llmrun
IBMGraniteGraniteMoeHybridForCausalLM

Granite 4.0 350M — Hardware Requirements & GPU Compatibility

ChatFunctions

Granite 4.0 350M is IBM's instruct model containing 350 million parameters; it is the smallest member of the Granite 4.0 Nano group, finetuned from Granite-4.0-350M-Base via supervised finetuning, reinforcement learning and model merging. Designed for lightweight assistant tasks, it handles tool calling, fill-in-the-middle code completion, and supports twelve languages, including English, German, Japanese, Arabic, and Chinese. Its config employs a hybrid architecture (GraniteMoeHybrid). Given this size, it runs on a CPU, phone-class devices, or small GPUs without quantization, making it suitable for edge and on-device use. The context window holds 32,768 tokens. Released under the Apache 2.0 license, it allows unrestricted commercial and research usage. The Hub repository appeared in October 2025, with the card listing a release date of October 28, 2025. This model is the smaller sibling of Granite 4.0 1B, and the card notes that users can fine-tune these Nano models to add languages beyond the supported list.

352M params · dense33K contextApache 2.018.1K downloads74 likes9.7K quant downloads
S

Runs great on a 24 GB GPU

Runs from8 GB Q4_K_M · 32KBest on24 GB Q4_K_M · ~1150 tok/s · 32KMacs from8 GB Q4_K_M · ~126 tok/s

Apache 2.0

Reference GPU · RTX 4090 24 GB

Estimates, ±20% · how we estimate

Run Granite 4.0 350M

One recommended way. Every other method, OS and quant is in the step-by-step guide.Each way opens a step-by-step guide for your OS and quant.

Recommended

Ollama
one command · needs Ollama installed
$ollama run hf.co/ibm-granite/granite-4.0-350m-GGUF:Q4_K_M

Q4_K_M fits fully on the RTX 4090 · 23.0 GB free · 16K context

Download: granite-4.0-350m-Q4_K_M.gguf0.24 GBibm-granite

install, OS, other quants

Ways to runOther ways

each opens step-by-step

Run in cloud

Runs on RTX 3060 12GB (11 GB headroom) · Q4_K_M

Generation speed
~411 tok/s
generation speed
Cost per 1M output tokens
$0.043
per 1M output tokens

How Much VRAM Does Granite 4.0 350M Need?

Select a quantization level; the listed hardware adheres to it.

QuantizationVRAM+32KOn RTX 4090 (reference)
0.5 GB1.4 GB3-bit small — 3-bit small quantizationFits: 32K context · ~1285 tok/s
0.5 GB1.4 GB2-bit · K-quant — 2-bit quantization with K-quant improvementsFits: 32K context · ~1285 tok/s
0.5 GB1.4 GB3-bit medium — 3-bit medium quantizationFits: 32K context · ~1236 tok/s
0.5 GB1.4 GB4-bit · legacy — 4-bit legacy quantizationFits: 32K context · ~1236 tok/s
0.6 GB1.5 GB4-bit medium — 4-bit medium quantization — most popular sweet spotFits: 32K context · ~1150 tok/s
0.6 GB1.5 GB5-bit medium — 5-bit medium quantization — good quality/size tradeoffFits: 32K context · ~1074 tok/s
0.7 GB1.5 GB6-bit, very good — 6-bit quantization, very good qualityFits: 32K context · ~1008 tok/s
0.7 GB1.6 GB8-bit, near-lossless — 8-bit quantization, near-losslessFits: 32K context · ~923 tok/s

Selected Q4_K_M

4-bit medium quantization

1.0GBat 16K context

Weights
0.2 GB
KV cache
0.5 GB
Overhead
0.3 GB

Fits RTX 4090 (reference) · 23.0 GB free · ~1150 tok/s

≥50 tok/s needs ≥44 GB/s · RTX 5090 ~2044, M5 Ultra ~1508

Which Hardware Can Run Granite 4.0 350M?

Q4_K_M · 0.6 GB

Granite 4.0 350M (Q4_K_M) requires 0.6 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 1+ GB is recommended. Using the full 33K context window can add up to 0.9 GB, bringing total usage to 1.4 GB. 114 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti. 185 devices with unified memory can run Granite 4.0 350M, including Mac Studio M5 Ultra (256 GB), Mac Studio M5 Ultra (512 GB).

By VRAM tier · typical GPUs and Macs at Q4_K_M
8 GB · RTX 4060Q4_K_M · 1 GB · ~310 tok/s · 32K8 GB tier →
12 GB · RTX 3060 12GBQ4_K_M · 1 GB · ~411 tok/s · 32K12 GB tier →
16 GB · RTX 5060 Ti 16GBQ4_K_M · 1 GB · ~511 tok/s · 32K16 GB tier →
24 GB · RTX 4090referenceQ4_K_M · 1 GB · ~1150 tok/s · 32K24 GB tier →
Mac 32 GB · Mac Mini M4Q4_K_M · 1 GB · ~147 tok/s · 32KMacs →
Mac 64 GB · Mac Studio M4 MaxQ4_K_M · 1 GB · ~671 tok/s · 32KMacs →

How good is Granite 4.0 350M?

llmrun Score blends public benchmark boards into one number; 100 = the average open model.

