DeepSeek-V4.1-Flash benchmarked on dual DGX Spark hardware
Nick Mermiges of PLI Labs benchmarked DeepSeek-V4.1-Flash-1017 at 2.75 bpw with EXL3 and vLLM across two NVIDIA DGX Spark systems. An EPYC system with two RTX 3090s handled Engram tables and speculative drafting. In the author's tests, reported as medians of three runs, prefill reached 2,618 tokens per second with a 256K prompt, while structured JSON output reached 322 tokens per second across four streams. The report published the code, launch scripts, benchmark tool and raw results, while noting lower prose throughput and slower decoding than some cited deployments. Planned 1M-token context, llama-benchy, BFCL and head-to-head KLD tests were not yet available, and the prefix-cache code was still pending publication.
The published configuration, scripts and raw benchmark data provide a reference for running and evaluating this DeepSeek-V4.1-Flash deployment on compact multi-GPU hardware.
Update from the lawyer with the V100 server. I went and got two DGX Sparks, because of course I did. The mid-life crisis remains on schedule.
Same deal as before: still a lawyer, still driving the whole thing through Claude Code, and Claude Code also ran the benchmarks, built these tables and helped me write this up. I'm Nick Mermiges, Esq., founder of PLI Labs. The repo below is ours.
The setup: DeepSeek-V4.1-Flash-1017 at 2.75 bpw EXL3, vLLM tensor parallel across two Sparks. The twist is a third box, my EPYC with the 3090s, doing two jobs for the Sparks: - it holds the Engram tables in system RAM and serves them over RDMA; - it runs the DSpark speculative drafter on an NVLinked pair of 3090s.
The memory that frees on the Sparks goes to KV cache: about 1.93M tokens of pool. There's also a prefix cache on the Sparks' NVMe, so a 150K prompt survives a server restart. First token comes back in 2.58 s instead of 58.6 s cold.
Numbers. I re-implemented MiaAI's sparkDash protocol so this would be apples to apples with the most popular 2x Spark build. Sparks at 2300 MHz, 3 reps, medians, today.
#DeepSeek-V4.1-Flash-1015#PLI Labs#NVIDIA DGX Spark#NVIDIA RTX 3090#vLLM#EXL3#Claude Code