GLM-5.3-Flash on 2× DGX Spark: 373.58 GiB does not fit one Spark
This URL is the GLM-5.3-Flash 2× DGX Spark path. GLM-5.3-Flash is not GLM-5.3. The catalog lists 321,323,031,390 parameters for zai-org/GLM-5.3-Flash. Auto-selected GGUF Q8_0 (Unsloth) is 340,981,966,112 bytes. At 8,192 tokens, one stream, the calculator estimates 318.04 GiB–373.58 GiB. Rough estimate, not a measurement. Official FP8 safetensors are 328,326,771,576 bytes (306.75 GiB–355.00 GiB). One NVIDIA DGX Spark (GB10, 128 GB unified) has a net budget of 112.00 GiB and does not fit GGUF Q8_0 (Unsloth). None of the 32 catalog devices fit GGUF Q8_0 (Unsloth). The largest unified devices are NVIDIA DGX Spark (GB10, 128 GB unified) and Apple M4 Max 128GB unified (MacBook Pro) and Ryzen AI Max+ 395. The largest discrete devices are NVIDIA A100 80GB SXM and NVIDIA H100 80GB SXM. Discrete GPUs are not summed, and unified memory is not added on top of discrete VRAM. This catalog has no EXL3 file for GLM-5.3-Flash, so this page does not restore a removed EXL3 recipe. GLM-5.3 GGUF Q8_0 (Unsloth) estimates 747.13 GiB–876.98 GiB at the same context. No can-i-run page is allowlisted for GLM-5.3 or GLM-5.3-Flash.