ggml CUDA on DGX Spark GB10: GLM-4.7-Flash Q8_0 pp2048 d16384 (b7941)
914.05 tok/s (pp2048, batch 1, per-request)
Fields
id
benchmarks/ggml-cuda-gb10-spark-glm47-flash-pp2048-d16384-b7941Metric
prefill_tok_s
Value
914.05
Unit
tok/s (pp2048, batch 1, per-request)
Model
GLM-4.7-Flash Q8_0 (ggml-org/GLM-4.7-Flash-GGUF; llama-bench labels the arch deepseek2 30B.A3B)
Methodology
MAINTAINER bench on one NVIDIA DGX Spark (GB10), build 11fb327bf (b7941). CUDA backend, ngl 99, flash attention on, ubatch 2048, mmap off, direct-io on. pp2048 is prefill of a 2048-token prompt, batch 1, per-request, NOT aggregate. tg32 is batch-1 decode of 32 new tokens, per-request. Depth d=16384 is the same measurement after 16384 tokens of prior context (plain row is depth 0). Value is the published mean; stddev is in notes. Single device, unified memory, no tensor parallel.
Measured by
third_party
Reproducible
yes
notes
Official ggml DGX Spark table, not a forum paste. stddev +/- 3.32. File size 29.65 GiB, params 29.94 B. A concurrent sweep at batch 1 through 32 is also on the page; those figures are aggregate and are not this record. Distinct from the earlier third-party Spark gpt-oss-120b CUDA sweep (different build, lower tok/s).
Notes
Official ggml DGX Spark table, not a forum paste. stddev +/- 3.32. File size 29.65 GiB, params 29.94 B. A concurrent sweep at batch 1 through 32 is also on the page; those figures are aggregate and are not this record. Distinct from the earlier third-party Spark gpt-oss-120b CUDA sweep (different build, lower tok/s).