This is the server I build back in 2023, based on a moded Dell Optiplex 3010 SFF motherboard with i5-2300 and Tesla M40. I wasn’t planning to put an Xeon E3 back then but something happened which changed my mind.
After approximately 1.5 years of 24/7 running, my CPU power extension cable melted.
This is just like the PCIe to EPS adapter situation last time, so I have to replace it with a heavier gauge (something better than the cheapest like above).
By this time, I found out that the TDP of i5-2300 is 95 W and E3-1230 v2 is 69 W, which is quite out of expectation. So I upgraded to E3-1230 v2 for $15, which is not only more powerful but also can lower a few bucks out of the running cost per year.
However, one more incident happened soon after I fixed the power cable.
My AIO cooler stopped pumping for unknown reason. This caused GPU overheating and must be fixed ASAP.
This Corsair AIO cooler is in a heavily used condition with broken lid when I first buying it ($25), so 1.5 years of service life is acceptable.
But I want to go something better for a longer service life this time, so I ordered a $29 Corsair AIO cooler, very similar model but in a never used open box condition.
Shipping takes around 1 week and I have to use my local AI models everyday, so I decided to make something temporarily out of my e-waste pile.
It ends up mounting a Pentium 4 stock heatsink with my Zip-tie method, and blow wind from sideways using a noctua case fan.
The idle temperature changed from 24C to75C:
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.54.14 Driver Version: 550.54.14 CUDA Version: 12.4 |
|-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 Tesla M40 24GB On | 00000000:01:00.0 Off | 0 |
| N/A 75C P0 65W / 250W | 2583MiB / 23040MiB | 0% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
Under a work-load, it changed from 52C to 86C:
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.54.14 Driver Version: 550.54.14 CUDA Version: 12.4 |
|-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 Tesla M40 24GB On | 00000000:01:00.0 Off | 0 |
| N/A 86C P8 65W / 250W | 14809MiB / 23040MiB | 98% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
The number is not good at all but I have to admitted the stock heatsink works pretty well. Because as long as the temperature is under 90C, my M40 would work normally without any risk of shutdown by overheating.
But this is only for cool weather, I still need the new AIO cooler before summer.
Idle temperature of the new AIO:
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.54.14 Driver Version: 550.54.14 CUDA Version: 12.4 |
|-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 Tesla M40 24GB On | 00000000:01:00.0 Off | 0 |
| N/A 28C P8 17W / 250W | 3MiB / 23040MiB | 0% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
Normal workload temperature of the new AIO:
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.54.14 Driver Version: 550.54.14 CUDA Version: 12.4 |
|-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 Tesla M40 24GB On | 00000000:01:00.0 Off | 0 |
| N/A 38C P0 80W / 250W | 13275MiB / 23040MiB | 45% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
Full-load temperature of the new AIO:
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.54.14 Driver Version: 550.54.14 CUDA Version: 12.4 |
|-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 Tesla M40 24GB On | 00000000:01:00.0 Off | 0 |
| N/A 55C P0 250W / 250W | 22011MiB / 23040MiB | 100% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
This results a much more pleasant temperature number after all hassle. There are around 3-4C difference between the old and new AIOs.
The reason is the old AIO has a 360mm radiator and the new AIO has a 120mm radiator. So size matters!