Zhipu on NVIDIA Blackwell

Can I run GLM-4 9B on an NVIDIA DGX Spark (GB10)?

Yes. GLM-4 9B at Q4_K_M uses 6.2 GB of the 96.0 GB available on a NVIDIA DGX Spark (GB10), and runs at about 32.2 tokens per second. You can push the context to 128K.

Runs comfortably

6.2 GB of 96.0 GB · 6%
096 GB
Weights 5.4 GB
KV cache 0.3 GB
Runtime overhead 0.5 GB

GLM-4 9B at Q4_K_M leaves 89.8 GB spare on a NVIDIA DGX Spark (GB10). There is room to raise the context length or move up a quantisation level.

Generation32.2tok/s
Prompt processing2793tok/s
Max context128Ktokens
KV per 1K tokens0GB

Every quantisation of GLM-4 9B 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 17.5 GB 18.3 GB Runs comfortably 128K 10.2
INT8 / W8A8 8.50 9.0 GB 9.9 GB Runs comfortably 128K 19.6
Q8_0 (GGUF) 8.50 9.0 GB 9.9 GB Runs comfortably 128K 19.6
FP8 (E4M3) 8.00 8.8 GB 9.6 GB Runs comfortably 128K 20.2
Q6_K 6.56 7.2 GB 8.0 GB Runs comfortably 128K 24.5
AWQ 4-bit 4.25 6.3 GB 7.2 GB Runs comfortably 128K 27.6
GPTQ 4-bit 4.25 6.3 GB 7.2 GB Runs comfortably 128K 27.6
MXFP4 4.25 6.3 GB 7.2 GB Runs comfortably 128K 27.6
Q5_K_M 5.67 6.2 GB 7.0 GB Runs comfortably 128K 28.2
Q5_K_S 5.52 6.0 GB 6.9 GB Runs comfortably 128K 28.9
Q4_K_M 4.85 5.4 GB 6.2 GB Runs comfortably 128K 32.2
Q4_K_S 4.58 5.1 GB 6.0 GB Runs comfortably 128K 33.8
Q4_0 4.55 5.1 GB 5.9 GB Runs comfortably 128K 33.9
IQ4_XS 4.25 4.8 GB 5.7 GB Runs comfortably 128K 35.9
Q3_K_M 3.91 4.5 GB 5.3 GB Runs comfortably 128K 38.3
IQ3_M 3.70 4.3 GB 5.1 GB Runs comfortably 128K 40.0
IQ3_XXS 3.06 3.7 GB 4.5 GB Runs comfortably 128K 46.3
Q2_K 2.63 3.3 GB 4.1 GB Runs comfortably 128K 51.7
IQ2_XXS 2.06 2.8 GB 3.6 GB Runs comfortably 128K 61.1
IQ1_M 1.75 2.5 GB 3.3 GB Runs comfortably 128K 67.9

Also worth checking

GLM-4 9B on every GPU
Full hardware tableEverything that runs on a NVIDIA DGX Spark (GB10)
Full model tableWhy the KV cache is the part that bites
Guide