Llama on NVIDIA Ada

Can I run Llama 3.1 70B on an L40S?

Yes. Llama 3.1 70B at Q4_K_M uses 43.4 GB of the 44.3 GB available on a L40S, and runs at about 12.5 tokens per second. You can push the context to 11K.

Fits, but tight

43.4 GB of 44.3 GB · 98%
044 GB
Weights 40.0 GB
KV cache 2.5 GB
Runtime overhead 0.9 GB

This fits with almost nothing to spare. A background application claiming VRAM will push it over. Drop to the next quantisation down, or quantise the KV cache to Q8_0 — that halves the cache for no meaningful quality loss.

Generation12.5tok/s
Prompt processing539tok/s
Max context11Ktokens
KV per 1K tokens0GB

Every quantisation of Llama 3.1 70B on a L40S

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 — 0.40
INT8 / W8A8 8.50 69.3 GB 72.7 GB Won't fit — 1.19
Q8_0 (GGUF) 8.50 69.3 GB 72.7 GB Won't fit — 1.19
FP8 (E4M3) 8.00 65.7 GB 69.1 GB Won't fit — 1.32
Q6_K 6.56 53.9 GB 57.3 GB Won't fit — 2.27
Q5_K_M 5.67 46.6 GB 50.0 GB Won't fit — 4.21
Q5_K_S 5.52 45.3 GB 48.7 GB Won't fit — 4.92
Q4_K_M 4.85 40.0 GB 43.4 GB Fits, but tight 11K 12.5
Q4_K_S 4.58 37.8 GB 41.2 GB Fits, but tight 18K 13.2
AWQ 4-bit 4.25 37.8 GB 41.2 GB Fits, but tight 18K 13.2
GPTQ 4-bit 4.25 37.8 GB 41.2 GB Fits, but tight 18K 13.2
MXFP4 4.25 37.8 GB 41.2 GB Fits, but tight 18K 13.2
Q4_0 4.55 37.6 GB 41.0 GB Fits, but tight 19K 13.3
IQ4_XS 4.25 35.2 GB 38.6 GB Runs comfortably 26K 14.1
Q3_K_M 3.91 32.5 GB 35.9 GB Runs comfortably 35K 15.2
IQ3_M 3.70 30.8 GB 34.2 GB Runs comfortably 40K 16.0
IQ3_XXS 3.06 25.7 GB 29.1 GB Runs comfortably 57K 19.0
Q2_K 2.63 22.3 GB 25.7 GB Runs comfortably 68K 21.8
IQ2_XXS 2.06 17.8 GB 21.2 GB Runs comfortably 82K 26.9
IQ1_M 1.75 15.3 GB 18.7 GB Runs comfortably 90K 30.9

Also worth checking

Llama 3.1 70B on every GPU
Full hardware tableEverything that runs on a L40S
Full model tableWhy the KV cache is the part that bites
Guide