The M.2 SSD Mushkin Vortex is known to thermal throttle under load. Combine this with the pathetic ASRock SSD cooling solution and its thermal throttles under what I consider normal use. Assuming you have access to CAD, CAM, 3D-printer, CNC-mill and aluminum laying around this isn’t a big deal:
After removing the sticker we see from the right to the left two TLC-NAND packages, DRAM and right next to the M.2 connector the Innogrit IG5236 controller. Noteworthy is that the die is exposed and is flush with the frame/package.
Why is this amazing?
As Innogrit publishes some specifications we know this controller has a package power of 3W. Approx. 5x5mm die size works out to 12W/cm^2 power density. Not comparable with modern CPUs ( > 50W/cm^2) but still enough to be problematic without a heatsink.
After drawing the initial concept, it was time for a test print: It did not fit.
With 40mm aluminum stock laying around I decided to make it 38mm but forgot about the AM4 mounting hardware. Additionally, the screws were too short.
At least the heatsink clears the m.2 slot and the cooler contacts the die like it should. In the end this test print was still a success.
Fixing the initial design errors I jumped directly to milling it out of solid aluminum. Special attention was given to the surface touching the SSD controller. Any imperfections might affect cooling or worse causes the die to crack which obviously means data loss. The end result is a 90 g chunk of aluminum with fins. As this cooler is overkill I skipped abrasive blasting.
At this point it might be worth mentioning some design considerations:
Did I call Asrock cooling pathetic at the beginning? This aluminum cover might have been good enough for nearly every SSD but they screwed up: The rear screw doesn’t line up with the M.2 resulting in very poor mounting pressure which is catastrophic for thermal pads. This issue is exaggerated by the detail that the only pressure there is, is from the M.2 contact springs pushing it up and the controller is right next to it meaning nearly zero pressure there. Thankfully this can be fixed by cutting a thin slice from a 6 mm pneumatic line.
The results speak for themselves. What was thermal throttling is now 55°C after multiple benchmarks writing and reading more than 500 GB. In the Atto benchmark, there is also no longer any thermal throttling. The anomaly around 256KB is probably caused by this being now my Windows boot drive.
After removing the sticker we see from the right to the left two TLC-NAND packages, DRAM and right next to the M.2 connector the Innogrit IG5236 controller. Noteworthy is that the die is exposed and is flush with the frame/package.
Why is this amazing?
- Direct die makes cooling easier compared to epoxy packages.
- The frame reduces the risk die cracking.
As Innogrit publishes some specifications we know this controller has a package power of 3W. Approx. 5x5mm die size works out to 12W/cm^2 power density. Not comparable with modern CPUs ( > 50W/cm^2) but still enough to be problematic without a heatsink.
After drawing the initial concept, it was time for a test print: It did not fit.
With 40mm aluminum stock laying around I decided to make it 38mm but forgot about the AM4 mounting hardware. Additionally, the screws were too short.
At least the heatsink clears the m.2 slot and the cooler contacts the die like it should. In the end this test print was still a success.
Fixing the initial design errors I jumped directly to milling it out of solid aluminum. Special attention was given to the surface touching the SSD controller. Any imperfections might affect cooling or worse causes the die to crack which obviously means data loss. The end result is a 90 g chunk of aluminum with fins. As this cooler is overkill I skipped abrasive blasting.
At this point it might be worth mentioning some design considerations:
- The surface touching the controller is raised to compensate for the thermal pad thickness (NAND cooling)
- cooler width is still 38mm and as such a cutout was added to clear the PCIE slot (Ryzen 7 5700G build)
- fin spacing is optimized for passive cooling. If you see an M.2 heatsink with fins tightly packed they either screwed up or it was designed for active cooling.
- Fin thickness could be smaller. This was mainly an attempt/compromise to push rapid feeds and speeds. Due to insufficient chip clearing this attempt caused the end mill to clog. and me testing if my workpiece holding can withstand a 1.5kW spindle/VFD (spoiler: workpiece holding was stronger and overpower protection works!). Thankfully the endmill didn't decide to friction stir weld itself to the aluminum stock.
- The label was sunken by it's thickness. As such it's flush to the surface.
Did I call Asrock cooling pathetic at the beginning? This aluminum cover might have been good enough for nearly every SSD but they screwed up: The rear screw doesn’t line up with the M.2 resulting in very poor mounting pressure which is catastrophic for thermal pads. This issue is exaggerated by the detail that the only pressure there is, is from the M.2 contact springs pushing it up and the controller is right next to it meaning nearly zero pressure there. Thankfully this can be fixed by cutting a thin slice from a 6 mm pneumatic line.
The results speak for themselves. What was thermal throttling is now 55°C after multiple benchmarks writing and reading more than 500 GB. In the Atto benchmark, there is also no longer any thermal throttling. The anomaly around 256KB is probably caused by this being now my Windows boot drive.
