01.AI on NVIDIA Blackwell

Can I run Yi-1.5 34B on an NVIDIA DGX Spark (GB10)?

Yes. Yi-1.5 34B at Q4_K_M uses 21.9 GB of the 96.0 GB available on a NVIDIA DGX Spark (GB10), and runs at about 8.80 tokens per second. You can push the context to 32K.

Runs comfortably

21.9 GB of 96.0 GB · 23%
096 GB
Weights 19.5 GB
KV cache 1.9 GB
Runtime overhead 0.6 GB

Yi-1.5 34B at Q4_K_M leaves 74.1 GB spare on a NVIDIA DGX Spark (GB10). There is room to raise the context length or move up a quantisation level.

Generation8.80tok/s
Prompt processing763tok/s
Max context32Ktokens
KV per 1K tokens0GB

Every quantisation of Yi-1.5 34B on a NVIDIA DGX Spark (GB10)

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

QuantisationbpwWeights TotalVerdictMax ctxtok/s
FP16 / BF16 16.00 64.1 GB 66.5 GB Runs comfortably 32K 2.77
INT8 / W8A8 8.50 33.8 GB 36.3 GB Runs comfortably 32K 5.18
Q8_0 (GGUF) 8.50 33.8 GB 36.3 GB Runs comfortably 32K 5.18
FP8 (E4M3) 8.00 32.0 GB 34.5 GB Runs comfortably 32K 5.46
Q6_K 6.56 26.3 GB 28.7 GB Runs comfortably 32K 6.62
Q5_K_M 5.67 22.7 GB 25.2 GB Runs comfortably 32K 7.61
Q5_K_S 5.52 22.1 GB 24.6 GB Runs comfortably 32K 7.81
Q4_K_M 4.85 19.5 GB 21.9 GB Runs comfortably 32K 8.80
Q4_K_S 4.58 18.4 GB 20.9 GB Runs comfortably 32K 9.28
Q4_0 4.55 18.3 GB 20.8 GB Runs comfortably 32K 9.34
AWQ 4-bit 4.25 18.3 GB 20.7 GB Runs comfortably 32K 9.36
GPTQ 4-bit 4.25 18.3 GB 20.7 GB Runs comfortably 32K 9.36
MXFP4 4.25 18.3 GB 20.7 GB Runs comfortably 32K 9.36
IQ4_XS 4.25 17.2 GB 19.6 GB Runs comfortably 32K 9.94
Q3_K_M 3.91 15.8 GB 18.3 GB Runs comfortably 32K 10.7
IQ3_M 3.70 15.0 GB 17.5 GB Runs comfortably 32K 11.3
IQ3_XXS 3.06 12.5 GB 15.0 GB Runs comfortably 32K 13.3
Q2_K 2.63 10.8 GB 13.3 GB Runs comfortably 32K 15.2
IQ2_XXS 2.06 8.6 GB 11.1 GB Runs comfortably 32K 18.8
IQ1_M 1.75 7.4 GB 9.9 GB Runs comfortably 32K 21.5

Also worth checking