32GB VRAM

Arc Pro B70 32G — what LLMs can it run?

88 of 112 indexed models fit comfortably in 32GB at 4K context, each at the highest-quality quant that still leaves headroom.

The short answer for Arc Pro B70 32G

32 GB, 608 GB/s GDDR6. 88 of 112 models in this index fit comfortably at 4K context; 91 load at all.

Biggest that fits: Nemotron 3.5 Bolt 74B-A9B — 23.1 GB at IQ2_XXS — 9.0 GB spare at 4K, so longer context comes out of a thin margin.
Room to grow: GLM-4.7-Flash — 19.2 GB at Q4_K_M — under 60% of the card, which leaves 12.8 GB for a long context window or a second process.
Speed ceiling: Llama 3.1 8B Instruct at Q4_K_M reads 4.6 GB of weights per token, so 608 GB/s puts a hard ceiling near 132 tok/s. That is arithmetic on two published numbers, not a benchmark — real throughput lands below it. No run on this card has been measured here, so there is nothing to compare the ceiling against.
The wall you will hit: Backend coverage, not capacity. At 608 GB/s the card has the memory system for its size class; what limits it is which runtimes reach it. llama.cpp's SYCL backend lists Arc A-series and B580 cards as verified, and Ollama runs them through Vulkan — so GGUF is the format this site recommends here. AWQ, GPTQ and EXL2 are not.

Best local LLM for 32GB

70B+ · 1

Arc Pro B70 32G — what LLMs can it run? — 70B+
Model	Quant	Est. VRAM	Headroom	tok/s
Nemotron 3.5 Bolt 74B-A9B74B-A9B	IQ2_XXS	23.05 GB	+9 GB	≤239Ceiling

32B · 29

Arc Pro B70 32G — what LLMs can it run? — 32B
Model	Quant	Est. VRAM	Headroom	tok/s
Seed-OSS 36B Instruct36B	Q5_K_M	27.81 GB	+4.2 GB	≤25Ceiling
Qwen3.6 35B-A3B35B-A3B	Q4_K_M	23.38 GB	+8.6 GB	≤341Ceiling
Command R 35B35B	Q4_K_M	22.86 GB	+9.1 GB	≤30Ceiling
Yi 1.5 34B Chat34B	Q4_K_M	22.82 GB	+9.2 GB	≤31Ceiling
Qwen3 32B Instruct32B	Q4_K_M	21.85 GB	+10.1 GB	21–28Derived
Qwen2.5 32B Instruct32B	Q4_K_M	21.69 GB	+10.3 GB	22–28Derived
Qwen2.5-Coder 32B Instruct32B	Q4_K_M	21.69 GB	+10.3 GB	22–28Derived
DeepSeek-R1-Distill-Qwen-32B32B	Q4_K_M	21.69 GB	+10.3 GB	22–28Derived
Nemotron 3.5 Lightning 30B-A3B30B-A3B	Q4_0	19.79 GB	+12.2 GB	≤356Ceiling
Gemma 4 31B IT31B	Q4_K_M	21.08 GB	+10.9 GB	23–30Derived
Qwen3 30B-A3B Instruct30B-A3B	Q5_K_M	23.04 GB	+9 GB	≤301Ceiling
Qwen3-Coder 30B-A3B Instruct30B-A3B	Q5_K_M	23.04 GB	+9 GB	≤301Ceiling
Qwen3-VL 30B-A3B Instruct30B-A3B	Q4_K_M	19.73 GB	+12.3 GB	≤352Ceiling
North Mini Code 1.0 30B-A3B30B-A3B	Q4_K_M	20.01 GB	+12 GB	≤347Ceiling
GLM-4.7-Flash30B-A3B	Q4_K_M	19.19 GB	+12.8 GB	≤352Ceiling
Granite 4.2 30B30B	Q6_K	26.19 GB	+5.8 GB	18–23Derived
Muse Glimmer 30B30B	Q6_K	24.61 GB	+7.4 GB	18–24Derived
Qwen3.8 27B27B	Q4_K_M	17.47 GB	+14.5 GB	26–34Derived
Qwen3.6 27B27B	Q5_K_M	22.19 GB	+9.8 GB	20–27Derived
Gemma 3 27B IT27B	Q4_K_M	18.37 GB	+13.6 GB	26–34Derived
Gemma 2 27B Instruct27B	Q5_K_M	21.76 GB	+10.2 GB	22–29Derived
Gemma 4 26B-A4B IT26B-A4B	Q4_K_M	16.39 GB	+15.6 GB	≤264Ceiling
Mistral Small 24B Instruct24B	Q4_K_M	15.89 GB	+16.1 GB	29–39Derived
LFM2.5 24B-A2B24B-A2B	Q8_0	26.73 GB	+5.3 GB	≤246Ceiling
Devstral Small 1.1 24B24B	Q6_K	20.91 GB	+11.1 GB	22–29Derived
Magistral Small 1.2 24B24B	Q6_K	20.91 GB	+11.1 GB	22–29Derived
Codestral 22B22B	Q4_K_M	15.03 GB	+17 GB	32–42Derived
ERNIE 4.5 21B-A3B21B-A3B	Q8_0	24.44 GB	+7.6 GB	≤201Ceiling
GPT-OSS 20B21B MoE	MXFP4	12.76 GB	+19.2 GB	≤307Ceiling

