Skip to content
32GB VRAM

Radeon AI PRO R9700 32G — what LLMs can it run?

125 of 151 indexed models fit comfortably in 32GB at 4K context, each at the highest-quality quant that still leaves headroom.

Context length
Concurrency (batch)

The short answer for Radeon AI PRO R9700 32G

32 GB, 640 GB/s GDDR6. 125 of 151 models in this index fit comfortably at 4K context; 126 load at all.

Biggest that fits
Kolibri-1 78B-A3.5B — 25.6 GB at IQ2_XS — 6.4 GB spare at 4K, so longer context comes out of a thin margin.Download: bartowski/Kolibri-1-GGUF · Kolibri-1-IQ2_XS.gguf · 24.34 GB · listing read 2026-11-12
Room to grow
Gemma 3 27B IT — 18.3 GB at Q4_K_M — under 60% of the card, which leaves 13.7 GB for a long context window or a second process.Download: bartowski/google_gemma-3-27b-it-GGUF · google_gemma-3-27b-it-Q4_K_M.gguf · 16.55 GB · + vision projector 0.86 GB, separate · listing read 2026-11-03
Speed ceiling
Llama 3.1 8B Instruct at Q4_K_M reads 4.6 GB of weights per token, so 640 GB/s puts a hard ceiling near 139 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 640 GB/s the memory system is competitive with NVIDIA cards of the same size; what varies is whether your runtime has a working ROCm build for your kernel and card. llama.cpp and vLLM both ship official ROCm support — this is a setup question, not a hardware ceiling.
Best local LLM for 32GB

50B+ · 4

Radeon AI PRO R9700 32G — what LLMs can it run? — 50B+
ModelQuantEst. VRAMHeadroomtok/s
Kolibri-1 78B-A3.5B78B-A3.5BPair fit detailsIQ2_XS25.59 GB+6.4 GB≤618Ceiling
Nemotron 3.5 Bolt 74B-A9B74B-A9BPair fit detailsIQ2_XXS23.05 GB+9 GB≤252Ceiling
Ripple-65B-Preview65BPair fit detailsIQ2_S27.18 GB+4.8 GB≤27Ceiling
Jamba 1.5 Mini52B-A12BAWQ INT427.24 GB+4.8 GB≤112Ceiling

