Spark Bench

DeepSeek V4 Flash DSpark: original July deployment, two Sparks (TP=2), thinking off

2× DGX Spark · vLLM (Anemll DSpark) ·

Tested configuration

Hardware
2x NVIDIA DGX Spark (128GB each)
Serving runtime
vLLM (Anemll DSpark)
Runtime version or build
the prebuilt ghcr.io/anemll/dspark-vllm-gx10:0.1.1 image (image ID 3430d6614a8e, repo digest sha256:a83948492cf13df455170fb42885f5ef4db54fefe0feff0f841ecbff464ac9d8, the same digest on both nodes). It was already present locally from the original deployment
Weight quantization or precision
Unavailable
Benchmark date
Recipe credit
MiaAI-Lab DeepSeek-v4-Flash DSpark 2x DGX Spark recipeBy MiaAI-Labb4cc551b0adb6aa62979d92686b9fe0f1326d640

Decode sweep

~256 input / 512 output tokens per request. Output throughput is aggregate across concurrent requests. TTFT means time to first token.

ConcurrencyOutput tok/sTTFT p95 (s)
142.124.12
267.920.83
481.949.80
892.5228.08
1299.2135.92
16104.8255.35

Prefill and context

Configured request window
1,048,576 tokens
Longest tested prompt
245,760 target input tokens
Validated retrieval
≈240K input tokens · passkey passed

Input throughput uses full request wall time. The passkey measures retrieval from one long prompt.

Workload (c1)Input tok/sTTFT p95 (s)
4,096 input tokens1,014.622.42
32,768 input tokens1,304.9030.62
131,072 input tokens1,132.52116.92
245,760 input tokensPasskey prompt1,400.67175.40

Sustained load

20-minute soak at concurrency 8 · ~1024 input / 512 output tokens per request.

Aggregate throughput
111.10 output tok/s
TTFT p95
20.24 s
Completed requests
266
Request errors
0

Serving regression checks 13/13 passed

Tool-use evaluation

Complete · 100% completionSafety gate passed

Compare only against complete results with 100% completion under the same protocol ID.

Protocol ID
tool-eval-standard-v1-b60e2f0
Workload
69 mock-tool scenarios per trial · 3 trials
Mean score
88.3 / 100
Pass@3 · passed at least once
84.1%
Pass³ · passed every trial
78.3%

Read the tool-eval report and scenario findings