Skip to content
- 0x0000
Intel D270 vs RTX PRO 6000: the 5.9x is real, on very narrow groundIntel's 5.9x comes from FP64 option pricing. On inference the gap shrinks to 12 %, at a little over half the power, and the D270's 160 to 480 GB of LPDDR5X matter more for local deployment.
2026-11-21
- 0x0040
TensorFold on Intel Arc Pro B70: from a Python port to a Zig engineHow we ported the TensorFold inference engine to the Intel Arc Pro B70 (Xe2), first in Python, then in Zig on Level Zero: Qwen3.8-27B at 122.9 tok/s, measured against vLLM and llama.cpp.
2026-10-23
- 0x0080
A private AI on a €1,900 card: what the Intel Arc Pro B70 changes for regulated professionsA 27-billion-parameter model on a €1,900 (ex-VAT) Intel card, with no document leaving the premises: speed against vLLM, office cases and answers an auditor can replay.
2026-10-23
- 0x00c0
Serving DeepSeek-V4-Flash-0731 on Two DGX Spark (TP=2): The Full Recipe96/100 correct code judged by execution, 78.55 tok/s single-stream and 188.9 tok/s at sixteen streams with DeepSeek-V4-Flash-0731 on two DGX Spark.
2026-08-04
- 0x0100
DeepSeek-V4-Flash 0731 on 2× DGX Spark: Setup, Measurements, TuningTwelve serving parameters ruled out of sixteen, a +31 s latency cliff with b12x, and 188.9 tok/s at sixteen streams with DeepSeek-V4-Flash on two DGX Spark.
2026-08-04
- 0x0140
Serving Laguna S 2.1 NVFP4 on a single NVIDIA DGX Spark (GB10): the complete runbookvLLM runbook to serve Laguna S 2.1 NVFP4 (117B, MoE) on a single DGX Spark GB10: avoiding the JIT OOM, 256K context, DFlash speculative decoding, detailed throughput benchmarks.
2026-07-23
- 0x0180
Running vLLM on an NVIDIA DGX Spark (GB10): from git clone to NVFP4 inferenceFull vLLM install on a DGX Spark GB10 (SM121): the three bugs that block the build and the launch of an NVFP4 model, and how to fix them.
2026-07-20