fine in my case (there may be be some production tolerance, it might be different for other)
I did a 100 minute burn @20°C ambient temperature and I have got this:
(so there is 6°C ambient temperature offset compared to your test)
Conclusion:
Don’t bother with messy thermal paste. For such low power consumption, it’s not required, a thin thermal pad is as good.
I did some Monero mining with xmrigCC for about 45 minutes just to see what the thermals would look like with Radxa’s metal case + thermal pad, and the case got pretty darn hot, and according to psensor, it was at 67 degrees C for four cores, and 64 for the other cores. It was hitting 100% CPU utilization the whole time, and the ambient room temp was 17 degrees C. I’m happy with the thermal performance under those conditions, and I won’t be stressing it to that degree ever again, because it’s not meant to be a mining computer for me.
I’m not sure how anyone gets these things assembled without modification. Regarding the (board-to-backplate) screw head interference (with the itty-bitty “shelf” in the extrusion), I divoted the shelf back to the main surface, gaining nearly 2mm clearance (1mm per side). For folks without a pile of tools lying around, this would be pretty inconvenient.
Checking my moderately large collection of M3 hardware, most screw heads vary between 5.25 and 5.5mm, with some larger. The only smaller heads I found were on sub-mini 120 degree countersunk screws, at 4.75mm. Nothing that would fit.
I also tapped the wee 2mm holes for the backplate-to-extrusion screws to 2.5mm (as the holes were conveniently the correct tap drill diameter) and reamed the backplate holes (very slightly) to 2.5mm, for M2.5 x 14mm screws. I’d have gone to M3, but there’s so little meat I didn’t want to compromise the extrusion, especially since mine is a bit squashed and I have to squeeze it together to assemble it. As for the faceplates, I could re-work them (either drill the screw boss to 2.5mm or trim it), but I doubt I’ll use them.
I’m a little surprised by your 49C temp given the description. Perhaps I have more work to do! Was your test with stress, or something else, and was the case sitting on the universally accepted test bench, hiding with it’s brethren behind the coffee cup, or mounted on a vertical surface in a draft?
I went the copper shim route not just because I expected (short term) heat transfer to be better, possibly giving lower average SOC temps under normally light loads (hard to quantify+test+compare), but also because I couldn’t justify spending MORE (per unit) on a thermal pad … I’ll be able to use the spare shims for something, right?
From that point of view, I’d have expected long(er)-term high-load temps to be all in the same ballpark (high 50’s to high 60’s) regardless of the TIM. Maybe, over that (medium?) time frame, the 20C ambient makes that much more difference to the SOC temp?
Speaking of coffee …
Yeah, it must have looked like a good idea in CAD
I reckon the trick is that the green panels are a little flexible (I assume they’re ABS), and carefully prying all the edges with a small screwdriver eventually frees them, but the problem with that is that I expect that the centre clips will be the first to droop+break from work-stress after frequent removals. I suspect that removing those centre clips would be the best option, and that the only minor issue with that approach would be that the panels are potentially not prevented from bowing after extended misuse.
It seems to me that this case was originally designed to use a thermal pad, but subsequently it was realised that repeated disassembly would be a problem, so paste was substituted as the cheapest option. Understandable, I guess, but ironic given how difficult it is to disassemble
The paste alone seems like a very poor choice, given the size of the gap. It’s like using mortar to fill the space a brick would fit!
That said, I probably wouldn’t have been so offended by the paste if it had been a THICK paste, rather than sloppy.