Local AI on a PC under $800: Tesla P100 and Strata
Qwen3.8-Flash-Next IQ3_S on a sub-$800 build with a 2016 Tesla P100 and Strata.
Machine P1005: Intel Xeon E5-2690 v4 (14 cores / 28 threads), NVIDIA Tesla P100 16 GB, 32 GB DDR4 ECC, 1 TB NVMe, Ubuntu Server 24.04 LTS, Strata 0.1.55, model Qwen3.8-Flash-Next IQ3_S (~83.6 GB).
Measured performance: decode 40–44 tok/s on structured outputs (peak 47.2 tok/s), prefill 294 tok/s at a 16K prompt. The listed price under $800 includes a new case, PSU, cooling and NVMe; other parts are used.
Raw benchmarks, logs and configs are published on GitHub. The point of the build is practical local AI on hardware that small businesses and homes can still afford.