I ran the memory test with just the fan blowing on the top surface of the P2 and while it failed the memory test after just a minute or so, it let me play the MSlug game for at least 15mins before I stopped playing. So having memory test failures early doesn't necessarily correlate with game lockups, although I didn't run it for an extended time. Ambient room temp is 22C. I probed the edge of the P2 and saw about 38C but this is not the hottest part.
If I do make another variant I think I could put the P2 on the top surface of my FireAnt P2 board with a single sided load. When I measured the inside of the case this leaves about 6mm from the PCB top to the inner top edge of the cartridge. A 1.2mm P2, a 1.5mm AL plate and a 3mm fan would just fit. The fan would sit over the P2 and blow air down onto the plastic case. Cuts in the AL plate would allow some airflow around the hot areas. Holes may need to be drilled above the fan in the case for intake/exhaust.
The only other way is to somehow route out the plastic case underneath the PCB for the P2 to sit exposing the ground on the top surface but it makes assembling the board very difficult with a double sided load, and I'm kind of wondering where all those bypass caps would go too. I'm still deliberating...
@Rayman said:
Would it run at 300 MHz with no special cooling?
I suspect it would depend on how active the Cogs are. If they are not all at 100% active on all peripherals it may potentially be achievable but can't tell for sure.
I'm thinking that some of this might be related to the PSRAMs getting hot and performing worse at high speeds. My board doesn't have a great path to PSRAMs vs the P2 Edge for example. Some IO has to snake around on Lachlan's board to reach the mezz connector to reach my own board's layout and despite fitting the 32MB at the far end of the bus to avoid an extra stub hanging off the end to generate reflections etc, it has some traces than hop from layer to layer and no impedance is controlled. I noticed this in my initial delay test results. The timing windows of operation already rapidly taper off close to the 338 MHz operation at room temp, and increasing the temp probably makes it worse by heating some chips differently to others. 300MHz may be safer. I wonder if these games can still work at that rate in a test, just that the pitches are off and the screen is refreshed slower - may try that out...EDIT: actually probably simpler to just run the RAM test at 300MHz..
I ran the RAMtest slower just by setting CLK_MULTIPLIER to 12 instead of 14 (~290MHz), and it has already lasted over 15mins with no issues, with no fan no heatsink and fully sealed up in my intact case cartridge (not the earlier one with the huge cutout at the back). I'll stop it at 20mins open it up and quickly measure the temp it gets to.
Update: reached 20mins okay and I stopped it there yanking it out and quickly measured the P2 temp. IR thermometer showed a 68C peak and then was seen reducing after that as it cooled down. 20C ambient. So even though the P2 was as hot or hotter than in the earlier failures, the RAM was still good at this frequency...interesting.
I do wonder if I layed out a new FireAnt P2 variant on a single PCB if the memory traces could be tidied up whether we'd get a better result even up at 338MHz. Maybe kill the 64MB option to eliminate any intermediate pad RF stubs with the second unpopulated bank. There are also probably ideal ways to design it to correctly terminate the data bus for best impedance matching, but I'm no expert at that sort of stuff. I did put a series resistor on the clock line which could be tweaked. Would need a decent high speed scope to see the benefits and for tuning and I don't have access to that right now. On a new board it wouldn't be a 0402 either, making it simpler to alter.
UPDATE2: Also just ran that RAM test for 20mins fully enclosed with no cooling at a CLK_MULTIPLIER of 13 (314MHz) and it passed the RAMtest there too. I'd kind of expect equilibrium to be reached by then at this load but it may still be rising slowly - ideally I need to try to feed a thermocouple in there to see when exactly it tops out. P2 temp got up to 77C (wow it's cooking), room temp was 20C. What happens if the P2 gets too hot????
Comments
Awesome, thanks for confirming that.
Vertical fins it is then... and perhaps a small 20mm fan on AL substrate
I ran the memory test with just the fan blowing on the top surface of the P2 and while it failed the memory test after just a minute or so, it let me play the MSlug game for at least 15mins before I stopped playing. So having memory test failures early doesn't necessarily correlate with game lockups, although I didn't run it for an extended time. Ambient room temp is 22C. I probed the edge of the P2 and saw about 38C but this is not the hottest part.
If I do make another variant I think I could put the P2 on the top surface of my FireAnt P2 board with a single sided load. When I measured the inside of the case this leaves about 6mm from the PCB top to the inner top edge of the cartridge. A 1.2mm P2, a 1.5mm AL plate and a 3mm fan would just fit. The fan would sit over the P2 and blow air down onto the plastic case. Cuts in the AL plate would allow some airflow around the hot areas. Holes may need to be drilled above the fan in the case for intake/exhaust.
The only other way is to somehow route out the plastic case underneath the PCB for the P2 to sit exposing the ground on the top surface but it makes assembling the board very difficult with a double sided load, and I'm kind of wondering where all those bypass caps would go too. I'm still deliberating...
Would it run at 300 MHz with no special cooling?
The thing I made could only really run MegaYume . That’s the easiest, lowest freq one…
I suspect it would depend on how active the Cogs are. If they are not all at 100% active on all peripherals it may potentially be achievable but can't tell for sure.
I'm thinking that some of this might be related to the PSRAMs getting hot and performing worse at high speeds. My board doesn't have a great path to PSRAMs vs the P2 Edge for example. Some IO has to snake around on Lachlan's board to reach the mezz connector to reach my own board's layout and despite fitting the 32MB at the far end of the bus to avoid an extra stub hanging off the end to generate reflections etc, it has some traces than hop from layer to layer and no impedance is controlled. I noticed this in my initial delay test results. The timing windows of operation already rapidly taper off close to the 338 MHz operation at room temp, and increasing the temp probably makes it worse by heating some chips differently to others. 300MHz may be safer. I wonder if these games can still work at that rate in a test, just that the pitches are off and the screen is refreshed slower - may try that out...EDIT: actually probably simpler to just run the RAM test at 300MHz..
I ran the RAMtest slower just by setting CLK_MULTIPLIER to 12 instead of 14 (~290MHz), and it has already lasted over 15mins with no issues, with no fan no heatsink and fully sealed up in my intact case cartridge (not the earlier one with the huge cutout at the back). I'll stop it at 20mins open it up and quickly measure the temp it gets to.
Update: reached 20mins okay and I stopped it there yanking it out and quickly measured the P2 temp. IR thermometer showed a 68C peak and then was seen reducing after that as it cooled down. 20C ambient. So even though the P2 was as hot or hotter than in the earlier failures, the RAM was still good at this frequency...interesting.
I do wonder if I layed out a new FireAnt P2 variant on a single PCB if the memory traces could be tidied up whether we'd get a better result even up at 338MHz. Maybe kill the 64MB option to eliminate any intermediate pad RF stubs with the second unpopulated bank. There are also probably ideal ways to design it to correctly terminate the data bus for best impedance matching, but I'm no expert at that sort of stuff. I did put a series resistor on the clock line which could be tweaked. Would need a decent high speed scope to see the benefits and for tuning and I don't have access to that right now. On a new board it wouldn't be a 0402 either, making it simpler to alter.
UPDATE2: Also just ran that RAM test for 20mins fully enclosed with no cooling at a CLK_MULTIPLIER of 13 (314MHz) and it passed the RAMtest there too. I'd kind of expect equilibrium to be reached by then at this load but it may still be rising slowly - ideally I need to try to feed a thermocouple in there to see when exactly it tops out. P2 temp got up to 77C (wow it's cooking), room temp was 20C. What happens if the P2 gets too hot????
Just so you know what I'm working with. I've gotta find a home for all these bypass caps, now sized as 0402.