14B · 19

Arc Pro B70 32G — what LLMs can it run? — 14B
Model	Quant	Est. VRAM	Headroom	tok/s
InternLM2 20B Chat20B	Q5_K_M	15.59 GB	+16.4 GB	30–40Derived
DeepSeek-Coder-V2-Lite Instruct16B-A2.4B	Q8_0	17.56 GB	+14.4 GB	≤251Ceiling
DeepSeek-V2-Lite Chat16B-A2.4B	Q4_K_M	10.08 GB	+21.9 GB	≤440Ceiling
StarCoder2 15B15B	Q4_K_M	10.19 GB	+21.8 GB	47–62Derived
Qwen3 14B Instruct14B	Q5_K_M	11.65 GB	+20.4 GB	40–54Derived
Qwen2.5 14B Instruct14B	Q5_K_M	11.73 GB	+20.3 GB	41–54Derived
DeepSeek-R1-Distill-Qwen-14B14B	Q4_K_M	10.14 GB	+21.9 GB	48–63Derived
Phi-4 14B14B	Q5_K_M	11.77 GB	+20.2 GB	41–54Derived
Phi-3 Medium 14B Instruct14B	Q6_K	12.86 GB	+19.1 GB	37–49Derived
Mistral Nemo 12B Instruct12B	Q6_K	11.14 GB	+20.9 GB	42–56Derived
Gemma 3 12B IT12B	Q5_K_M	9.84 GB	+22.2 GB	49–65Derived
Mellum2.1 12B-A2.5B Thinking12B-A2.5B	Q8_0	13.6 GB	+18.4 GB	≤241Ceiling
Stable LM 2 12B Chat12B	Q4_K_M	8.35 GB	+23.7 GB	58–77Derived
Gemma 4 12B IT12B	Q8_0	13.82 GB	+18.2 GB	33–44Derived
Llama 3.2 11B Vision Instruct11B	Q8_0	12.9 GB	+19.1 GB	36–48Derived
Solar 10.7B Instruct11B	Q4_K_M	7.6 GB	+24.4 GB	65–87Derived
Falcon 3 10B Instruct10B	Q4_K_M	7.28 GB	+24.7 GB	67–89Derived
Gemma 2 9B Instruct9B	Q8_0	11.7 GB	+20.3 GB	43–57Derived
Qwen3.5 9B9B	Q8_0	10.37 GB	+21.6 GB	43–57Derived

