3 ms·
tl;dr: CPU overclock performs the same (as expected) as a Pi 4 model B, and was stable in all my testing up to 2.0 GHz. The board won't normally overheat/throt
by geerlingguy 6y ago
tl;dr: CPU overclock performs the same (as expected) as a Pi 4 model B, and was stable in all my testing up to 2.0 GHz.
The board won't normally overheat/throttle if you keep it at the stock 1.5 GHz clock, but when overclocked, you do need better cooling, whether a very large passive heatsink or any kind of fan to move the air over the SoC better (heatsink not required but many people enjoy sticking one on).
The result that surprised me was the NVMe drive I was testing performed 15% faster (overall) when I overclocked the CPU. I wasn't expecting that much of a difference in NVMe performance, and looking around elsewhere I couldn't find a similar kind of result—maybe other CPUs/busses are just so fast at their base frequencies that it isn't an issue at all.
I mean, it's not like the Pi CM4 can put through 1500 MB/sec on its PCIe 1x lane.
- justinclift 6y agoAt the moment with the cooling, the air is only cooling the top of the board. eg: -> | -> | -> | _____ Wonder if raising the board, so air cools both the top and bottom, would make a difference? eg: -> | _____ -> | -> |
- CobaltFire 6y agoGiven the simplicity of the RPi system, it's quite possible that the PCIe bus is clocked from the same source that you are raising the frequency of to overclock the CPU. Since we saw a ~15% increase in NVME performance with a ~33% increase in CPU clock, I'd even go so far as to say that the 100MHz clock to the PCIe bus is somehow derived from the base clock in a way that means that any n% increase in CPU clock would yield an n/2% increase in PCIe performance. This tracks with USB not seeing an increase in speed as that's likely clock stabilized separately.