Mistral Nemo Base 2407 — Hardware Requirements & GPU Compatibility
Chat
Mistral-Nemo-Base-2407 is a 12-billion-parameter pretrained base language model jointly developed by Mistral AI and NVIDIA, intended as a drop-in replacement for the earlier Mistral 7B rather than a chat assistant; it has not been instruction-tuned or aligned, and the card notes it carries no built-in moderation mechanisms. It uses a standard transformer architecture with grouped-query attention and SwiGLU activations, and was trained on a large share of multilingual and code data across nine languages. At 12 billion parameters it fits comfortably on a single consumer GPU once quantized. Context length is 131,072 tokens (a 128k window). It is released under the Apache 2.0 license, permitting unrestricted commercial and research use, and was published in July 2024, alongside an instruction-tuned Nemo variant.
Runs great on a 24 GB GPU
Runs from 8 GB (IQ2_M · 16K)Best on 24 GB (Q4_K_M · ~81 tok/s · 64K)Macs from 16 GB (Q4_K_M · ~17 tok/s)
Apache 2.0
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Mistral Nemo Base 2407
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 run hf.co/mradermacher/Mistral-Nemo-Base-2407-i1-GGUF:Q4_K_MQ4_K_M fits fully on the RTX 4090 · 13.0 GB free · 16K context
Download: Mistral-Nemo-Base-2407.i1-Q4_K_M.gguf7.48 GBmradermacher
Ways to runOther ways
each opens step-by-stepRun in cloud
Your own private copy · Q4_K_M · 24 GB GPU
- The easiest way to try this model
- Running in a few minutes, easy setup
- Chat right here or connect any app
$5+ bonus for new users
Runs on RTX 3090 (24 GB) (13 GB headroom) · Q4_K_M
- Generation speed
- ~75 tok/s
- generation speed
- Cost per 1M output tokens
- $0.62
- per 1M output tokens
How Much VRAM Does Mistral Nemo Base 2407 Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 5.9 GB | 12.2 GB | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Fits: 64K context · ~111 tok/s | |
| 6.1 GB | 12.4 GB | 3-bit small — 3-bit small quantization | Fits: 64K context · ~108 tok/s | |
| 6.7 GB | 13 GB | 3-bit medium — 3-bit medium quantization | Fits: 64K context · ~98 tok/s | |
| 6.8 GB | 13.1 GB | 4-bit · legacy — 4-bit legacy quantization | Fits: 64K context · ~96 tok/s | |
| 8.1 GB | 14.4 GB | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Fits: 64K context · ~81 tok/s | |
| 9.4 GB | 15.7 GB | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Fits: 64K context · ~69 tok/s | |
| 10.8 GB | 17.1 GB | 6-bit, very good — 6-bit quantization, very good quality | Fits: 64K context · ~61 tok/s | |
| 13.0 GB | 19.3 GB | 8-bit, near-lossless — 8-bit quantization, near-lossless | Fits: 32K context · ~51 tok/s |
Selected Q4_K_M
4-bit medium quantization11.0GBat 16K context
- Weights
- 7.4 GB
- KV cache
- 3.4 GB
- Overhead
- 0.3 GB
Fits RTX 4090 (reference) · 13.0 GB free · ~81 tok/s
≥50 tok/s needs ≥620 GB/s · RTX 5090 ~144, RTX 3090 Ti ~81
Which Hardware Can Run Mistral Nemo Base 2407?
Q4_K_M · 8.1 GBMistral Nemo Base 2407 (Q4_K_M) requires 8.1 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 11+ GB is recommended. Using the full 131K context window can add up to 26.4 GB, bringing total usage to 34.5 GB. 40 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti, NVIDIA GeForce RTX 3080 Ti. 150 devices with unified memory can run Mistral Nemo Base 2407, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB).
