Llama on NVIDIA Blackwell

Can I run Llama 4 Scout 109B-A17B on an RTX 5090?

Not at Q4_K_M — it needs 64.0 GB against 29.3 GB available. You would need 3 of these cards.

Won't fit

64.0 GB of 29.3 GB · 219%
067 GB
Weights 61.7 GB
KV cache 1.5 GB
Runtime overhead 0.8 GB
Over the limit 34.8 GB

Short by 34.8 GB. You can run it with 20 of 48 layers on the RTX 5090 and the rest in system RAM, at roughly 6.75 tok/s — usable for batch work, painful for chat. A smaller quantisation or a shorter context is usually the better trade.

Generation6.75tok/s
Prompt processing2582tok/s
Max context0tokens
KV per 1K tokens0GB

Every quantisation of Llama 4 Scout 109B-A17B on a RTX 5090

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

QuantisationbpwWeights TotalVerdictMax ctxtok/s
FP16 / BF16 16.00 203.0 GB 205.4 GB Won't fit — 1.48
INT8 / W8A8 8.50 107.4 GB 109.7 GB Won't fit — 3.17
Q8_0 (GGUF) 8.50 107.4 GB 109.7 GB Won't fit — 3.17
FP8 (E4M3) 8.00 101.5 GB 103.9 GB Won't fit — 3.34
Q6_K 6.56 83.2 GB 85.6 GB Won't fit — 4.38
Q5_K_M 5.67 71.9 GB 74.3 GB Won't fit — 5.32
Q5_K_S 5.52 70.0 GB 72.4 GB Won't fit — 5.63
Q4_K_M 4.85 61.7 GB 64.0 GB Won't fit — 6.75
Q4_K_S 4.58 58.3 GB 60.7 GB Won't fit — 7.61
Q4_0 4.55 58.0 GB 60.3 GB Won't fit — 7.66
AWQ 4-bit 4.25 56.8 GB 59.1 GB Won't fit — 7.81
GPTQ 4-bit 4.25 56.8 GB 59.1 GB Won't fit — 7.81
MXFP4 4.25 56.8 GB 59.1 GB Won't fit — 7.81
IQ4_XS 4.25 54.2 GB 56.6 GB Won't fit — 8.44
Q3_K_M 3.91 50.0 GB 52.3 GB Won't fit — 9.81
IQ3_M 3.70 47.4 GB 49.7 GB Won't fit — 11.2
IQ3_XXS 3.06 39.4 GB 41.8 GB Won't fit — 16.8
Q2_K 2.63 34.1 GB 36.4 GB Won't fit — 26.3
IQ2_XXS 2.06 27.0 GB 29.3 GB Won't fit 8K 130
IQ1_M 1.75 23.1 GB 25.4 GB Runs comfortably 28K 209

Also worth checking

Llama 4 Scout 109B-A17B on every GPU
Full hardware tableEverything that runs on a RTX 5090
Full model tableWhy the KV cache is the part that bites
Guide