01.AI on NVIDIA Ampere

Can I run Yi-1.5 34B on an A100 80GB?

Yes. Yi-1.5 34B at Q4_K_M uses 22.2 GB of the 74.4 GB available on a A100 80GB, and runs at about 61.1 tokens per second. You can push the context to 32K.

Runs comfortably

22.2 GB of 74.4 GB · 30%
074 GB
Weights 19.5 GB
KV cache 1.9 GB
Runtime overhead 0.9 GB

Yi-1.5 34B at Q4_K_M leaves 52.2 GB spare on a A100 80GB. There is room to raise the context length or move up a quantisation level.

Generation61.1tok/s
Prompt processing1905tok/s
Max context32Ktokens
KV per 1K tokens0GB

Every quantisation of Yi-1.5 34B on a A100 80GB

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

QuantisationbpwWeights TotalVerdictMax ctxtok/s
FP16 / BF16 16.00 64.1 GB 66.8 GB Runs comfortably 32K 19.5
INT8 / W8A8 8.50 33.8 GB 36.6 GB Runs comfortably 32K 36.2
Q8_0 (GGUF) 8.50 33.8 GB 36.6 GB Runs comfortably 32K 36.2
FP8 (E4M3) 8.00 32.0 GB 34.8 GB Runs comfortably 32K 38.2
Q6_K 6.56 26.3 GB 29.0 GB Runs comfortably 32K 46.1
Q5_K_M 5.67 22.7 GB 25.5 GB Runs comfortably 32K 53.0
Q5_K_S 5.52 22.1 GB 24.9 GB Runs comfortably 32K 54.3
Q4_K_M 4.85 19.5 GB 22.2 GB Runs comfortably 32K 61.1
Q4_K_S 4.58 18.4 GB 21.2 GB Runs comfortably 32K 64.3
Q4_0 4.55 18.3 GB 21.1 GB Runs comfortably 32K 64.7
AWQ 4-bit 4.25 18.3 GB 21.0 GB Runs comfortably 32K 64.9
GPTQ 4-bit 4.25 18.3 GB 21.0 GB Runs comfortably 32K 64.9
MXFP4 4.25 18.3 GB 21.0 GB Runs comfortably 32K 64.9
IQ4_XS 4.25 17.2 GB 19.9 GB Runs comfortably 32K 68.8
Q3_K_M 3.91 15.8 GB 18.6 GB Runs comfortably 32K 74.1
IQ3_M 3.70 15.0 GB 17.8 GB Runs comfortably 32K 77.8
IQ3_XXS 3.06 12.5 GB 15.3 GB Runs comfortably 32K 91.7
Q2_K 2.63 10.8 GB 13.6 GB Runs comfortably 32K 104
IQ2_XXS 2.06 8.6 GB 11.4 GB Runs comfortably 32K 127
IQ1_M 1.75 7.4 GB 10.2 GB Runs comfortably 32K 145

Also worth checking

Yi-1.5 34B on every GPU
Full hardware tableEverything that runs on a A100 80GB
Full model tableWhy the KV cache is the part that bites
Guide