Qwen on NVIDIA Blackwell

Can I run Qwen3 32B on an NVIDIA DGX Spark (GB10)?

Yes. Qwen3 32B at Q4_K_M uses 21.2 GB of the 96.0 GB available on a NVIDIA DGX Spark (GB10), and runs at about 9.17 tokens per second. You can push the context to 128K.

Runs comfortably

21.2 GB of 96.0 GB · 22%
096 GB
Weights 18.6 GB
KV cache 2.0 GB
Runtime overhead 0.5 GB

Qwen3 32B at Q4_K_M leaves 74.8 GB spare on a NVIDIA DGX Spark (GB10). There is room to raise the context length or move up a quantisation level.

Generation9.17tok/s
Prompt processing801tok/s
Max context128Ktokens
KV per 1K tokens0GB

Every quantisation of Qwen3 32B 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 61.0 GB 63.6 GB Runs comfortably 128K 2.91
INT8 / W8A8 8.50 32.1 GB 34.6 GB Runs comfortably 128K 5.45
Q8_0 (GGUF) 8.50 32.1 GB 34.6 GB Runs comfortably 128K 5.45
FP8 (E4M3) 8.00 30.5 GB 33.0 GB Runs comfortably 128K 5.71
Q6_K 6.56 25.0 GB 27.6 GB Runs comfortably 128K 6.92
Q5_K_M 5.67 21.6 GB 24.2 GB Runs comfortably 128K 7.95
Q5_K_S 5.52 21.1 GB 23.6 GB Runs comfortably 128K 8.16
Q4_K_M 4.85 18.6 GB 21.2 GB Runs comfortably 128K 9.17
AWQ 4-bit 4.25 18.3 GB 20.9 GB Runs comfortably 128K 9.30
GPTQ 4-bit 4.25 18.3 GB 20.9 GB Runs comfortably 128K 9.30
MXFP4 4.25 18.3 GB 20.9 GB Runs comfortably 128K 9.30
Q4_K_S 4.58 17.6 GB 20.2 GB Runs comfortably 128K 9.65
Q4_0 4.55 17.5 GB 20.1 GB Runs comfortably 128K 9.70
IQ4_XS 4.25 16.4 GB 19.0 GB Runs comfortably 128K 10.3
Q3_K_M 3.91 15.2 GB 17.7 GB Runs comfortably 128K 11.1
IQ3_M 3.70 14.4 GB 17.0 GB Runs comfortably 128K 11.6
IQ3_XXS 3.06 12.1 GB 14.7 GB Runs comfortably 128K 13.7
Q2_K 2.63 10.6 GB 13.1 GB Runs comfortably 128K 15.5
IQ2_XXS 2.06 8.5 GB 11.0 GB Runs comfortably 128K 18.9
IQ1_M 1.75 7.4 GB 9.9 GB Runs comfortably 128K 21.4

Also worth checking