Blog

Engineering notes on local inference, heterogeneous computing, GPU kernels, speculative decoding, and hands-on benchmarks.

3 articles

Latest

New releases and benchmark reports.

Ling 3.0 Flash running with Lucebox on an NVIDIA DGX Spark

Ling 3.0 Flash: up to 141.9 tok/s with adaptive DSpark and FlashKDA

Our result on one DGX Spark: up to 36.4% faster prompt reading, near-tied ordinary generation, and 141.9 tok/s in the best matched DSpark case.

Qwen3.8-27B running on an AMD Radeon AI PRO R9700 with the DFlash2 block-diffusion drafter

Qwen3.8-27B on the AMD R9700: up to 227 tok/s

One Radeon AI PRO R9700 serves Qwen3.8-27B with the z-lab DFlash2 drafter: 208 tok/s HumanEval average, 3.8x llama.cpp decode with the same drafter, on a stock quant that matches an 8-bit reference.

Lucebox and Geometric with AMD Strix Halo hardware and a blue DeepSeek whale under a starry sky

Lucebox × Geometric: DeepSeek V4 Flash 0731 reaches 32.7 tok/s

On AMD Strix Halo, the 98.29 GB model scores 82/92 and reaches 32.7 tok/s with DSpark. Across our fixed HumanEval, GSM8K, and MATH evaluation, it averages 27.9 tok/s.

15 articles

More articles

Previous posts, newest first.

An amber tool prefix feeding three progressively growing blue agent turns

Tool prefix caching: 48× faster warm prefill for agent loops

Agent turns resend thousands of tool-definition tokens. Lucebox restores that stable prefix and processes only the new conversation: 1.04 seconds warm after a 50.35 second cold turn.

An AMD-Powered Lucebox combining a Radeon AI PRO R9700 with Ryzen AI MAX+ 395

Lucebox beats DGX Spark by 3.63× on DeepSeek V4 Flash decode

51.1 tok/s on Lucebox versus our 14.09 tok/s average on one NVIDIA DGX Spark for the full 284B model. The complete $6,499 Lucebox costs 31% less than two DGX Sparks.

DeepSeek V4 Flash running locally on AMD Strix Halo unified-memory hardware

DeepSeek V4 Flash: 284B model, up to 32 tok/s on AMD Strix Halo

The full target runs locally with 128 GB unified memory, reaching up to 32 tok/s decode and roughly 250 tok/s indexed sparse prefill.

Laguna XS 2.1 on an RTX 3090: 296 tok/s peak, flat 152 tok/s at 256K context

Laguna XS 2.1 on a RTX 3090: 296 tok/s peak, 152 tok/s at 256K

poolside's coding MoE with its official DFlash drafter: 296 tok/s peak, a flat 152 tok/s at 256K tokens, and prefill at 3,500 tok/s on one 24 GB card.

Luce KVFlash keeping a small resident pool of KV on the GPU and paging the rest to host RAM

Luce KVFlash: 256K context with 72 MiB of KV on the GPU

KVFlash pages cold 64-token chunks to host RAM bit-exact, holding Qwen3.6-27B decode at 38.6 tok/s from 64K to 256K with unchanged accuracy.

Lucebox shipping as one Docker image that runs across the supported GPU range

Lucebox in a container: one image for every supported GPU

A prebuilt image spans the RTX 2080 Ti through RTX 5090. The fat-binary compile happens once in CI, with two host dependencies, self-tuning, and build provenance included.

Luce Spark serving a 33-35B MoE from a fraction of the experts on consumer memory

Luce Spark: fit Qwen3.6 35B and Laguna XS.2 on a 16 GB GPU

Spark keeps only the experts traffic uses resident and swaps the rest: Qwen3.6 35B-A3B in 13.3 GiB and Laguna XS.2 in 14.6 GiB, self-tuning with one flag.

Gemma 4 26B on an RTX 5090 Laptop next to DeepSeek V4 Flash on a MacBook

Gemma 4 26B edges out DeepSeek V4 Flash at 5× the speed

A ds4-eval-92 head-to-head: Gemma 4 26B on a 24 GB RTX 5090 Laptop ties DeepSeek V4 Flash on a 192 GB Mac at 78.3%, and decodes about five times faster.

Lucebox client harness experiments on an RTX 3090

Launch and tune Lucebox with real agent harnesses

Real-client profiles, launch scripts, and TQ3/DDTree results for OpenCode, Hermes, OpenClaw, Open WebUI, Codex, Claude Code, and Pi.

AMD Strix Halo running Qwen3.6-27B locally through Lucebox

DFlash + PFlash on AMD Strix Halo: 2.5× end-to-end versus llama.cpp

Qwen3.6-27B on the Ryzen AI MAX+ 395 iGPU: 26.85 tok/s DFlash decode and a 2.51× end-to-end gain at 16K plus 1K generation.

Laguna XS.2 running on a single RTX 3090 inside the DFlash daemon

Laguna XS.2 on a 3090: 111 tok/s and 5.4× prefill

Poolside Laguna XS.2 ported into DFlash and PFlash as the first MoE target supported by PFlash: about 107 tok/s decode and 5.4× faster 128K prefill than llama.cpp.

PFlash speculative prefill compression for DFlash

PFlash: 10× prefill speedup over llama.cpp at 128K on a RTX 3090

PFlash compresses 128K to 2.6K tokens before DFlash sees the prompt: 24.8 seconds to first token versus about 257 seconds for llama.cpp, with measured retrieval preserved.

Qwen3.5-27B DFlash on ggml

DFlash on ggml: up to 207 tok/s Qwen3.5-27B on a RTX 3090

A standalone C++ and ggml speculative decoder with a DFlash block-diffusion draft and DDtree verifier: 3.43× AR and 128K context on 24 GB.

RTX 3090, eGPU dock, and MacBook running NVIDIA on macOS over USB4

The eGPU myth: why a $300 dock will not make an AI workstation

tinygrad wrote an NVIDIA driver from scratch. We tested real models on an RTX 3090 over USB4: brilliant engineering, but the performance numbers are not there yet.

RTX 3090, the GPU behind the megakernel

Megakernel: matching Apple Silicon efficiency at 2× the throughput

The first megakernel for hybrid DeltaNet and attention LLMs fuses all 24 layers into one CUDA dispatch, reaching 1.87 tok/J on an RTX 3090.