readygpu

granite vision 4.1 4b

4B parameters · 128k context · Apache 2.0 · released 2026-04-29

vision

In short

granite vision 4.1 4b is an open model from IBM with 4B parameters. At Q4_K_M, granite vision 4.1 4b needs about 3 GB of memory at 8k context and 5 GB at 32k, counting the weights, the KV cache and runtime buffers. The cheapest hardware in the UK that runs granite vision 4.1 4b fully in memory at 32k context is the NVIDIA Tesla P40, £261 at eBay, at about 34 tok/s (estimated).

Updated . Speeds are estimates; how they are worked out.

What to run it on in the UK

Fully in memory at 32k context, priced at retailers today.

All hardware that runs it

Everything we price, best value first. Two cards split the model by layers.

HardwareQuantFitsDecodePricetok/s per 1,000
NVIDIA Tesla P40Q8_0fits34 tok/s£261130.4
AMD Radeon RX 7600 XTQ8_0fits25 tok/s£34872.0
Intel Arc B580Q8_0fits36 tok/s£349103.9
NVIDIA GeForce RTX 3060 12GBQ8_0fits35 tok/s£39090.8
NVIDIA GeForce RTX 3080 10GBQ8_0fits75 tok/s£466160.4
AMD Radeon RX 9060 XT 16GBQ8_0fits28 tok/s£49955.9
NVIDIA GeForce RTX 5060 Ti 16GBQ8_0fits44 tok/s£55080.2
AMD Radeon RX 7800 XTQ8_0fits54 tok/s£56895.8
Intel Arc A770 16GBQ8_0fits44 tok/s£58875.5
AMD Radeon RX 9070 XTQ8_0fits56 tok/s£67083.3
NVIDIA GeForce RTX 5070Q8_0fits66 tok/s£69595.1
Intel Arc Pro B60Q8_0fits36 tok/s£74248.8
Mac mini M4 16GBQ8_0fits12 tok/s£76516.2
AMD Radeon RX 7900 XTQ8_0fits70 tok/s£78988.5
NVIDIA GeForce RTX 4070 SuperQ8_0fits50 tok/s£87456.8
NVIDIA GeForce RTX 4070 Ti SuperQ8_0fits66 tok/s£95069.6
NVIDIA GeForce RTX 5070 TiQ8_0fits88 tok/s£97590.4
NVIDIA GeForce RTX 4060 Ti 16GBQ8_0fits28 tok/s£97729.0
AMD Radeon RX 7900 XTXQ8_0fits84 tok/s£99983.9
NVIDIA GeForce RTX 5080Q8_0fits94 tok/s£1,19978.7
Mac mini M4 32GBQ8_0fits12 tok/s£1,20010.3
NVIDIA GeForce RTX 4080 SuperQ8_0fits72 tok/s£1,42950.7
AMD Radeon AI PRO R9700Q8_0fits56 tok/s£1,67433.3
NVIDIA GeForce RTX 3090Q8_0fits92 tok/s£1,70754.0
NVIDIA GeForce RTX 3090 TiQ8_0fits99 tok/s£2,19945.1
Mac mini M4 Pro 48GBQ8_0fits28 tok/s£3,0009.4
GMKtec EVO-X2 Ryzen AI Max+ 395 128GBQ8_0fits22 tok/s£3,3206.7
Mac Studio M4 Max 64GBQ8_0fits56 tok/s£3,39916.6
NVIDIA GeForce RTX 5090Q8_0fits176 tok/s£5,00035.2
Mac Studio M3 Ultra 96GBQ8_0fits84 tok/s£5,49915.3
NVIDIA GeForce RTX 4090Q8_0fits99 tok/s£6,01516.5
Mac Studio M3 Ultra 256GBQ8_0fits84 tok/s£8,4999.9
Mac Studio M3 Ultra 512GBQ8_0fits84 tok/s£17,4954.8
NVIDIA RTX PRO 6000 BlackwellQ8_0fits176 tok/s£29,5046.0
Mac Studio M4 Max 128GBQ8_0fits56 tok/s––
Mac mini M4 Pro 64GBQ8_0fits28 tok/s––
Framework Desktop Ryzen AI Max+ 395 128GBQ8_0fits22 tok/s––
NVIDIA DGX Spark 128GBQ8_0fits27 tok/s––
ASUS Ascent GX10 128GBQ8_0fits27 tok/s––
Framework Desktop Ryzen AI Max+ 395 64GBQ8_0fits22 tok/s––

