IBM

Granite Code 34B

IBM·34B·Granite·8K ctx·2026-08

Check on my hardwareAdd to compare →Ollama →Hugging Face →

Model size→Hardware requirement→Actual fit

Granite Code 34B needs ~20.2–37.5 GB of VRAM depending on quantization. The bar is what it needs; the marker is the card's VRAM. Speed figures below are planning estimates.

8 GB

RTX 4060 · 3060 Ti

20.2
8
Q4_K_M · Est. ~3 tok/sSlow · CPU offload

12 GB

RTX 4070 · 3060 12G

20.2
12
Q4_K_M · Est. ~5 tok/sSlow · CPU offload

16 GB

RTX 4080 · M-series 16G

20.2
16
Q4_K_M · Est. ~7 tok/sSlow · CPU offload

24 GB

RTX 4090 · 3090

20.2
24
Q4_K_M · Est. ~76 tok/sComfortable

48 GB

A6000 · 2×24G

37.5
48
Q8_0 · Est. ~30 tok/sComfortable

02  — Quantizations

QuantFile sizeEst. VRAM
Q4_K_M19 GB20.2 GB
Q5_K_M22.5 GB23.7 GB
Q8_036.3 GB37.5 GB

03  — Runtimes & use

llama.cpp · Ollama · LM Studio

General · Coding · Max intelligence

Memory and speed figures are planning estimates, not measured benchmarks. Memory includes a fixed allowance for KV cache and runtime overhead. Long-context prompts push VRAM use above the numbers shown here.