How fast does Llama 3.1 8B Instruct run on the Intel Arc Pro B70?

Llama 3.1 8B Instruct runs at about 82.6 tokens/s on the Intel Arc Pro B70 (measured, Q4_K_M, SYCL, llama-bench). Based on 1 measured run(s), each linked to its public source.

Does Llama 3.1 8B Instruct Q4_K_M fit in the Intel Arc Pro B70's memory?

Estimate: the Q4_K_M file is 4.9 GB and needs about 5.9 GB with an ~8K context; the Intel Arc Pro B70 has about 31.5 GB usable, so it fits.

Home › Models › Llama 3.1 8B Instruct › Intel Arc Pro B70

Llama 3.1 8B Instruct on the Intel Arc Pro B70

Llama 3.1 8B Instruct runs at about 82.6 tokens/s on the Intel Arc Pro B70 (measured, Q4_K_M, SYCL, llama-bench).

Best measured generation82.6tok/s

Q4_K_M · SYCL · FA on

Prompt processing2,452tok/s

same run

For comparison, our size-scaling estimate for this combination is 92.9 tok/s.

Every run

Quant	Backend	Generate tok/s	Prompt tok/s	Test	Build / by / notes	Source
Q4_K_M	SYCL FA on	82.6	2,452	tg128	ec6f7a6a5c PMZFX SYCL on one Arc Pro B70, context 4096, KV f16/f16; build b8840-12-gec6f7a6a5-dirty: the author's notes say the local change is a Q2_K DMMV work-in-progress patch	PMZFX intel-arc-pro-b70-benchmarks (MIT)

tg128 = generating 128 tokens; pp512 = reading a 512-token prompt (llama-bench). LocalScore averages several prompt and output lengths, so its numbers run lower than tg128. "decode, llama-server" = single-stream generation speed measured against a running llama-server (no speculative decoding).

Check prices

Amazon Newegg

As an Amazon Associate I earn from qualifying purchases.

No card yet?

Try it on a rented GPU (RunPod)

Affiliate link: if you sign up, RigBench may earn a referral credit. It never affects rankings or numbers.

All results for the Intel Arc Pro B70 · Llama 3.1 8B Instruct on other hardware

Embed this result

HTML

<a href="https://rigbench.pages.dev/model/llama-3-1-8b-instruct/on/intel-arc-pro-b70/"><img src="https://rigbench.pages.dev/badge/llama-3-1-8b-instruct--intel-arc-pro-b70.svg" alt="Llama 3.1 8B Instruct · Intel Arc Pro B70: 82.6 tok/s (Q4_K_M, measured)"></a>

Markdown (GitHub README, forums)

[![Llama 3.1 8B Instruct · Intel Arc Pro B70: 82.6 tok/s (Q4_K_M, measured)](https://rigbench.pages.dev/badge/llama-3-1-8b-instruct--intel-arc-pro-b70.svg)](https://rigbench.pages.dev/model/llama-3-1-8b-instruct/on/intel-arc-pro-b70/)