# DeepSeek v4.1 Flash

**URL:** <https://forums.developer.nvidia.com/t/deepseek-v4-1-flash/382725>\
**Category:** DGX Spark / GB10\
**Tags:** deepseek\
**Created:** [September 8, 2026, 5:15pm UTC](https://forums.developer.nvidia.com/t/deepseek-v4-1-flash/382725 "2026-09-08T17:15:40Z")\
**Posts on this page:** 1\
**Showing post:** 11

<div class="post-metadata">

**Author:** ![PILCOTHINK](https://sea2.discourse-cdn.com/nvidia/user_avatar/forums.developer.nvidia.com/pilcothink/32/528945_2.png) [@PILCOTHINK](https://forums.developer.nvidia.com/u/PILCOTHINK)\
**Post date:** [September 10, 2026, 7:07am UTC](https://forums.developer.nvidia.com/t/deepseek-v4-1-flash/382725/11 "2026-09-10T07:07:42Z")

</div>

It looks like a lot of testing will be needed to run it on dual Sparks.

In particular, we don’t yet know how the higher active parameter count during decoding, n-gram offloading, and DSpark will interact when quantization is also involved. It seems we’ll need many people to test this configuration and share their results.

---

_[View the full topic](https://forums.developer.nvidia.com/t/deepseek-v4-1-flash/382725)._