Llama on NVIDIA Ampere

Can I run Llama 3.1 70B on an RTX 3090?

Not at Q4_K_M — it needs 43.4 GB against 21.8 GB available. You would need 2 of these cards.

Won't fit

43.4 GB of 21.8 GB · 199%
045 GB
Weights 40.0 GB
KV cache 2.5 GB
Runtime overhead 0.9 GB
Over the limit 21.6 GB

Short by 21.6 GB. You can run it with 36 of 80 layers on the RTX 3090 and the rest in system RAM, at roughly 1.67 tok/s — usable for batch work, painful for chat. A smaller quantisation or a shorter context is usually the better trade.

Generation1.67tok/s
Prompt processing211tok/s
Max context0tokens
KV per 1K tokens0GB

Every quantisation of Llama 3.1 70B on a RTX 3090

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

QuantisationbpwWeights TotalVerdictMax ctxtok/s
FP16 / BF16 16.00 131.4 GB 134.8 GB Won't fit — 0.35
INT8 / W8A8 8.50 69.3 GB 72.7 GB Won't fit — 0.75
Q8_0 (GGUF) 8.50 69.3 GB 72.7 GB Won't fit — 0.75
FP8 (E4M3) 8.00 65.7 GB 69.1 GB Won't fit — 0.80
Q6_K 6.56 53.9 GB 57.3 GB Won't fit — 1.06
Q5_K_M 5.67 46.6 GB 50.0 GB Won't fit — 1.31
Q5_K_S 5.52 45.3 GB 48.7 GB Won't fit — 1.37
Q4_K_M 4.85 40.0 GB 43.4 GB Won't fit — 1.67
Q4_K_S 4.58 37.8 GB 41.2 GB Won't fit — 1.83
AWQ 4-bit 4.25 37.8 GB 41.2 GB Won't fit — 1.83
GPTQ 4-bit 4.25 37.8 GB 41.2 GB Won't fit — 1.83
MXFP4 4.25 37.8 GB 41.2 GB Won't fit — 1.83
Q4_0 4.55 37.6 GB 41.0 GB Won't fit — 1.88
IQ4_XS 4.25 35.2 GB 38.6 GB Won't fit — 2.10
Q3_K_M 3.91 32.5 GB 35.9 GB Won't fit — 2.48
IQ3_M 3.70 30.8 GB 34.2 GB Won't fit — 2.75
IQ3_XXS 3.06 25.7 GB 29.1 GB Won't fit — 4.40
Q2_K 2.63 22.3 GB 25.7 GB Won't fit — 6.96
IQ2_XXS 2.06 17.8 GB 21.2 GB Fits, but tight 10K 30.4
IQ1_M 1.75 15.3 GB 18.7 GB Runs comfortably 18K 34.8

Also worth checking

Llama 3.1 70B on every GPU
Full hardware tableEverything that runs on a RTX 3090
Full model tableWhy the KV cache is the part that bites
Guide