Memory needed

Weights, KV cache and runtime buffers at each context length (KV cache at fp16).

QuantFile@8k@32k@64k@128k
Q4_K_Mgguf2.0 GB3.3 GB5.3 GB8.0 GB13.4 GB
Q5_K_Mgguf2.3 GB3.6 GB5.7 GB8.3 GB13.7 GB
Q8_0gguf3.4 GB4.7 GB6.8 GB9.4 GB14.8 GB

Run it

Copy a command for the runtime you use.

  • Q4_K_M
    • Ollamaollama run hf.co/ibm-granite/granite-vision-4.1-4b-GGUF:Q4_K_M
    • LM Studiolms get ibm-granite/granite-vision-4.1-4b-GGUF@q4_k_m
    • llama.cppllama-server -hf ibm-granite/granite-vision-4.1-4b-GGUF:Q4_K_M
  • Q5_K_M
    • Ollamaollama run hf.co/ibm-granite/granite-vision-4.1-4b-GGUF:Q5_K_M
    • LM Studiolms get ibm-granite/granite-vision-4.1-4b-GGUF@q5_k_m
    • llama.cppllama-server -hf ibm-granite/granite-vision-4.1-4b-GGUF:Q5_K_M
  • Q8_0
    • Ollamaollama run hf.co/ibm-granite/granite-vision-4.1-4b-GGUF:Q8_0
    • LM Studiolms get ibm-granite/granite-vision-4.1-4b-GGUF@q8_0
    • llama.cppllama-server -hf ibm-granite/granite-vision-4.1-4b-GGUF:Q8_0

Notes

Specialised for chart, table and key-value extraction from documents; may not generalise well to open-ended vision tasks and is English-only. Requires transformers>=5.8.0 (or trust_remote_code on older versions) or vLLM>=0.21.0.

Questions

How much VRAM does granite vision 4.1 4b need?
At Q4_K_M, granite vision 4.1 4b needs about 3 GB of memory at 8k context and 5 GB at 32k, counting the weights, the KV cache and runtime buffers.
What is the best GPU for running granite vision 4.1 4b locally?
The cheapest hardware in the UK that runs granite vision 4.1 4b fully in memory at 32k context is the NVIDIA Tesla P40, £261 at eBay, at about 34 tok/s (estimated). The best value is the NVIDIA GeForce RTX 3080 10GB (about 75 tok/s for £466). The fastest we price is the NVIDIA GeForce RTX 5090, at about 176 tok/s.
Can I run granite vision 4.1 4b on a 16 GB GPU?
Yes: at Q8_0 it fits fully in 16 GB at 32k context. On the NVIDIA GeForce RTX 5080 it runs at about 94 tok/s (estimated).
Can I run granite vision 4.1 4b on a 24 GB GPU?
Yes: at Q8_0 it fits fully in 24 GB at 32k context. On the NVIDIA GeForce RTX 3090 Ti it runs at about 99 tok/s (estimated).
How do I run granite vision 4.1 4b locally?
With Ollama: ollama run hf.co/ibm-granite/granite-vision-4.1-4b-GGUF:Q4_K_M (the Q4_K_M build). LM Studio and llama.cpp commands are on this page.
What licence is granite vision 4.1 4b released under?
Apache 2.0, which allows commercial use.