Skip to content
Run LLM Locally

Mistral Small 4 119B on the RTX 5090

Mistral Small 4 119B at Q4_K_M needs about 72 GB, more than the 32 GB of the RTX 5090: part of the model runs from system RAM, which slows generation. At Q3_K_M it does not fit in VRAM.

ShareFacebookXWhatsAppLinkedInEmail

Mistral Small 4 119B

Q4_K_M · 8K context

Possible but slow

The model needs about 72 GB but only 31 GB of VRAM is usable: 41 GB must be loaded into system RAM. It runs, 43% on the GPU, but more slowly.

Your hardware
RTX 5090 32 GB · 64 GB RAM
Estimated memory
~72 GB estimate
VRAM (usable memory)
31 GB
GPU only
No
CPU/GPU offload
Required (41 GB)
Generation speed
8–16 tok/s estimate
Largest context in fast memory
None
Confidence: low

Based on: memory bandwidth (1,792 GB/s) and model size · system RAM bandwidth (~51 GB/s)

Memory breakdown

~72 GB (70.3 GB–73.8 GB) needed · estimate

41 GB must be offloaded to system RAM

  • Model weights67.6 GB
  • KV cache0.2 GB
  • Runtime overhead0.8 GB
  • Safety margin3.4 GB
Assumptions: context 8K tokens · KV cache f16 · llama.cpp (CUDA) · flash attention on · weights = real file size · compressed MLA KV cache

EstimateData updated 40 days agoSource: Run LLM Locally engine, Hugging Face config.json + model card

What makes this estimate less reliable

  • no benchmark for this hardware and model yet
  • split CPU/GPU runs vary a lot with the CPU and PCIe link
  • system RAM speed unknown (dual-channel DDR4-3200 assumed)
  • mixture-of-experts: speed depends on expert routing

Other quantizations on the RTX 5090

Mistral Small 4 119B on RTX 5090 + 64 GB RAM · 8K context
Mistral Small 4 119B Q8_08K~125 GB · No dataIncompatible
Mistral Small 4 119B Q6_K8K~102 GB · No dataIncompatible
Mistral Small 4 119B Q5_K_M8K~84.1 GB · No dataIncompatible
Mistral Small 4 119B Q4_K_M8K~72 GB · 8–16 tok/s estimatePossible but slow
Mistral Small 4 119B IQ4_XS8K~63.4 GB · 10–20 tok/s estimatePossible but slow
Mistral Small 4 119B Q3_K_M8K~54.3 GB · 14–27 tok/s estimateCPU/GPU offload

Alternatives

GPUs that run Mistral Small 4 119B Q4_K_M entirely in VRAM

Other models that fit the RTX 5090

Hardware that runs Mistral Small 4 119B Q4_K_M entirely in memory

Prices and links follow the market you choose.

Radeon PRO W7900

AMD

ExcellentRuns Mistral Small 4 119B Q4_K_M entirely in memory

Price not available

Availability unknown

View on amazon.com

Affiliate link

NVIDIA DGX Spark (GB10)

Unified memory PC

ExcellentRuns Mistral Small 4 119B Q4_K_M entirely in memory

Price not available

Availability unknown

View on amazon.com

Affiliate link

No current offer imported for this market: the link opens a marketplace search.

Frequently asked questions

Can the RTX 5090 run Mistral Small 4 119B?
The model needs about 72 GB but only 31 GB of VRAM is usable: 41 GB must be loaded into system RAM. It runs, 43% on the GPU, but more slowly.
Which quantization should I use on the RTX 5090?
None fits entirely in VRAM at 8K context; the smallest, Q3_K_M, still needs offload.
What is the largest context on the RTX 5090?
At Q4_K_M, the model does not fit entirely in VRAM even at a short context.
How fast is Mistral Small 4 119B (RTX 5090)?
We estimate 8–16 tok/s (estimate, confidence low). Based on: memory bandwidth (1,792 GB/s) and model size, system RAM bandwidth (~51 GB/s).