sacrelege

6 ms·

1.
Rvidia-exporter – Prometheus metrics exporter for Nvidia GPUs(github.com)
3 points by sacrelege 5mo ago|1 comments
2.
by sacrelege 5mo ago
Exposes 16 GPU telemetry metrics (temperature, utilization, power, clocks, fan speed, memory, processes, etc.) via a single ~2 MB binary built in rust. Includes a minimal NVML FFI layer via libloading — no nvml-wrapper, no build script, no
3.
by sacrelege 5mo ago
no I don't, since there seem to be a silent degradation bug
4.
by sacrelege 5mo ago
Ah thanks, I love coffee At a high level, it's a mix of our own GPU capacity plus the ability to burst into external nodes when things get busy. Right now we're running a bunch of RTX PRO 6000s, which basically forces you into wor
5.
by sacrelege 5mo ago
We typically observe throughput of around 100–110 toks/s, and for larger context sizes this ranges between 90–100 toks/s. While we don't guarantee a fixed toks/s rate, we scale by provisioning external GPU nodes during p
6.
by sacrelege 5mo ago
Thanks for putting N-Day-Bench together - really interesting benchmark design and results. I'd love to see how the model we serve, Qwen3.5 122B A10B, stacks up against the rest on this benchmark. AI Router Switzerland (aiRouter.ch) can
7.
by sacrelege 6y ago
the posted thread is rather interesting and I would love to see more critical discussions like this on the issues of bitcoin. unlike the often seen circle jerk of bitcoin maxis, this could actually lead to solutions or issues getting addres
8.
3D printed 3-sided cryptocurrency themed LED lamp(bitlamp.ch)
1 points by sacrelege 6y ago|1 comments
9.
by sacrelege 6y ago
I'm proud of my work, I hope all of you like it too :)