Llama on NVIDIA Blackwell

Can I run Llama 3.2 3B on an RTX 5070 Ti?

Yes. Llama 3.2 3B at Q4_K_M uses 3.5 GB of the 14.2 GB available on a RTX 5070 Ti, and runs at about 249 tokens per second. You can push the context to 106K.

Runs comfortably

3.5 GB of 14.2 GB · 25%
014 GB
Weights 1.8 GB
KV cache 0.9 GB
Runtime overhead 0.8 GB

Llama 3.2 3B at Q4_K_M leaves 10.7 GB spare on a RTX 5070 Ti. There is room to raise the context length or move up a quantisation level.

Generation249tok/s
Prompt processing5757tok/s
Max context106Ktokens
KV per 1K tokens0GB

Every quantisation of Llama 3.2 3B on a RTX 5070 Ti

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

QuantisationbpwWeights TotalVerdictMax ctxtok/s
FP16 / BF16 16.00 6.0 GB 7.7 GB Runs comfortably 68K 90.9
INT8 / W8A8 8.50 3.1 GB 4.8 GB Runs comfortably 95K 163
Q8_0 (GGUF) 8.50 3.1 GB 4.8 GB Runs comfortably 95K 163
FP8 (E4M3) 8.00 3.0 GB 4.7 GB Runs comfortably 96K 168
Q6_K 6.56 2.5 GB 4.1 GB Runs comfortably 100K 198
AWQ 4-bit 4.25 2.1 GB 3.8 GB Runs comfortably 103K 222
GPTQ 4-bit 4.25 2.1 GB 3.8 GB Runs comfortably 103K 222
MXFP4 4.25 2.1 GB 3.8 GB Runs comfortably 103K 222
Q5_K_M 5.67 2.1 GB 3.8 GB Runs comfortably 103K 223
Q5_K_S 5.52 2.1 GB 3.7 GB Runs comfortably 104K 228
Q4_K_M 4.85 1.8 GB 3.5 GB Runs comfortably 106K 249
Q4_K_S 4.58 1.8 GB 3.4 GB Runs comfortably 107K 259
Q4_0 4.55 1.7 GB 3.4 GB Runs comfortably 107K 260
IQ4_XS 4.25 1.6 GB 3.3 GB Runs comfortably 108K 271
Q3_K_M 3.91 1.5 GB 3.2 GB Runs comfortably 109K 286
IQ3_M 3.70 1.5 GB 3.1 GB Runs comfortably 109K 296
IQ3_XXS 3.06 1.3 GB 2.9 GB Runs comfortably 111K 330
Q2_K 2.63 1.1 GB 2.8 GB Runs comfortably 113K 359
IQ2_XXS 2.06 0.9 GB 2.6 GB Runs comfortably 114K 404
IQ1_M 1.75 0.8 GB 2.5 GB Runs comfortably 115K 435

Also worth checking

Llama 3.2 3B on every GPU
Full hardware tableEverything that runs on a RTX 5070 Ti
Full model tableWhy the KV cache is the part that bites
Guide