Qwen Vulkan PRs 28489 and 28501: below noise
I benchmark llama.cpp PR 28489 and PR 28501 with the 103.7 GiB Qwen3.8 Flash Next model on Strix Halo. Both are correct; neither is a production speed-up.
Field notes35 published
The complete field-note archive.
Every published investigation in date order. Use the Strix Halo and Storage pages when you want a shorter route into one subject.
Browse the archive
Two draft changes delivered most of Qwen3.8's stack gain
I isolate five Qwen3.8 27B optimisations on Strix Halo. Two improve the current Vulkan stack; IQ4 main and TOP_K do not.
Qwen3.8 can now roll back recurrent state, but that is only half the MTP story
PR 28123 makes Qwen3.8 recurrent rollback practical for MTP. It removes a large copy cost, but does not make concurrent drafting safe.
The Qwen3.8 correctness merge I would carry before chasing speed
PR 27941 fixes five Qwen3.8 state and QSA correctness faults. I would carry it into the next EVO-X3 canary, but not promote it untested.
Two QSA gather patches attack Qwen3.8's long-context decode cost
PRs 28213 and 28244 gather QSA-selected K/V cells instead of masking the full cache. The long-context gains are real leads, not Vulkan results.
A Qwen3.8 prompt-cache slot can loop forever
Issue 28280 records a Qwen3.8 prompt-cache checkpoint livelock on Strix Halo. Health alone can miss it, so slot progress needs watching.
Six Qwen3.8 MTP heads; only one fits my next test
Unsloth's six Qwen3.8 MTP GGUF sidecars range from 1.776 to 7.237 GiB. Shared Q8_0 is the sensible canary, not a production download.
Qwen3.8 MTP can mix users' answers across slots
A gfx1151 report shows Qwen3.8 draft-MTP content crossing between concurrent slots. That is a correctness blocker for my dual-slot service.
A one-line Vulkan fix may unlock Q8 K/V prefill, but 125B still needs proving
Merged PR 28190 makes Vulkan's Flash Attention dequant path engage before the KV cache is full. It is promising, not yet a 125B result.
Lazy auto mode can halve Qwen3.8 prefill on Strix Halo
An exact Strix Halo report measures 216 versus 406 prompt tokens/s from one llama.cpp lazy-mode switch. New canaries need it made explicit.
Lemonade 11.9 on EVO-X3: the control plane changed, inference did not
I upgraded EVO-X3 from a patched Lemonade 11.8 daemon to official 11.9. Fixed prefill held, decode moved -1.93% and every route passed.
Two llama.cpp Qwen fixes on Strix Halo: no free speed-up
I backported two merged llama.cpp Qwen fixes into the qualified Strix Halo runtime. One was 3–5% slower; the other lost the Vulkan device during model load.
Qwen3.8 Flash Next: Vulkan 0.7.1 and Q8 in production
I moved Qwen3.8 Flash Next to a Q8 dual-slot Vulkan 0.7.1 profile, measured 5–8% gains and passed a 30-minute cached concurrency soak.
Qwen3.8 Flash Next on AMD Strix Halo: ROCm vs Vulkan
I compare ROCm and Vulkan for Qwen3.8 Flash Next on AMD Strix Halo, then tune a 104GiB model to 15.6 tok/s single-stream and 24.3 tok/s aggregate.
Ornith 1.5 on Strix Halo: the 32K production profile
I qualified a patched Vulkan build for Ornith 1.5 35B on Strix Halo, lifting 32K prefill by 24% and passing two 30-minute production soaks.
Qwen3.8 at 262K: fixing DFlash2 for production
I repaired Qwen3.8 DFlash2 state reuse and a deterministic Vulkan stall, then deployed separate fast and exact 262K routes with memory-aware eviction.
Vulkan 0.6.10 on Strix Halo: the MTP fix has a cost
Vulkan 0.6.10 fixes the long MTP stall on my Strix Halo machine, but Qwen3.6 decode falls about 11% and DeepSeek does not improve.
Vulkan 0.6.4 on Strix Halo: the coding models moved
I compared Vulkan 0.6.2 and 0.6.4 with ROCm 7.14 on Strix Halo. The new release improved prefill by up to 72%, while decode performance stayed unchanged.
23 coding deployments on Strix Halo: what I would run
I ran 23 local coding-model deployments through 230 executable tasks on a 128GB Strix Halo workstation, comparing accuracy, latency and context limits.
Muse Glimmer 30B on Strix Halo: Vulkan passed the 8K test
Muse Glimmer 30B reaches 42 tok/s with DFlash on ROCm, but an 8K correctness test makes Vulkan the safer Strix Halo text route.
Strix Halo LLM upgrades: what got faster and what failed
Matched before-and-after tests of llama.cpp, vLLM and Lemonade on Strix Halo, plus quantised KV, BF16, MTP, DFlash and ROCmFPX experiments.
SGLang vs vLLM vs llama.cpp on Strix Halo: five models, three winners
Twenty matched SGLang-vLLM points plus llama.cpp and DeepSeek tests show why vLLM wins native Qwen while specialised runtimes win large quants.
DeepSeek V4 Flash on Strix Halo: tuning ROCm, matching Vulkan
Matched DeepSeek V4 Flash tests on Strix Halo show a 44% ROCm prefill gain from tuning, but patched Vulkan still wins all four llama.cpp workloads.
EVO-X3 from BIOS to Lemonade: a reproducible build
A tested route from EVO-X3 firmware settings to a signed Ubuntu USB image, 120 GiB dynamic GTT, SSH-key-only access, Lemonade and verified updates.
DeepSeek V4 Flash 0731 on EVO-X3: the repeat that changed deployment
A measured DeepSeek V4 Flash 0731 study on Strix Halo: ROCm optimisation, a four-hour 32K thermal soak and 20 cold model loads.
GMKtec EVO-X3: the hardware, BIOS and NVMe baseline
A product-led examination of my Linux-based GMKtec EVO-X3: Strix Halo hardware, its provable firmware state and measured NVMe performance.
A local LLM stack worth keeping
The Strix Halo stack I would keep after the first benchmark sequence: hardware, Linux memory, runtimes, models and operating controls.
The long-session test: 122B at a 256K context
A 101-minute, 165-request soak of Qwen3.5 122B at a nearly full 256K context, including throughput, memory, thermals and request-retention fixes.
Finding the useful quant: what fits is not what wins
A measured 0.8B-to-397B Strix Halo model sweep showing why file size, active parameters, context cost and answer quality must be separate decisions.
llama.cpp on Strix Halo: Vulkan versus ROCm
Matched llama.cpp benchmarks show why Strix Halo has no single fastest backend: ROCm wins prompt processing while Vulkan can win generation.
ROCm on Strix Halo, without the folklore
A measured Linux setup for ROCm on the Ryzen AI Max+ 395: memory, build provenance, model lifecycle, failure modes and a recovery runbook.
S3 benchmarking with parallel ranged downloads
Benchmark parallel S3 ranged-download performance with s3bench, a Go tool for measuring throughput against AWS S3 and compatible object stores.
SPDK NVMe-oF target setup on Ubuntu with RDMA
Build SPDK v26.05 and configure an NVMe over Fabrics target on Ubuntu with RDMA, explicit host access and a persistent JSON configuration.
NVMe-oF with RoCE Configuration Guide - Ubuntu Server
Configure a persistent Linux NVMe over Fabrics target with RoCE, stable device identities, host access control and explicit network prerequisites.
Changing an NVMe LBA format safely on Linux
A safety-first guide to inspecting and changing a supported NVMe LBA format, with separate notes on namespace management and secure erase.