oyd@lab:~oyd bench --list --verbose

Benchmarks

How fast local language models run on our hardware: which hardware, which model, which settings and what result. The list grows with every new test.

stdout

> 0 results

> First benchmarks are on the way

> We will publish only what we measure ourselves on our own hardware: the model, the settings and the result. For now, the hardware is ready.

oyd@lab:~lsdev --lab --all

Our hardware

We measure on our own hardware: from a single consumer graphics card to several NVIDIA DGX Spark units and Apple Silicon machines.

qty	device	mem	status	note
4×	NVIDIA RTX 3090	24 GB	in use	One runs as an eGPU server. We have a 4-slot NVLink bridge.
3×	NVIDIA DGX Spark	128 GB	in use	Linked over ConnectX-7. We measure how models scale from 1 to 4 units.
1×	Mac Studio M5 Ultra	—	arriving
1×	NVIDIA RTX 2080 Ti	11 GB	in use	We plan a memory upgrade to 22 GB and a two-card 44 GB rig.
1×	NVIDIA RTX 3060	12 GB	in use
2×	AMD RX 480	8 GB	in use
2×	Mac (Apple Silicon)	—	in use	Two Macs linked over Thunderbolt RDMA.
2×	X99 (Intel)	—	in use
1×	X299 (Intel)	—	arriving

oyd@lab:~cat METHOD.md

[1] We measure on our own hardware, with the models and settings listed next to each result.
[2] We publish only what we measured ourselves. We do not repeat vendor numbers.
[3] Where we tuned a setup, we show the result before and after.