20–50B · 43

Radeon AI PRO R9700 32G — what LLMs can it run? — 20–50B
ModelQuantEst. VRAMHeadroomtok/s
Mixtral 8x7B Instruct47B MoEAWQ INT424.95 GB+7.1 GB≤104Ceiling
Recursive Flint 40B-A2.5B40B-A2.5BPair fit detailsQ4_K_M26.68 GB+5.3 GB≤424Ceiling
K2 Horizon MoVA 36B-A4B36B-A4BPair fit detailsQ4_K_M24.21 GB+7.8 GB≤282Ceiling
Seed-OSS 36B Instruct36BPair fit detailsQ5_K_M27.83 GB+4.2 GB≤26Ceiling
Qwen3.6 35B-A3B35B-A3BPair fit detailsQ4_K_M23.38 GB+8.6 GB≤358Ceiling
Apex 35B V1.035B-A3BPair fit detailsQ4_K_M23.91 GB+8.1 GB≤350Ceiling
Command R 35B35BQ4_K_M22.86 GB+9.1 GB≤32Ceiling
K2 Horizon 32B32BPair fit detailsQ5_K_M26.91 GB+5.1 GB≤27Ceiling
MiniCPM-V 4.7 35B-A3B35B-A3BPair fit detailsQ5_K_M25.91 GB+6.1 GB≤315Ceiling
Clover-35B-A3B-Preview35B-A3BPair fit detailsQ4_K_M23.4 GB+8.6 GB≤349Ceiling
Yi 1.5 34B Chat34BQ4_K_M22.82 GB+9.2 GB≤32Ceiling
Nanbeige4.3 33B33BPair fit detailsQ4_K_M22.61 GB+9.4 GB≤32Ceiling
Qwen3 32B Instruct32BQ4_K_M21.85 GB+10.1 GB18–24Derived
Qwen2.5 32B Instruct32BQ4_K_M21.69 GB+10.3 GB18–24Derived
Qwen2.5-Coder 32B Instruct32BQ4_K_M21.69 GB+10.3 GB18–24Derived
DeepSeek-R1-Distill-Qwen-32B32BQ4_K_M21.69 GB+10.3 GB18–24Derived
Nemotron 3.5 Lightning 30B-A3B30B-A3BPair fit detailsQ4_019.79 GB+12.2 GB≤375Ceiling
Xing4.0 29B-A4B29B-A4BPair fit detailsQ5_K_M25.28 GB+6.7 GB≤219Ceiling
Gemma 4 31B IT31BPair fit detailsQ5_K_M25.25 GB+6.8 GB16–21Derived
Qwen3-Coder 30B-A3B Instruct30B-A3BPair fit detailsQ5_K_M23.12 GB+8.9 GB≤316Ceiling
Qwen3-VL 30B-A3B Instruct30B-A3BPair fit detailsQ4_K_M19.89 GB+12.1 GB≤368Ceiling
Qwen3 30B-A3B Instruct30B-A3BQ5_K_M23.04 GB+9 GB≤316Ceiling
North Mini Code 1.0 30B-A3B30B-A3BPair fit detailsQ4_K_M20.01 GB+12 GB≤365Ceiling
Hy-Memory-30B-A3B30B-A3BPair fit detailsQ6_K26.21 GB+5.8 GB≤273Ceiling
GLM-4.7-Flash30B-A3BPair fit detailsQ4_K_M19.51 GB+12.5 GB≤364Ceiling
Granite 4.2 30B30BPair fit detailsQ6_K26.19 GB+5.8 GB15–20Derived
Muse Glimmer 30B30BPair fit detailsQ6_K24.61 GB+7.4 GB15–20Derived
Qwen3.8 27B27BPair fit detailsQ6_K25.2 GB+6.8 GB15–20Derived
Arete 27B27BPair fit detailsQ5_K_M21.89 GB+10.1 GB17–23Derived
Qwen3.6 27B27BPair fit detailsQ5_K_M22.19 GB+9.8 GB17–22Derived
DaisyRose 27B27BPair fit detailsQ5_K_M22 GB+10 GB17–23Derived
Gemma 2 27B Instruct27BQ5_K_M21.76 GB+10.2 GB18–24Derived
Gemma 3 27B IT27BPair fit detailsQ4_K_M18.3 GB+13.7 GB21–28Derived
Ternary Bonsai 2 27B27BPair fit detailsQ2_08.24 GB+23.8 GB47–62Derived
Clairic 27B27BPair fit detailsQ5_K_M20.51 GB+11.5 GB18–24Derived
Gemma 4 26B-A4B IT26B-A4BPair fit detailsQ4_K_M18.22 GB+13.8 GB≤250Ceiling
Mistral Small 24B Instruct24BQ4_K_M15.89 GB+16.1 GB24–32Derived
LFM2.5 24B-A2B24B-A2BPair fit detailsQ8_026.73 GB+5.3 GB≤259Ceiling
Devstral Small 1.1 24B24BPair fit detailsQ6_K20.89 GB+11.1 GB18–24Derived
Magistral Small 1.2 24B24BPair fit detailsQ6_K20.89 GB+11.1 GB18–24Derived
Codestral 22B22BQ4_K_M15.03 GB+17 GB26–35Derived
ERNIE 4.5 21B-A3B21B-A3BPair fit detailsQ8_024.48 GB+7.5 GB≤211Ceiling
GPT-OSS 20B21B MoEMXFP412.76 GB+19.2 GB≤323Ceiling

10–20B · 19

Radeon AI PRO R9700 32G — what LLMs can it run? — 10–20B
ModelQuantEst. VRAMHeadroomtok/s
InternLM2 20B Chat20BQ5_K_M15.59 GB+16.4 GB25–33Derived
DeepSeek-Coder-V2-Lite Instruct16B-A2.4BQ8_017.56 GB+14.4 GB≤264Ceiling
DeepSeek-V2-Lite Chat16B-A2.4BQ4_K_M10.08 GB+21.9 GB≤463Ceiling
StarCoder2 15B15BQ4_K_M9.85 GB+22.2 GB39–52Derived
Qwen3 14B Instruct14BQ5_K_M11.65 GB+20.4 GB34–45Derived
Qwen2.5 14B Instruct14BQ5_K_M11.73 GB+20.3 GB34–45Derived
DeepSeek-R1-Distill-Qwen-14B14BQ4_K_M10.14 GB+21.9 GB40–53Derived
Phi-4 14B14BQ5_K_M11.77 GB+20.2 GB34–45Derived
Phi-3 Medium 14B Instruct14BQ6_K12.86 GB+19.1 GB31–41Derived
Ministral 3 14B Instruct14BPair fit detailsQ8_015.7 GB+16.3 GB25–33Derived
Mistral Nemo 12B Instruct12BQ6_K11.14 GB+20.9 GB36–47Derived
Gemma 3 12B IT12BQ5_K_M9.84 GB+22.2 GB41–54Derived
Mellum2.1 12B-A2.5B Thinking12B-A2.5BQ8_013.6 GB+18.4 GB≤254Ceiling
Stable LM 2 12B Chat12BQ4_K_M8.52 GB+23.5 GB48–64Derived
Gemma 4 12B IT12BPair fit detailsQ8_013.82 GB+18.2 GB28–37Derived
Llama 3.2 11B Vision Instruct11BQ8_012.9 GB+19.1 GB30–40Derived
Solar 10.7B Instruct11BQ4_K_M7.6 GB+24.4 GB55–73Derived
Falcon 3 10B Instruct10BQ4_K_M7.28 GB+24.7 GB56–75Derived
Metis-1.6 Logos10B MoEPair fit detailsQ8_011.67 GB+20.3 GB≤242Ceiling