7B · 25

Arc Pro B70 32G — what LLMs can it run? — 7B
Model	Quant	Est. VRAM	Headroom	tok/s
GLM-4-9B-Chat9B	Q8_0	10.16 GB	+21.8 GB	44–59Derived
Qwen3-VL 8B Instruct8B	Q8_0	10.39 GB	+21.6 GB	45–60Derived
Granite 4.2 8B8B	Q8_0	10.45 GB	+21.6 GB	45–60Derived
Ministral 3 8B Instruct8B	Q8_0	10.02 GB	+22 GB	47–62Derived
Qwen2-VL 7B Instruct7B	Q4_K_M	5.49 GB	+26.5 GB	84–112Derived
Granite 3.1 8B Instruct8B	Q4_K_M	5.88 GB	+26.1 GB	85–113Derived
Qwen3 8B Instruct8B	Q6_K	7.64 GB	+24.4 GB	63–84Derived
Llama 3.1 8B Instruct8B	Q8_0	9.47 GB	+22.5 GB	50–66Derived
Nous Hermes 3 Llama 3.1 8B8B	Q4_K_M	5.64 GB	+26.4 GB	87–115Derived
Aya 23 8B8B	Q4_K_M	5.64 GB	+26.4 GB	87–115Derived
OpenChat 3.6 8B8B	Q4_K_M	5.64 GB	+26.4 GB	87–115Derived
DeepSeek-R1-Distill-Llama-8B8B	Q5_K_M	6.51 GB	+25.5 GB	74–98Derived
InternLM2 7B Chat7B	Q4_K_M	5.45 GB	+26.6 GB	90–120Derived
Qwen2.5 7B Instruct7B	Q6_K	6.77 GB	+25.2 GB	68–90Derived
Qwen2.5-Coder 7B Instruct7B	Q4_K_M	5.07 GB	+26.9 GB	92–121Derived
DeepSeek-R1-Distill-Qwen-7B7B	Q4_K_M	5.07 GB	+26.9 GB	92–121Derived
Gemma 4 E4B ITE4B	Q8_0	8.46 GB	+23.5 GB	53–70Derived
OLMo 2 7B Instruct7B	Q8_0	10.3 GB	+21.7 GB	55–72Derived
OLMo 3 7B Instruct7B	Q8_0	10.3 GB	+21.7 GB	55–72Derived
WizardLM-2 7B7B	Q4_K_M	5.14 GB	+26.9 GB	97–128Derived
Mistral 7B Instruct v0.37B	Q6_K	6.75 GB	+25.3 GB	71–95Derived
Zephyr 7B Beta7B	Q6_K	6.75 GB	+25.3 GB	71–95Derived
Gemma 4 E2B ITE2B	Q8_0	5.2 GB	+26.8 GB	≤130Ceiling
Qwen3.5 4B4B	Q8_0	4.97 GB	+27 GB	≤138Ceiling
Gemma 3 4B IT4B	Q8_0	5.05 GB	+27 GB	≤140Ceiling

≤3B · 14

Arc Pro B70 32G — what LLMs can it run? — ≤3B
Model	Quant	Est. VRAM	Headroom	tok/s
Qwen3 4B Instruct4B	Q6_K	4.05 GB	+28 GB	≤195Ceiling
Phi-4 Mini Instruct3.8B	Q8_0	4.81 GB	+27.2 GB	≤157Ceiling
Phi-3.5 Mini Instruct3.8B	Q8_0	5.89 GB	+26.1 GB	≤158Ceiling
Granite 4.2 3B3B	Q8_0	4.41 GB	+27.6 GB	≤165Ceiling
Llama 3.2 3B Instruct3B	Q8_0	4.05 GB	+28 GB	≤188Ceiling
Qwen2.5 3B Instruct3B	Q8_0	3.59 GB	+28.4 GB	≤195Ceiling
SmolLM3 3B3B	Q8_0	3.73 GB	+28.3 GB	≤195Ceiling
Gemma 2 2B Instruct2B	Q8_0	3.34 GB	+28.7 GB	≤231Ceiling
Qwen3.5 2B2B	Q8_0	2.23 GB	+29.8 GB	≤307Ceiling
Qwen3 1.7B Instruct1.7B	Q8_0	2.39 GB	+29.6 GB	≤349Ceiling
Qwen2.5 1.5B Instruct1.5B	Q8_0	1.83 GB	+30.2 GB	≤392Ceiling
Llama 3.2 1B Instruct1B	Q8_0	1.51 GB	+30.5 GB	≤486Ceiling
Qwen3.5 0.8B0.8B	Q8_0	0.94 GB	+31.1 GB	≤751Ceiling
Qwen2.5 0.5B Instruct0.5B	Q8_0	0.6 GB	+31.4 GB	≤1241Ceiling

