NVIDIA · 24 GB GDDR6X · 1008 GB/s bandwidth ·
Meta · 405.9B · model ctx up to 128K
No — Llama-3.1-405B-Instruct at Q4 needs ≈ 264.75 GB even at 4K context, beyond the RTX 3090 Ti 24GB's 24 GB GDDR6X. It would only run with heavy CPU/disk offload. FP16 needs ≈ 893.8 GB — does not fit on this card. Estimates come from memory-bandwidth math; rows tagged measured override estimates. Relative ranking is more reliable than absolute tok/s. Methodology.
Fit & speed by quant and context
Quant
Context
VRAM needed
Fits?
tok/s (decode)
Q4
4K
264.75 GB
No
—
Q4
8K default
305.25 GB
No
—
Q4
32K
548.25 GB
No
—
FP16
4K
853.3 GB
No
—
FP16
8K default
893.8 GB
No
—
FP16
32K
1136.8 GB
No
—
Fits = weights + KV(ctx) + 1 GB overhead ≤ 95% of VRAM. Measured
anchors are context-agnostic; the fit verdict is recomputed per context. A missing tok/s
means the model is far beyond this card (offload-only territory).
VRAM breakdown at 8K context
Quant
Weights
KV cache
Overhead
Total needed
RTX 3090 Ti 24GB VRAM
Q4
223.25 GB
81.0 GB
1 GB
305.25 GB
24 GB
FP16
811.8 GB
81.0 GB
1 GB
893.8 GB
24 GB
Other GPUs that run Llama-3.1-405B-Instruct
No catalog GPU fully fits Llama-3.1-405B-Instruct at Q4/8K — see the model page for offload estimates.