Hardware
The lab testbed
The machines that actually exist in the lab, with their operational status and their limits stated. No result generalizes beyond the tested unit and versions; environment fingerprints are frozen at flagship preregistration.
- online
2× Beelink GTR9 Pro — AMD Strix Halo
Ryzen AI Max+ 395, 128 GB LPDDR5X-8000 unified, Radeon 8060S (gfx1151), dual 10GbE — two commercially identical units
- What it is for
- Flagship qualification target: backend comparisons (Vulkan vs ROCm), unit-to-unit replication, multi-slot serving, RAG side-services
- Limit
- Two units do not represent the whole production batch
- online
2× NVIDIA DGX Spark — GB10
GB10 Grace Blackwell, 20-core Arm, 128 GB unified, sm_121, ConnectX-7 — linked point-to-point over 200G RoCE
- What it is for
- aarch64/sm_121 portability, multi-node topologies, long-context and speculative decoding studies
- Limit
- Expensive narrow testbed; results do not generalize to datacenter Blackwell
- online
RTX 5090 workstation
Consumer Blackwell, 32 GB GDDR7, x86 host
- What it is for
- CUDA control lane, fine-tuning/distillation, consumer-GPU baselines
- Limit
- One configuration; not an enterprise server
- online
Apple M1 Max, 64 GB
Apple Silicon, 64 GB unified, macOS / MLX lane
- What it is for
- Apple/MLX smoke tests and small cross-platform anchors
- Limit
- Not the current high-end Apple generation
- online
MacBook Pro — Apple M4 Pro, 24 GB
Apple Silicon M4 Pro, 24 GB unified, macOS / MLX lane
- What it is for
- Current-generation Apple anchor: laptop-class local inference, MLX vs llama.cpp Metal comparisons
- Limit
- 24 GB caps model size; thermals of a laptop, not a desktop
- online
Laptop — Ryzen 9 9955HX + RTX 5070 Ti
Ryzen 9 9955HX (Zen 5, 16C), RTX 5070 Ti Laptop 12 GB GDDR7, 32 GB DDR5
- What it is for
- Discrete-GPU laptop lane: mobile Blackwell CUDA, VRAM-constrained inference and offload studies vs unified-memory machines
- Limit
- 12 GB VRAM forces offload for mid-size models; laptop thermals
Measured on this fleet
Hardware is lab-owned; there are no loaner or client-shipped units. Provenance and modification status are disclosed per project on the disclosures page.