localLLMriglocal, on your table

estimate

Can I run Llama 3.3 70B on an RTX 3090?

No: needs ~45.8 GB, this card has ~23.4 GB usable (estimate)

Can I run Llama 3.3 70B on an RTX 3090? No: needs ~45.8 GB, this card has ~23.4 GB usable (estimate). About 45.8 GB at Q4_K_M (estimate).

Updated . The later of the hardware price date and the formula date.

Llama 3.3 70B on RTX 3090, formula estimate
QuantContextWeightsKVOverheadTotalUsableFitEst. tok/s
Q4 Q4_K_M8,192 tokens42.4 GB estimate2.7 GB estimate0.8 GB estimate45.8 GB estimate23.4 GB estimateno—
Q5 Q5 (~0.71 bytes/param)8,192 tokens50.1 GB estimate2.7 GB estimate0.8 GB estimate53.6 GB estimate23.4 GB estimateno—
Q8 Q88,192 tokens75.5 GB estimate2.7 GB estimate0.8 GB estimate79.0 GB estimate23.4 GB estimateno—
FP16 FP168,192 tokens141.2 GB estimate2.7 GB estimate0.8 GB estimate144.7 GB estimate23.4 GB estimateno—
Q4, longer context32,768 tokens42.4 GB estimate10.7 GB estimate0.8 GB estimate53.9 GB estimate23.4 GB estimateno—

Formula estimate, not a benchmark. Method →

Cheapest catalog option that fits at Q4

Mac mini M4 Pro 64GB, catalog price, €1,890, dated 2026-10-16. No buy link.

This hardware

$1,370€1,224

used (median of sold listings)

price as of 16 Oct 2026

new (lowest in-stock EU): $2,483€2,218. GPU PRIX. New low €2,218, Zotac Trinity OC, France. Page last updated 16 Oct 2026.

pcprice.watch

Used median €1,224, typical €1,136–€1,312, 232 eBay sold sales over 90 days, updated 16 Oct 2026. Not an EU retail listing.

Embed this

Llama 3.3 70B on RTX 3090

[![Llama 3.3 70B Q4 on RTX 3090: ✗ · needs 45.8 GB (estimate)](https://localllmrig.com/badge/llama-3.3-70b/rtx-3090.svg)](https://localllmrig.com/can-i-run/llama-3.3-70b/rtx-3090?utm_source=badge)

Questions

Can I run Llama 3.3 70B on an RTX 3090?
No: needs ~45.8 GB, this card has ~23.4 GB usable (estimate).
How much memory does Llama 3.3 70B need at Q4?
About 45.8 GB at 8,192 tokens (estimate): 42.4 GB weights, 2.7 GB KV cache, 0.8 GB overhead.
How fast is Llama 3.3 70B on RTX 3090 (estimate)?
No tokens-per-second estimate. The Q4 fit is not full and not offload, so this page leaves speed blank.
Llama 3.3 70B on RTX 3090 (estimate) · localLLMrig