Llama on NVIDIA Blackwell

Can I run Llama 3.1 70B on an RTX 5090?

Not at Q4_K_M — it needs 43.4 GB against 29.3 GB available. Drop to IQ3_XXS and it fits, at about 43.0 tokens per second.

Won't fit

43.4 GB of 29.3 GB · 148%
045 GB
Weights 40.0 GB
KV cache 2.5 GB
Runtime overhead 0.9 GB
Over the limit 14.1 GB

Short by 14.1 GB. You can run it with 51 of 80 layers on the RTX 5090 and the rest in system RAM, at roughly 2.65 tok/s — usable for batch work, painful for chat. A smaller quantisation or a shorter context is usually the better trade.

Generation2.65tok/s
Prompt processing622tok/s
Max context0tokens
KV per 1K tokens0GB

Every quantisation of Llama 3.1 70B on a RTX 5090

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 — 0.39
INT8 / W8A8 8.50 69.3 GB 72.7 GB Won't fit — 0.92
Q8_0 (GGUF) 8.50 69.3 GB 72.7 GB Won't fit — 0.92
FP8 (E4M3) 8.00 65.7 GB 69.1 GB Won't fit — 1.01
Q6_K 6.56 53.9 GB 57.3 GB Won't fit — 1.41
Q5_K_M 5.67 46.6 GB 50.0 GB Won't fit — 1.88
Q5_K_S 5.52 45.3 GB 48.7 GB Won't fit — 1.98
Q4_K_M 4.85 40.0 GB 43.4 GB Won't fit — 2.65
Q4_K_S 4.58 37.8 GB 41.2 GB Won't fit — 3.09
AWQ 4-bit 4.25 37.8 GB 41.2 GB Won't fit — 3.09
GPTQ 4-bit 4.25 37.8 GB 41.2 GB Won't fit — 3.09
MXFP4 4.25 37.8 GB 41.2 GB Won't fit — 3.09
Q4_0 4.55 37.6 GB 41.0 GB Won't fit — 3.22
IQ4_XS 4.25 35.2 GB 38.6 GB Won't fit — 3.84
Q3_K_M 3.91 32.5 GB 35.9 GB Won't fit — 5.18
IQ3_M 3.70 30.8 GB 34.2 GB Won't fit — 6.82
IQ3_XXS 3.06 25.7 GB 29.1 GB Fits, but tight 9K 43.0
Q2_K 2.63 22.3 GB 25.7 GB Runs comfortably 19K 49.1
IQ2_XXS 2.06 17.8 GB 21.2 GB Runs comfortably 34K 60.5
IQ1_M 1.75 15.3 GB 18.7 GB Runs comfortably 42K 69.2

Also worth checking

Llama 3.1 70B on every GPU
Full hardware tableEverything that runs on a RTX 5090
Full model tableWhy the KV cache is the part that bites
Guide