Most of the RISC-V systems we've been able to add into the jenkins CI have been too slow to put into the main node-test-commit runs.
While this could have been mitigated if we take the "fanned" approach as we have for other platforms that is not ideal.
There are newer boards based on the SpacemiT-K3 SoC which are starting to become available and I have acquired one today with a view to evaluating the performance. Initial tests suggest it can build Node.js in around 3h with an unpopulated ccache which puts it in line with some of our other platforms meaning that it would not unduly delay PR testing if we were to integrate such systems into the PR testing process.
Scaleway are looking at having K3 based servers in the near future to supplement the existing slower TH1520 boards that they have available, so it is likely that we would be able to obtain capacity along the lines of what I'm currently testing. Based on other testing I've done the K3 boards can do things around 4-5 times faster than the TH1520 ones in many scenarios (even excluding the use of the eight additional slower A100 cores on the board)
Adding it to the CI would be conditional on it being able to run cleanly, and while my initial runs have shown some unexpected issues/crashes I will work through some of those in this issue and #4099 for any other issues related to getting to green on this platform.
Most of the RISC-V systems we've been able to add into the jenkins CI have been too slow to put into the main
node-test-commitruns.While this could have been mitigated if we take the "fanned" approach as we have for other platforms that is not ideal.
There are newer boards based on the SpacemiT-K3 SoC which are starting to become available and I have acquired one today with a view to evaluating the performance. Initial tests suggest it can build Node.js in around 3h with an unpopulated ccache which puts it in line with some of our other platforms meaning that it would not unduly delay PR testing if we were to integrate such systems into the PR testing process.
Scaleway are looking at having K3 based servers in the near future to supplement the existing slower TH1520 boards that they have available, so it is likely that we would be able to obtain capacity along the lines of what I'm currently testing. Based on other testing I've done the K3 boards can do things around 4-5 times faster than the TH1520 ones in many scenarios (even excluding the use of the eight additional slower A100 cores on the board)
Adding it to the CI would be conditional on it being able to run cleanly, and while my initial runs have shown some unexpected issues/crashes I will work through some of those in this issue and #4099 for any other issues related to getting to green on this platform.