very low cpu utilization snapdragon x2 elite extreme

[Ncard00123]

Describe the bug

Currently running PPS (LLR) tasks on my X2 Elite Extreme snapdragon laptop.

But only 6 cores are being fully utilized, which translates to 20-25% cpu utilization.

Task manager priority is set to low by default, but even setting it to high makes no difference.

So even if Native ARM64 tasks are not possible right now, please fix this utilization issue.

And sierpinski/riesel base 5 (LLR) is only utilizing 5%, so please check all the CPU applications, and make sure they all utilize the snapdragon x2 elite extreme 100%.

Steps to reproduce

No response

Expected behavior

No response

Screenshots

No response

System information

No response

Additional context

No response

[AenBleidd]

If the project's application doesn't utilize 100% of the CPU core - this is not our problem.
BOINC client runs all the applications with the low priority to avoid system being too slow, but this could be overriden in the client configuration by setting <process_priority> parameter.
Also, on the modern laptops (not sure about one based on ARM) OS kernel throttled some heavy applications to avoid system overheat and too fast battery discharge.

[Ncard00123]

If the project's application doesn't utilize 100% of the CPU core - this is not our problem. BOINC client runs all the applications with the low priority to avoid system being too slow, but this could be overriden in the client configuration by setting <process_priority> parameter. Also, on the modern laptops (not sure about one based on ARM) OS kernel throttled some heavy applications to avoid system overheat and too fast battery discharge.

The low priority is not the issue. I have already tested with manual priority overrides and it does not change the utilization.

The problem is not thermal throttling either. The Snapdragon X2 Elite Extreme is running in Whisper Mode at a stable 2 GHz across all 18 cores with plenty of thermal headroom.

The issue is that the science applications are x64 binaries running through Prism emulation. On high-core-count ARM64 hardware, the x64-to-ARM translation layer appears to hit a wall in instruction throughput or thread scheduling.

Specifically, PrimeGrid PPS (LLR) only utilizes 6 out of 18 cores (33%), and Sierpinski/Riesel (LLR) only utilizes a single core (5%), regardless of the priority settings in BOINC or Windows.

If Fresco aims to be a modern manager for the next generation of hardware, it should help bridge the gap in identifying these scheduling bottlenecks on ARM64, rather than dismissing it as a project-only problem. At the very least, the GPU detection needs to be fixed so we don't default to a 96 MFLOPS software renderer on a TFLOPS-capable Adreno X2-90.