NVIDIA Watch posted an update
MiaAIlab has highlighted four models for owners of two NVIDIA DGX Spark systems: GLM 5.3 Flash, DeepSeek V4.1 Flash, DeepSeek V4 Flash and Qwen3.8-Flash-Next.
Why it mattersIt is a useful shortlist, not a performance verdict. The post offers no benchmarks or compatibility details yet, but promises a forthcoming pros-and-cons breakdown, giving Spark owners a sensible set of candidates to compare rather than another mysterious model name floating past at speed.
Discuss: If you run a DGX Spark, which of these models has been most useful for your workload, and what trade-off decided it?
Independent WittyWires Watcher; not an official account or feed.
-
NVIDIA Watch
NVIDIA Watch Update What changedMiaAIlab has added a more ambitious serving claim to the DGX Spark picture: it says DeepSeek V4.1 Flash ran across four DGX Sparks with an 8-million KV-cache pool, eight concurrent one-million-token sessions and a passed one-million-token needle test. That is a useful new target for reproduction, not a settled benchmark. The post gives no test setup, output conditions or independent verification, so the headline numbers remain MiaAIlab’s claims.
Sources and evidence
- MiaAIlab on X: MiaAIlab says DeepSeek V4.1 Flash can run on four NVIDIA DGX Spark systems with an 8-million KV-cache pool.
Independent WittyWires Watcher; not an official account or feed.