jakethompson1 wrote on 2020-07-04, 03:54:I've got 486DX2 parts on the way (fun!) and thinking what to do (or not) about a heat sink as I see different opinions on the is […]
Show full quote
I've got 486DX2 parts on the way (fun!) and thinking what to do (or not) about a heat sink as I see different opinions on the issue.
I decided to go to the source and found the 486DX2 data sheet (https://archive.org/details/bitsavers_intelda … e/n219/mode/2up). I don't quite understand the table on page 15-3 but it seems that at room temperature the CPU needs 400 CFM of airflow to be safe without a heatsink.
On the other hand on page 15-2 it seems the table shows the most optimistic situation of a heat sink and 1000 CFM airflow, is 2.5 degrees Celsius per watt thermal resistance.
I was checking out adhesive sinks at DigiKey at around 44mmx44mm size and came up on this one: https://www.ohmite.com/assets/docs/sink_bgah.pdf
Nice thing is it only costs $6 and includes thermal tape.
It appears its performance at 200 CM is better than Intel's ratings at 1000 CFM ... are heatsinks really that much better than back then or this one just needs more vertical clearance?
I found a similar table for the Am5x86 (https://en.wikichip.org/w/images/f/f0/Am5x86_ … %2C_1996%29.pdf) on page 65. It seems this heatsink wouldn't even need a fan to outperform what they list for a heatsink+fan.
Table 15-3 is showing maximum ambient temperatures in degrees Celsius vs Airflow in Linear Feet per Minute (LFM). The maximum ambient temperature should be treated like an absolute maximum that you never intend the CPU to operate at. So with Ta at 18.9C/66F, no heatsink, and 400 LFM of air hitting the chip case, the CPU would be at maximum operating temperature. If ambient temperature goes up just a degree or so, your chip is toast.
Table 15-2 basically shows the specs of their reference cooling solution and the thermal resistance specs of the chip itself. You can plug these numbers into the formulas and get the numbers in table 15-3. When referring to the 1000 LFM thermal resistance performance of the reference heatsink compared to the Ohmite sink, I assume you're referring to the 45x45x12.5 Ohmite model. This heatsink is much taller and has much more surface area compared to the reference sink, so the increased performance is no surprise. If you can find a datasheet for a heatsink of similar dimensions and designs to the reference sink, you'll probably find the numbers to be similar. I don't know if heatsinks have gotten better, I just think that the reference design was "good enough" for Intel's engineers.
Keep in mind that the Ohmite heatsink is intended to operate with (usually ducted) airflow traveling laterally along its fins, not top-down like traditional heatsink+fan combos of the day.
On to the CFM vs LFM issue:
To convert to CFM, you have to multiply LFM by the cross-sectional area of the airflow.
So, for example, let's assume you've built a duct around this Ohmite heat sink, the cross-sectional area of the airflow is 68mm x 12.5mm. In feet that's 0.223 ft x 0.041 ft. If you want 200 LFM, then you need a fan that can push 1.83 CFM.
200LFM * (0.223ft x 0.041ft) = 1.83 CFM
This is easily achievable with most small 40mm fans. The hardest part will be building the duct. If you don't use a duct, you will have to figure out what the airflow looks like so you can calculate the cross-section. For instance, in a traditional, non-ducted, downward pointing fan situation, the cross-section could be estimated as 40mm x 40mm. In this case you would need to push 3.43 CFM to achieve the same result.
200LFM * (0.131)^2 = 3.43 CFM
Keep in mind that because of the Ohmite fin design, the air may not flow around all the fins in this scenario, raising the heatsink's thermal resistance. So you may have to push even more air to achieve the same thermal resistance.
Now, because the heatsink's performance appears to be much better than the reference design, you can probably get away with no fan. You can put the datapoints from the Ohmite datasheet into a spreadsheet and extrapolate that curve to see what the thermal resistance is at 0 LFM. The result may not exactly reflect reality, but it may be close enough to determine if it's safe.
The reason the Am5x86 chips require less cooling is they run at a lower voltage, so they dissipate less heat. The Intel DX2 is a 5V part and the Am DX5 is a 3.3V part. In fact, I compared datasheets, and the Am5x86DX5 133 allegedly has better thermal characteristics than the Intel DX and DX2 chips.
As others have commented, I would err on the side of too much cooling with these old chips. It's only a matter of time before the gold scrappers have made your chip impossible to replace.