benchmarks / qwen3-6-35b-a3b--dgx-spark-x1--llama-cpp--q4-k-m--dflash-greedy-baseline--p2048-o256-c1--2026-07-06
Qwen3.6 35B-A3B on NVIDIA DGX Spark
compare66.7tok/s
1.04sttft
2.1kprefill
llama.cpp b15-ee445f9Q4_K_Mspec-decodingJul 6, 2026
Run configuration
nodes
1
batch size
—
concurrency
1
context
10240
prompt / output
2041 / 256
weights unsloth/Qwen3.6-35B-A3B-GGUF ↗
Metrics
decode tok/s
66.7
prefill tok/s
2.1k
ttft
1.04s
total tok/s
52.5
inter-token
15ms
e2e latency
4.88s
req/s
0.205
peak memory
29.7 GB
avg power
41.7 W
energy/req
203.4 J
temperature
0
Reproduce
~/Dev/llama.cpp/build/bin/llama-server -m ~/models/gguf/Qwen3.6-35B-A3B-UD-Q4_K_M.gguf -c 10240 --parallel 1 -ngl 99 -fa on --host 127.0.0.1 --port 8080 python3 bench/harness.py bench/configs/qwen3-6-35b-a3b--dflash-baseline--llama-cpp.json --rows-out data/benchmarks/qwen3-6-35b-a3b.json
spec-decoding profile: greedy no-draft Q4_K_M baseline for the DFlash comparison (original target).