How this list is built

Each row is the lowest-perplexity-loss quant of that model whose estimated total — weights plus KV cache at 4K context plus activation buffer — uses at most 88% of the card. That is the calculator's "green" threshold, so every row here has real headroom rather than only just fitting. Raise the context length and the list shortens; the calculator lets you check any combination directly. The speed column is this card: a measured run, a range scaled from this card's own Llama 2 7B benchmark, or a bandwidth ceiling. A figure that only exists for the RTX 4090 is labelled as that card.

A further 3 models load but with no headroom to spare (up to 105% of VRAM) — the Quant Hub’s GPU chips count those too, which is why its number is higher.

Open this card in the VRAM calculator Feel this speed Browse all 112 models

Measured on this card

No benchmark runs in this index were recorded on a Arc Pro B70 32G. Every figure on this page is calculated from the model architecture and the quant level — treat them as estimates, not measurements.

Cards with the same budget

The same 32GB in system memory is a different machine — everything on this page fits there, at a speed set by your DIMMs rather than a graphics bus: 32 GB RAM (CPU)

Setup guides for this card

Step-by-step guides written for the Arc Pro B70 32G's runtime or memory size, with the commands and the checks that prove the model is running on the card.

Intel Arc: llama.cpp (SYCL) or Ollama (Vulkan)

Common questions

What is the best local LLM for an Arc Pro B70 32G?

For everyday use, GLM-4.7-Flash at Q4_K_M — about 19.2 GB of the card's 32 GB at 4K context, leaving 12.8 GB for a longer window. If you want the largest thing that will load, that is Nemotron 3.5 Bolt 74B-A9B at IQ2_XXS (23.1 GB, 9.0 GB spare). "Best" here means best fit for the memory budget — this index does not run task benchmarks, so it cannot tell you which model is smarter.

Can an Arc Pro B70 32G run Llama 3.2 90B Vision Instruct?

No. Its smallest build here, Q3_K_M, needs about 46.2 GB and this card has 32.0 GB usable — short by 14.2 GB before any context beyond 4K. The largest model this card does clear is Nemotron 3.5 Bolt 74B-A9B.

How many tokens per second does an Arc Pro B70 32G do on an 8B model at Q4?

This index has no measured run on this card, so it does not publish a figure. What can be stated from specifications: generating a token requires reading every weight once, Llama 3.1 8B Instruct at Q4_K_M is 4.6 GB of weights, and this card moves 608 GB/s — a ceiling near 132 tok/s. Batch size, context length, the runtime and how much of the model sits in cache all take you below it.

Arc Pro B70 32G or RTX 5090 for local LLMs?

They do not hold the same set: Arc Pro B70 32G fits 88 and RTX 5090 90 of 112 at 4K — the cards load different formats. Speed is the other difference — 608 GB/s against 1,792 GB/s, a 2.9× gap in how fast the weights can be read, which is what token generation is bound by. The RTX 5090 generates faster on any model both can hold.

This page's figures change when the model or the runtime does.Last updated 2026-10-20 RSS → /feed.xml