Not enough data yet for an llmrun Score.

A Score needs results on enough independent boards. How the Composite Score is built

Benchmarks

Here are all 3 boards that have results for this model, sorted by their weight in the llmrun Score; badges show who ran the test.

  • GPQA Diamond

    Epoch AI · Oct 2025Independent

    11.2%

    #99/100 open

    Independent
  • AIME 2024/2025

    Epoch AI · Oct 2025Independent

    0.3%

    #86/88 open

    Independent
  • CritPt

    Epoch AI · Oct 2025Lab-reported

    0.0%

    #61/85 open

    Lab-reported

Alternatives to Granite 4.0 350M

Granite family · 28 models →

Similar models

6
ModelScoreVRAMtok/sWhy it is similar
SmolLM 360MHugging Face—0.6 GB~1092same size
LFM2.5 350MLiquid AI—0.8 GB~1130same size
MobileLLM R1 360MMeta—1.3 GB~2730same size
ERNIE 4.5 0.3B PTBaidu—0.7 GB~1213same size
Opt 350MMeta—0.8 GB~851same size
Codegen 350M MonoSalesforce—0.8 GB~851same size

Specifications

Publisher
IBM
Family
Granite
Parameters
352M
Architecture
GraniteMoeHybridForCausalLM
Context Length
32,768 tokens
Vocabulary Size
100,352
Release Date
2025-10-07
License
Apache 2.0

Frequently Asked Questions

How much VRAM does Granite 4.0 350M need?

Granite 4.0 350M requires 0.6 GB of VRAM at Q4_K_M, or 1.1 GB at BF16. Full 33K context adds up to 0.9 GB (1.4 GB total).

VRAM = Weights + KV Cache + Overhead

Weights = 352M × 4.8 bits ÷ 8 = 0.2 GB

KV Cache + Overhead ≈ 0.4 GB (at 2K context + ~0.3 GB framework)

Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.

KV Cache + Overhead ≈ 1.3 GB (at full 33K context)

VRAM usage by quantization

0.6 GB
1.4 GB

Learn more about VRAM estimation →

What's the best quantization for Granite 4.0 350M?

For Granite 4.0 350M, Q4_K_M (0.6 GB) offers the best balance of quality and VRAM usage. Q5_0 (0.6 GB) provides better quality if you have the VRAM. The smallest option is IQ3_XXS at 0.5 GB.

VRAM requirement by quantization

IQ3_XXS
0.5 GB
Q4_0
0.5 GB
Q4_K_S
0.6 GB
Q4_K_M ★
0.6 GB
Q5_K_S
0.6 GB
BF16
1.1 GB

★ Recommended — best balance of quality and VRAM usage.

Learn more about quantization →

Can I run Granite 4.0 350M on a Mac?

Yes — MacBook Air 13" M3 (8 GB) and 158 other Macs can run Granite 4.0 350M. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 0.6 GB of usable unified memory (RAM minus macOS overhead).

Can I run Granite 4.0 350M locally?

Yes — Granite 4.0 350M can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.

How fast is Granite 4.0 350M?

At Q4_K_M, Granite 4.0 350M can reach ~8421 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~1150 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.

tok/s = (bandwidth GB/s ÷ model GB) × efficiency

Example: NVIDIA B200 → 8000 ÷ 0.57 × 0.65 = ~9123 tok/s

Estimated speed at Q4_K_M (0.6 GB)

~9123 tok/s
~1150 tok/s
~9123 tok/s
~8421 tok/s

Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.

Learn more about tok/s estimation →

What's the download size of Granite 4.0 350M?

At Q4_K_M, the download is about 0.21 GB. The full-precision BF16 version is 0.70 GB. The smallest option (IQ3_XXS) is 0.14 GB.

Which GPUs can run Granite 4.0 350M?

114 consumer GPUs can run Granite 4.0 350M at Q4_K_M. Top options include AMD Radeon RX 5600 XT, AMD Radeon RX 5700 XT, AMD Radeon RX 580 8GB. 114 GPUs have plenty of headroom for comfortable inference.

Which devices can run Granite 4.0 350M?

185 devices with unified memory can run Granite 4.0 350M at Q4_K_M (0.6 GB), including AMD Ryzen 7 8845HS (Hawk Point) Laptop, AMD Ryzen AI 7 350 (Krackan Point) Laptop, AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10. Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.