Candidate package · llama.cpp SYCL · 1× Intel Arc Pro B70

Qwen3.8 27B 256K + vision + MTP draft on one Intel Arc Pro B70

Alibaba's Qwen3.8 27B on one Arc Pro B70 with its full 262K-token context, image input, and draft head all resident at once. A capability-first fit, not a speed record.

What we measured Lab-measured

26.7tok/s decode (MTP-assisted, full package resident)

Conventional 99-interval median, cold 12-prompt suite, up to 512-token responses, cache-zero, at 262144 configured capacity with q8_0 K/V, vision mmproj, and MTP draft loaded; active prompts were 48–78 tokens and the second fresh-server run measured 26.640510. Speculation-assisted and labeled as such; target-only tg128 at depth 0 is 22.64 tok/s (raw engine). proof file

target-only = no draft model assisting · cache-zero = a fresh start, nothing pre-computed · 99-interval median = the middle rate across 99 measured stretches of a long answer · tg128 = a raw 128-token generation benchmark, not the headline suite · MTP = multi-token prediction, a small draft the main model verifies

Model
Qwen3.8 27B flagship package
Publisher
Alibaba / Qwen (GGUF by unsloth)
Checkpoint
unsloth/Qwen3.8-27B-GGUF
Compression
UD-Q5_K_S (shipped) / UD-Q4_K_XL (alternative)
Software
llama.cpp SYCL
Cards
1× Intel Arc Pro B70 32 GiB
Model weight bytes
18.7 GB
Operating systems
Linux
Delivery
native
Good for
long context vision general coding
Published
2026-08-22
Clean-host replay
not yet

Still missing before this becomes an install guide

  • tested clean-host platform installation
  • broader multimodal quality beyond the objective smoke
  • deep-context validation beyond canaries on the multi-chunk prefill path

What to know

  • 256K KV requires q8_0 K/V on one card; f16 KV limits context to roughly 96K at these quants.
  • MTP-assisted rate varies with content (p10 24.08 vs median 26.67); target-only numbers published alongside.
  • Vision verified by objective two-question smoke (dominant color, corner color) on a deterministic test image; broader multimodal quality untested.
  • Above ~1024-token prompts this package relies on multi-chunk prefill paths that the lab's separate vLLM lane found corruption-prone; the llama.cpp path here passed its canaries but has not had equivalent deep validation at depth.

Measured performance profiles Lab-measured

Raw decode over existing context depth Lab-measured

21.019.817.614.111.910.32K4K8K16K24K32KExisting context depth before tg128 · y: tok/s
llama-bench raw engine rates, TARGET-ONLY Q5_K_S with q8_0 KV (no draft; llama-bench has no speculation). The MTP-assisted serving rate above is the package headline. The directly measured zero-depth point remains in the linked raw evidence; no missing depth is interpolated. evidence

Raw pp2048 over existing context depth Lab-measured

8318228087607516792K4K8K16K24K32KExisting context depth before pp2048 · y: tok/s
llama-bench raw engine rates, TARGET-ONLY Q5_K_S with q8_0 KV (no draft; llama-bench has no speculation). The MTP-assisted serving rate above is the package headline. The directly measured zero-depth point remains in the linked raw evidence; no missing depth is interpolated. evidence

Many people at once Not published

No qualified multi-user aggregate profile is published for this exact package. Diagnostic or unsupported boundaries may still appear under “What to know” or in the full guide. Nothing is interpolated or promoted from a different model, quantization, runtime, or card count.

How much faster could this get? No projection

This package's measured workload does not map cleanly onto a single model + compression + card-count shape, so no like-for-like projection is shown. The measured numbers above stand on their own.