Arc Pro B70

Model configurations

Each card is a model and hardware setup, including any documented host RAM, storage or CPU requirements. Select a recipe to update its speeds, context and exact command.

Search configurations

Sort byModelHardware setupTotal memory (GB)RecipeContextDecode (tok/s)Prefill (tok/s)Evidence↑ A–Z

15 configurations · 16 recipes

Gemma 4 12B

Arc Pro B7032 GB1 card per machine · 1 machine

llama.cppFormat not recorded

48.1decode tok/s

910prefill tok/s

256Kcontext

Tested on this card · Oct 16, 2026

CommandFull recipe ›

Gemma 4 26B A4B

Arc Pro B7032 GB1 card per machine · 1 machine

llama.cppFormat not recorded

75.7decode tok/s

1,568prefill tok/s

256Kcontext

Tested on this card · Oct 16, 2026

CommandFull recipe ›

Gemma 4 31B

Arc Pro B7032 GB1 card per machine · 1 machine

llama.cppFormat not recorded

22.7decode tok/s

465prefill tok/s

128Kcontext

Tested on this card · Oct 16, 2026

CommandFull recipe ›

Muse Glimmer 30B

Arc Pro B7032 GB1 card per machine · 1 machine

llama.cppFormat not recorded

29.6decode tok/s

915prefill tok/s

128Kcontext

Tested on this card · Oct 16, 2026

CommandFull recipe ›

Nemotron 3.5 Bolt 74B-A9B

Arc Pro B7032 GB1 card per machine · 1 machine

llama.cppFormat not recorded

26.3decode tok/s

603prefill tok/s

256Kcontext

Tested on this card · Oct 17, 2026

CommandFull recipe ›

Ornith-1.5-35B-A3B

Arc Pro B7032 GB1 card per machine · 1 machine

llama.cppFormat not recorded

101.8decode tok/s

1,250prefill tok/s

256Kcontext

Tested on this card · Oct 16, 2026

CommandFull recipe ›

Qwen3.6-27B

Arc Pro B7032 GB1 card per machine · 1 machine

llama.cppFormat not recorded

26decode tok/s

628prefill tok/s

256Kcontext

Tested on this card · Oct 16, 2026

CommandFull recipe ›

Qwen3.6-35B-A3B

Arc Pro B7032 GB1 card per machine · 1 machine

llama.cppFormat not recorded

90.4decode tok/s

1,212prefill tok/s

256Kcontext

Tested on this card · Oct 16, 2026

CommandFull recipe ›

Qwen3.8-27B

Arc Pro B7032 GB1 card per machine · 1 machine

Exact recipellama.cpp · 256K contextvllm · EXL3 4 bpw · 256K context

26.9decode tok/s

653prefill tok/s

256Kcontext

Tested on this card · Oct 16, 2026

CommandFull recipe ›

Qwen3.8-Flash-Next

Arc Pro B7032 GB1 card per machine · 1 machine

Host RAM

≥ 12 GB free

Disk

≥ 132 GB NVMe

sglang.256k.nvme12EXL3 3.05 bpw

29.9decode tok/s

873prefill tok/s

256Kcontext

Tested on this card · Oct 17, 2026

CommandFull recipe ›

Qwen3.8-Flash-Next

Arc Pro B7032 GB1 card per machine · 1 machine

Host RAM

≥ 25 GB free

Disk

≥ 131 GB NVMe

sglang.256k.nvme28EXL3 3.05 bpw

29.6decode tok/s

1,025prefill tok/s

256Kcontext

Tested on this card · Oct 17, 2026

CommandFull recipe ›

Qwen3.8-Flash-Next

Arc Pro B7032 GB1 card per machine · 1 machine

Host RAM

≥ 54 GB free

Disk

≥ 131 GB NVMe

sglang.256k.nvme58EXL3 3.05 bpw

31.8decode tok/s

1,020prefill tok/s

256Kcontext

Tested on this card · Oct 17, 2026

CommandFull recipe ›

Qwen3.8-Flash-Next

Arc Pro B7032 GB1 card per machine · 1 machine

Host RAM

≥ 63 GB free

Disk

≥ 85.14 GB NVMe

SGLangEXL3 3.05 bpw

36.1decode tok/s

1,087prefill tok/s

256Kcontext

Tested on this card · Oct 16, 2026

CommandFull recipe ›

Swift-1.5-Qwen3.8-27B

Arc Pro B7032 GB1 card per machine · 1 machine

llama.cppFormat not recorded

25.6decode tok/s

661prefill tok/s

256Kcontext

Tested on this card · Oct 16, 2026

CommandFull recipe ›

Welp-35B-A3B

Arc Pro B7032 GB1 card per machine · 1 machine

llama.cppFormat not recorded

82.9decode tok/s

1,216prefill tok/s

256Kcontext

Tested on this card · Oct 16, 2026

CommandFull recipe ›

Speeds are the selected recipe’s recorded measurements, not estimates; “—” means unmeasured. Evidence distinguishes checks on this card, sibling-card checks, earlier checks and publisher reports.

Sources: Intel, Intel (2), bhphotovideo, microcenter