Building on the amazing work of @eugr / @eugr_nv , @arthurdroz , @jasl and everyone else in this community, I wrote this blog post: DeepSeek V4 Flash: Bringing Frontier AI to the Home Hope it’s useful!

Can you comfortablely get 1M context with TP=2? The recipe you referred to was using 200K max_model_len.

I believe GB10 *2 max context would be around 512K, memory-limited, wish GB10 successor can give us 256GB+ memory, and at least RTX Pro 6000 class DIE size

@davwu The startup log says: (EngineCore pid=158) INFO 05-17 13:06:07 [kv_cache_utils.py:1710] GPU KV cache size: 820,713 tokens (EngineCore pid=158) INFO 05-17 13:06:07 [kv_cache_utils.py:1711] Maximum concurrency for 200,000 tokens per request: 4.10x so this is looking good for a 800K token, fou…

@davwu If I configure the recipe with the maximum context size (1048576) and tell vLLM to use 0.9 of my RAM then it seems to run stably, and offers just short of 4 concurrent requests: (APIServer pid=84) INFO 05-17 13:42:09 [model.py:1697] Using max model len 1048576 [snip] (Worker_TP0_EP0 pid=207)…

@jasl do you think supporting long contexts will be eventually supported? Is the slow prefill something that can be developed away or is there something about the deepseek architecture that makes that really difficult on spark?

I’ll try my best. The major challenge is that Spark 5070-grade design and the limited memory bandwidth. Luckily, profiling metrics show we still have room to optimize, so it’s possible, but we need time to make it true. I super wish next gen we could have an Apple Ultra class SoC. My equipment i…

Slight nuance to my response. It works stably up to 3 concurrent requests (66 tps overall, 22 tps per request). When I try 4 concurrent Inspect Evals connections, the performance fluctuates terribly: [DGX Dashboard with fluctuating GPU utilization] We have clearly reached a resource bottleneck. I…

I managed to run DeepSeek V4 Flash locally on a single GX10/Spark using a llama.cpp-based setup rather than vLLM. My working configuration is based on antirez/ds4 with the DeepSeek-V4-Flash GGUF quantized build. I run it as a user systemd service exposing an OpenAI-compatible endpoin then connect …

Thanks @marco.palaferri . In response to your and @jasl ’s comments I’ve updated the blog to highlight this scaling behavior that we’ve seen. I also corrected it to say that both TP and EP are used.

DeepSeek V4 Flash: Bringing Frontier AI to the Home

Accelerated Computing DGX Spark / GB10 User Forum DGX Spark / GB10

jasl May 17, 2026, 3:07pm 9

Decode is memory bandwidth-limited.

Topic		Replies	Views
Anyone having luck with Deepseek V4 Flash on Dual Sparks? DGX Spark / GB10 deepseek	13	1227	June 4, 2026
Deepseek v4 Flash on 2 Nodes DGX Spark / GB10 Projects deepseek	70	5366	June 12, 2026
DeepSeek-V4-Flash (official FP8) running across 2x DGX Spark — TP=2, MTP, 200K ctx, recipe + numbers DGX Spark / GB10 deepseek	219	12811	June 12, 2026
DeepSeek v4 Flash (Aiden Recipe from Reddit) - 1M token session operational, Cuda 12.1 tailored for DGX Spark GB10 DGX Spark / GB10 deepseek	82	3921	June 13, 2026
Deepseek V4 released DGX Spark / GB10 deepseek	143	15622	May 18, 2026
DeepSeek v4 Flash (IQ2XXS) on a single GB10! DGX Spark / GB10 Projects llm , llama , deepseek	9	3094	June 13, 2026
Fully custom CUDA-native Deepseek 4 Flash optimized for 1x Spark! antirez/ds4 DGX Spark / GB10 Projects gaming , llama , deepseek	67	5631	June 12, 2026
DeepSeekV4-Flash hybrid quant, 1x DGX Spark: antirez's optimized 128 GB MLX recipe ported to vLLM for GB10 DGX Spark / GB10 Projects deepseek	18	1759	May 11, 2026
DeepSeek V4 Flash MXFP4 proof-of-life on a single GB10/GX10 DGX Spark / GB10 cuda , kernel , deepseek	4	1256	May 8, 2026
Does Qwen3.5-35B-A3B on GB10 leave a lot of performance on the table? DGX Spark / GB10 agentic-ai	40	5902	March 16, 2026

DeepSeek V4 Flash: Bringing Frontier AI to the Home

Related topics