Mistral on NVIDIA Ada

Can I run Mistral Small 24B on an RTX 4060 Ti 16GB?

Not at Q4_K_M — it needs 15.5 GB against 14.2 GB available. Drop to IQ4_XS and it fits, at about 13.8 tokens per second.

Won't fit

15.5 GB of 14.2 GB · 109%
016 GB
Weights 13.4 GB
KV cache 1.3 GB
Runtime overhead 0.8 GB
Over the limit 1.3 GB

Short by 1.3 GB. You can run it with 36 of 40 layers on the RTX 4060 Ti 16GB and the rest in system RAM, at roughly 9.09 tok/s — usable for batch work, painful for chat. A smaller quantisation or a shorter context is usually the better trade.

Generation9.09tok/s
Prompt processing392tok/s
Max context0tokens
KV per 1K tokens0GB

Every quantisation of Mistral Small 24B on a RTX 4060 Ti 16GB

Highlighted row is the highest quality that still fits at 8K context.

QuantisationbpwWeights TotalVerdictMax ctxtok/s
FP16 / BF16 16.00 43.9 GB 46.0 GB Won't fit 1.09
INT8 / W8A8 8.50 23.0 GB 25.1 GB Won't fit 2.74
Q8_0 (GGUF) 8.50 23.0 GB 25.1 GB Won't fit 2.74
FP8 (E4M3) 8.00 22.0 GB 24.0 GB Won't fit 2.96
Q6_K 6.56 18.0 GB 20.1 GB Won't fit 4.33
Q5_K_M 5.67 15.6 GB 17.6 GB Won't fit 5.95
Q5_K_S 5.52 15.1 GB 17.2 GB Won't fit 6.42
AWQ 4-bit 4.25 13.5 GB 15.6 GB Won't fit 9.03
GPTQ 4-bit 4.25 13.5 GB 15.6 GB Won't fit 9.03
MXFP4 4.25 13.5 GB 15.6 GB Won't fit 9.03
Q4_K_M 4.85 13.4 GB 15.5 GB Won't fit 9.09
Q4_K_S 4.58 12.7 GB 14.8 GB Won't fit 4K 11.0
Q4_0 4.55 12.6 GB 14.7 GB Won't fit 5K 11.1
IQ4_XS 4.25 11.9 GB 13.9 GB Fits, but tight 10K 13.8
Q3_K_M 3.91 11.0 GB 13.1 GB Fits, but tight 16K 14.8
IQ3_M 3.70 10.4 GB 12.5 GB Runs comfortably 19K 15.6
IQ3_XXS 3.06 8.8 GB 10.9 GB Runs comfortably 30K 18.3
Q2_K 2.63 7.7 GB 9.8 GB Runs comfortably 32K 20.7
IQ2_XXS 2.06 6.2 GB 8.3 GB Runs comfortably 32K 25.2
IQ1_M 1.75 5.4 GB 7.5 GB Runs comfortably 32K 28.5

Also worth checking