Home / Models / Llama 3.1 70B Instruct / NVIDIA RTX 3090 Ti 24 GB

Calculated memory fit · not a benchmark

Can Llama 3.1 70B Instruct run on NVIDIA RTX 3090 Ti 24 GB?

Can it run? Yes, with CPU/RAM offload. Full GPU fit? No.

Can it run?

Can it run?
Yes, with CPU/RAM offload. This is not equivalent to full-GPU inference.
Full GPU fit?
No
Model
Llama 3.1 70B Instruct · 71B
GPU
NVIDIA RTX 3090 Ti 24 GB · 24 GB advertised

Memory calculation

Breakdown for Q4 at 8K context, batch size 1. Calculator v1.0.0.

Model weights41.4 GB
KV cache2.5 GB
Runtime reserve2.6 GB
Safety margin1.5 GB
Required (estimated)48.0 GB
GPU usable VRAM21.6 GB

Result: CPU RAM OFFLOAD REQUIRED

Quantization table

Quantization2K4K8K16K32K64K128K
BF16NoNoNoNoNoNoNo
FP16NoNoNoNoNoNoNo
Q3OffloadOffloadOffloadOffloadOffloadOffloadOffload
Q4OffloadOffloadOffloadOffloadOffloadOffloadOffload
Q5OffloadOffloadOffloadOffloadOffloadOffloadOffload
Q6NoNoNoNoNoNoNo
Q8NoNoNoNoNoNoNo

What fits entirely in VRAM?

No calculated full-VRAM fit at the evaluated quantizations and context lengths.

What requires RAM offload?

Weights may run with CPU/RAM offload. That is not equivalent to full-GPU inference. RigForAI does not estimate tokens/sec in this release.

Available graphics cards

These SKUs use this GPU chip. Amazon CTAs appear only for EXACT/HIGH matches. Absence of a price does not change the compatibility result above.

Check price on Amazon

Sources

Model: official config meta-llama/Llama-3.1-70B-Instruct (official-fallback-config). GPU/product specs: Icecat. Compatibility: RigForAI calculator v1.0.0 estimated VRAM · Q8 at 8K: DOES_NOT_FIT. Amazon: affiliate commerce match, not a spec source.