brucethemoose

Oh yeah, anything private sounds fishy. I think the only "public" OpenJDK builds that provide any kind of gain are Intel's build for Clear Linux and Oracle's Graal CE/EE. For...

Here are the logs from that above scenario: Host: https://smapi.io/log/fd36daf80ab4470abb47b4c0bcd1a721 Client: https://smapi.io/log/fd36daf80ab4470abb47b4c0bcd1a721 I documented this issue with more screenshots, albeit in a less minimal environment and with some unrelated MP...

My tests show it hits server throughput by approximately 40%, and their own description claims its optimized for startup time, not throughput... That's quite a hit to overcome with flag...

Hmm, there are some interesting flags documented here: https://www.eclipse.org/openj9/docs/x_jvm_commands/

I was recommended these flags for low memory usage, and they seem to do just that: https://gist.github.com/FluffyFoxUwU/69f8f156feefae3d826ad0d15c694002

> Ooh, in the same configuration CE results in unbearable log spam with GL errors. Oofie. Oh interesting. CE is basicaly OpenJDK, so I'm surprised it has any issues. The...

@HyperSoop you can fix this for now by rolling back to GraalVM 22.1.0 and removing `-Dgraal.EarlyGVN=true -Dgraal.StripMineCountedLoops=true` Also, I can't replicate your OpenGL errors with CE

> I see you have already figured out the problem with EE. Not exactly :(. I know its some kind of enterprise compiler regression between 22.2.0 and, 22.1.0, but I...

> > > > I actually have iris on that instance, so it's just poor compatibility with immptl and iris which is somewhat known. Oh actually I get that as...

Oh I've confirmed this breaks Iris too, you were right. I've updated the args to remove the offending flag.