4–10B · 37

Radeon AI PRO R9700 32G — what LLMs can it run? — 4–10B
ModelQuantEst. VRAMHeadroomtok/s
Gemma 2 9B Instruct9BQ8_011.7 GB+20.3 GB36–48Derived
Qwen3.5 9B9BPair fit detailsQ8_010.39 GB+21.6 GB36–48Derived
K2 Horizon 7B7BPair fit detailsQ8_010.62 GB+21.4 GB37–49Derived
GLM-4-9B-Chat9BQ8_010.16 GB+21.8 GB37–49Derived
MiMo-V2.6-Distill-Qwen-9B9BPair fit detailsQ8_010.11 GB+21.9 GB37–49Derived
Clairic Flash 9B9BPair fit detailsQ8_010.1 GB+21.9 GB37–49Derived
Qwen3-VL 8B Instruct8BQ8_010.39 GB+21.6 GB38–50Derived
TwIL 8B balanced8.8BQ4_K_M6.28 GB+25.7 GB66–88Derived
Granite 4.2 8B8BPair fit detailsQ8_010.46 GB+21.5 GB38–50Derived
Ministral 3 8B Instruct8BQ8_010.02 GB+22 GB39–52Derived
LFM2.5 8B-A1B8B-A1.5BPair fit detailsQ8_09.46 GB+22.5 GB≤422Ceiling
Qwen2-VL 7B Instruct7BQ4_K_M5.49 GB+26.5 GB71–94Derived
Granite 3.1 8B Instruct8BQ4_K_M5.88 GB+26.1 GB72–95Derived
Shot Reader 8B8BPair fit detailsQ8_09.72 GB+22.3 GB41–54Derived
Qwen3 8B Instruct8BQ6_K7.64 GB+24.4 GB53–70Derived
Llama 3.1 8B Instruct8BQ8_09.47 GB+22.5 GB42–55Derived
Nous Hermes 3 Llama 3.1 8B8BQ4_K_M5.64 GB+26.4 GB73–97Derived
Aya 23 8B8BQ4_K_M5.64 GB+26.4 GB73–97Derived
OpenChat 3.6 8B8BQ4_K_M5.64 GB+26.4 GB73–97Derived
DeepSeek-R1-Distill-Llama-8B8BQ5_K_M6.51 GB+25.5 GB62–82Derived
InternLM2 7B Chat7BQ4_K_M5.45 GB+26.6 GB76–100Derived
Qwen2.5 7B Instruct7BQ6_K6.77 GB+25.2 GB57–75Derived
Qwen2.5-Coder 7B Instruct7BQ4_K_M5.07 GB+26.9 GB77–102Derived
DeepSeek-R1-Distill-Qwen-7B7BQ4_K_M5.07 GB+26.9 GB77–102Derived
Gemma 4 E4B ITE4BPair fit detailsQ8_08.68 GB+23.3 GB43–57Derived
Nemotron 3.5 Bolt E4BE4BPair fit detailsQ8_08.5 GB+23.5 GB44–59Derived
OLMo 2 7B Instruct7BQ8_010.3 GB+21.7 GB46–61Derived
OLMo 3 7B Instruct7BQ8_010.3 GB+21.7 GB46–61Derived
WizardLM-2 7B7BQ4_K_M5.14 GB+26.9 GB81–107Derived
Mistral 7B Instruct v0.37BQ6_K6.75 GB+25.3 GB60–79Derived
Zephyr 7B Beta7BQ6_K6.75 GB+25.3 GB60–79Derived
K2 Horizon 3.7B3.7BQ8_06.24 GB+25.8 GB≤125Ceiling
Gemma 4 E2B ITE2BQ8_05.2 GB+26.8 GB≤136Ceiling
Qwen3.5 4B4BQ8_04.97 GB+27 GB≤146Ceiling
Gemma 3 4B IT4BQ8_05.05 GB+27 GB≤147Ceiling
Spark-X2.6 4B4BQ8_04.84 GB+27.2 GB≤154Ceiling
Qwen3 4B Instruct4BQ6_K4.05 GB+28 GB≤205Ceiling