12 GB
12 GB tier →fits · 1.0 GB free · up to 16K ctx
10 GPUs
RTX 3080 Ti~74Simulate output on RTX 3080 TiRTX 5070~54Simulate output on RTX 5070RTX 4070~41Simulate output on RTX 407020 GB
fits · 9.0 GB free · up to 32K ctx
1 GPU
RX 7900 XT~60Simulate output on RX 7900 XT32–40 GB
fits · 21.0 GB free · up to 64K ctx
1 GPU · 3 Mac classes · 4 device families
RTX 5090~144Simulate output on RTX 5090M5 Max36 GB~40Simulate output on Mac Studio M5 Max (36 GB)M532 GB~13Simulate output on MacBook Air 13" M5 (32 GB)Orin~17Simulate output on NVIDIA Jetson AGX Orin 32GBSnapdragon X Elite (X1E-80-100)~11Simulate output on Snapdragon X Elite Copilot+ PC
By VRAM tier · typical GPUs and Macs at Q4_K_M
How good is Mistral Nemo Base 2407?
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 4 boards that have results for this model, sorted by their weight in the llmrun Score; badges show who ran the test.
Alternatives to Mistral Nemo Base 2407
Mistral family · 35 models →Similar models
6| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| Gemma 3 12B ITGoogle | 74.2 | 10.9 GB | ~82 | same size |
| Gemma 4 12BGoogle | — | 9 GB | ~80 | same size |
| Baichuan2 13B Basebaichuan-inc | — | 11.5 GB | ~76 | same size |
| Llama 2 13B Chat HFMeta | — | 11.5 GB | ~76 | same size |
| Vicuna 13B V1.3LMSYS | — | 8.6 GB | ~76 | same size |
| Vllm Translategemma 12B ITchbae624 | — | 9.8 GB | ~73 | same size |
Fine-tunes & variants(2)
Specifications
- Publisher
- Mistral AI
- Family
- Mistral
- Parameters
- 12.2B
- Architecture
- MistralForCausalLM
- Context Length
- 131,072 tokens
- Vocabulary Size
- 131,072
- Release Date
- 2024-07-18
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does Mistral Nemo Base 2407 need?
Mistral Nemo Base 2407 requires 8.1 GB of VRAM at Q4_K_M, or 25.2 GB at BF16. Full 131K context adds up to 26.4 GB (34.5 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 12.2B × 4.8 bits ÷ 8 = 7.3 GB
KV Cache + Overhead ≈ 0.8 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 ≈ 27.2 GB (at full 131K context)
VRAM usage by quantization
Can NVIDIA GeForce RTX 4090 run Mistral Nemo Base 2407?
Yes, at Q8_0 (13.0 GB) or lower. Higher quantizations like BF16 (25.2 GB) exceed the NVIDIA GeForce RTX 4090's 24 GB.
What's the best quantization for Mistral Nemo Base 2407?
For Mistral Nemo Base 2407, Q4_K_M (8.1 GB) offers the best balance of quality and VRAM usage. Q5_K_S (9.1 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 4.1 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run Mistral Nemo Base 2407 on a Mac?
Yes — Mac Mini M2 Pro (16 GB) and 133 other Macs can run Mistral Nemo Base 2407. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 8.1 GB of usable unified memory (RAM minus macOS overhead).
Can I run Mistral Nemo Base 2407 locally?
Yes — Mistral Nemo Base 2407 can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Mistral Nemo Base 2407?
At Q4_K_M, Mistral Nemo Base 2407 can reach ~595 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~81 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 ÷ 8.07 × 0.65 = ~644 tok/s
Estimated speed at Q4_K_M (8.1 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Mistral Nemo Base 2407?
At Q4_K_M, the download is about 7.35 GB. The full-precision BF16 version is 24.50 GB. The smallest option (IQ2_XXS) is 3.37 GB.
Which GPUs can run Mistral Nemo Base 2407?
40 consumer GPUs can run Mistral Nemo Base 2407 at Q4_K_M. Top options include AMD Radeon RX 6800, AMD Radeon RX 6800 XT, AMD Radeon RX 6900 XT, AMD Radeon RX 6700 XT. 26 GPUs have plenty of headroom for comfortable inference.
Which devices can run Mistral Nemo Base 2407?
150 devices with unified memory can run Mistral Nemo Base 2407 at Q4_K_M (8.1 GB), including AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.