<4B · 22

Radeon AI PRO R9700 32G — what LLMs can it run? — <4B
ModelQuantEst. VRAMHeadroomtok/s
Phi-4 Mini Instruct3.8BQ8_04.81 GB+27.2 GB≤165Ceiling
Phi-3.5 Mini Instruct3.8BQ8_05.89 GB+26.1 GB≤166Ceiling
Granite 4.2 3B3BQ8_04.41 GB+27.6 GB≤173Ceiling
Ministral 3 3B Instruct3BQ8_04.26 GB+27.7 GB≤184Ceiling
Llama 3.2 3B Instruct3BQ8_04.05 GB+28 GB≤198Ceiling
Qwen2.5 3B Instruct3BQ8_03.59 GB+28.4 GB≤205Ceiling
SmolLM3 3B3BQ8_03.73 GB+28.3 GB≤206Ceiling
Metis-1.6 Praxis3B MoEQ8_03.4 GB+28.6 GB≤566Ceiling
Gemma 2 2B Instruct2BQ8_03.34 GB+28.7 GB≤243Ceiling
Qwen3.5 2B2BQ8_02.23 GB+29.8 GB≤323Ceiling
LFM3 1.7B1.7BQ8_01.99 GB+30 GB≤366Ceiling
LFM3-VL 2B2BQ8_01.99 GB+30 GB≤366Ceiling
Qwen3 1.7B Instruct1.7BQ8_02.39 GB+29.6 GB≤368Ceiling
Qwen2.5 1.5B Instruct1.5BQ8_01.83 GB+30.2 GB≤413Ceiling
Llama 3.2 1B Instruct1BQ8_01.51 GB+30.5 GB≤512Ceiling
LFM2.5 1.2B Instruct1.2BQ8_01.35 GB+30.7 GB≤542Ceiling
K2 Horizon 0.9B0.9BQ8_01.44 GB+30.6 GB≤587Ceiling
Qwen3.5 0.8B0.8BQ8_00.94 GB+31.1 GB≤790Ceiling
BotPsy 0.8B0.75BQ8_00.9 GB+31.1 GB≤831Ceiling
Qwen2.5 0.5B Instruct0.5BQ8_00.6 GB+31.4 GB≤1306Ceiling
MNN-Pocket 0.3B Instruct0.36BQ8_00.46 GB+31.5 GB≤1730Ceiling
MNN-Pocket 0.3B QAT0.30BQ4_10.26 GB+31.7 GB≤3556Ceiling

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 each estimate leaves headroom. Raise the context length and the list shortens; the calculator lets you check any combination directly. Speed figures are labelled as measurements, estimates scaled from this card's Llama 2 7B benchmark, or theoretical bandwidth ceilings. A figure that only exists for the RTX 4090 is labelled as that card.

A further 1 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.

Measured on this card

No benchmark runs in this index were recorded on a Radeon AI PRO R9700 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

What fits is decided by memory, so every 32GB card of this type returns the same list. These pages are not different answers — they differ in throughput, which this index does not measure per card.

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)

Same list, different speed — play one sentence on both cards: Radeon AI PRO R9700 32G vs Instinct MI100 32G (Qwen3 32B Instruct)

Setup guides for this card

Step-by-step guides written for the Radeon AI PRO R9700 32G's runtime or memory size, with the commands and the checks that prove the model is running on the card.

Stepping up

A Radeon PRO W7900 48G (48GB) fits 6 more of the indexed models than this card. Radeon PRO W7900 48G →

Before buying, hear the speed difference on a model both cards hold: Radeon AI PRO R9700 32G vs Radeon PRO W7900 48G (Qwen3 32B Instruct) →

Common questions

What is the best local LLM for a Radeon AI PRO R9700 32G?

For everyday use, Gemma 3 27B IT at Q4_K_M — about 18.3 GB of the card's 32 GB at 4K context, leaving 13.7 GB for a longer window. If you want the largest thing that will load, that is Kolibri-1 78B-A3.5B at IQ2_XS (25.6 GB, 6.4 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 a Radeon AI PRO R9700 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 Kolibri-1 78B-A3.5B.

How many tokens per second does a Radeon AI PRO R9700 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 640 GB/s — a ceiling near 139 tok/s. Batch size, context length, the runtime and how much of the model sits in cache all take you below it.

Radeon AI PRO R9700 32G or RTX 5090 for local LLMs?

They hold the same models: 32 GB against 32 GB fits 125 and 125 of 151 respectively at 4K. The difference is throughput — 640 GB/s against 1,792 GB/s, a 2.8× 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.

125 of 151 indexed models fit comfortably in 32GB at 4K context, each at the highest-quality quant that still leaves headroom.

This page's figures change when the model or the runtime does.Last updated 2026-11-22 RSS → /feed.xml