# Some new development work for Qwen3 on the Spark **URL:** \ **Category:** DGX Spark / GB10\ **Created:** [January 29, 2026, 9:07pm UTC](https://forums.developer.nvidia.com/t/some-new-development-work-for-qwen3-on-the-spark/359100 "2026-01-29T21:07:23Z")\ **Posts on this page:** 6\ **Page:** 1
**Author:** ![christopher\_owen](https://sea2.discourse-cdn.com/nvidia/user_avatar/forums.developer.nvidia.com/christopher_owen/32/469895_2.png) [@christopher\_owen](https://forums.developer.nvidia.com/u/christopher_owen)\ **Post date:** [January 29, 2026, 9:07pm UTC](https://forums.developer.nvidia.com/t/some-new-development-work-for-qwen3-on-the-spark/359100/1 "2026-01-29T21:07:23Z")
[https://github.com/seli-equinix/docker-swarm-stacks/tree/main/asus-dgx-spark/vllm-docker](https://github.com/seli-equinix/docker-swarm-stacks/tree/main/asus-dgx-spark/vllm-docker) I came across this. The developer is interested in feedback. He says: I’ve got vLLM fully working on the **NVIDIA DGX Spark** (Latest OS update) with the new GB10 Blackwell GPU (SM121). I have it running v0.14.0rc1 code as of this morning. It is a complete implementation with all the SM121-specific fixes needed to run on this hardware. 🎯 What’s Working ✅ **Full 256K context length** (model’s max capacity) ✅ **~45 tok/s** on Qwen3-Next-80B-A3B-FP8 ✅ **FP8 quantization** with Triton MoE backend ✅ **Blackwell-class detection** (`is_blackwell_class()` for SM10x/SM11x/SM12x) - ✅ **Proper backend fallbacks** (TRTLLM → CUTLASS → Triton) 🔧 Key Technical Changes The GB10 is SM121 (major=12), different from B100/B200 which are SM100/SM103 (major=10). This required: 1. **New `is_blackwell_class()` method** - Unified detection for all Blackwell variants 2. **TRITON\_ATTN backend** - FlashInfer TRTLLM doesn’t support SM121 yet 3. **Correct backend gating** - TRTLLM/CUTLASS MLA restricted to SM100 only 4. **KV cache layout fix** - HND layout for SM121 like SM100 📦 Pre-built Docker Image (Easiest) ```auto # Pull the image docker pull hellohal2064/vllm-dgx-spark-gb10:latest # Run docker run -d --name vllm-server --gpus all -p 8000:8000 \ -v /path/to/models:/models:ro \ -e MODEL_PATH=/models/Qwen3-Next-80B-A3B-FP8 \ -e ATTENTION_BACKEND=TRITON_ATTN \ -e MAX_MODEL_LEN=262144 \ -e GPU_MEMORY_UTIL=0.85 \ hellohal2064/vllm-dgx-spark-gb10:latest ``` 🛠️ Build From Source ```auto # Clone Docker setup git clone https://github.com/seli-equinix/docker-swarm-stacks.git cd docker-swarm-stacks/asus-dgx-spark/vllm-docker # Clone vLLM with SM121 support git clone https://github.com/seli-equinix/vllm.git cd vllm && git checkout feature/sm121-gb10-support && cd .. # Build docker build -t vllm-gb10:latest . ``` ⚙️ Environment Variables Variable Description Recommended`MODEL_PATH`Model path in container`/models/YourModelATTENTION_BACKEND` **Must be TRITON\_ATTN for GB10** `TRITON_ATTNMAX_MODEL_LEN`Context length (up to 256K)`262144GPU_MEMORY_UTIL`GPU memory fraction`0.85` 📊 Hardware Specs (DGX Spark) Component Spec GPUNVIDIA GB10 (SM121 Blackwell)Compute Capability 12.1 Memory128GB unified (CPU+GPU shared) CPUARM64 NVIDIA Grace (20 cores) CUDA13.1+ required 🔗 Links **Docker Image** `hellohal2064/vllm-dgx-spark-gb10:latest` **Docker Setup & README** [https://github.com/seli-equinix/docker-swarm-stacks/tree/main/asus-dgx-spark/vllm-docker](https://github.com/seli-equinix/docker-swarm-stacks/tree/main/asus-dgx-spark/vllm-docker) **vLLM Fork (SM121 branch)** [https://github.com/seli-equinix/vllm/tree/feature/sm121-gb10-support](https://github.com/seli-equinix/vllm/tree/feature/sm121-gb10-support) **Upstream PR** #31740 [https://github.com/vllm-project/vllm/pull/31740](https://github.com/vllm-project/vllm/pull/31740) ⚠️ Known Limitations - FlashInfer TRTLLM attention not supported on SM121 (uses Triton) - MoE configs not tuned for GB10 yet (works with defaults) - DeepGEMM not supported on SM121 🙏 Looking For - **Testers with DGX Spark hardware** - Please try it and report issues! - **Review on PR** #31740 - Would appreciate maintainer feedback - **MoE tuning help** - Anyone interested in generating GB10-optimized configs? Happy to answer questions. I am working to build the MoE Config for the GB10/SM121 Spark ---
**Author:** ![space11](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@space11](https://forums.developer.nvidia.com/u/space11)\ **Post date:** [February 2, 2026, 8:48am UTC](https://forums.developer.nvidia.com/t/some-new-development-work-for-qwen3-on-the-spark/359100/2 "2026-02-02T08:48:37Z")
404 – ---
**Author:** ![notmy.reward438](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@notmy.reward438](https://forums.developer.nvidia.com/u/notmy.reward438)\ **Post date:** [February 2, 2026, 2:46pm UTC](https://forums.developer.nvidia.com/t/some-new-development-work-for-qwen3-on-the-spark/359100/3 "2026-02-02T14:46:00Z")
> [@christopher\_owen](#): > > v0.14.0rc1 v0.14.0rc1 or 0.15 release candidate? 0.14 release was out few weeks ago. Looks promising though! Did you try Qwen3 Coder 30B by any chance? ---
**Author:** ![kyrylo.gorbachov](https://sea2.discourse-cdn.com/nvidia/user_avatar/forums.developer.nvidia.com/kyrylo.gorbachov/32/465121_2.png) [@kyrylo.gorbachov](https://forums.developer.nvidia.com/u/kyrylo.gorbachov)\ **Post date:** [February 2, 2026, 4:40pm UTC](https://forums.developer.nvidia.com/t/some-new-development-work-for-qwen3-on-the-spark/359100/4 "2026-02-02T16:40:47Z")
> [@christopher\_owen](#): > > [seli-equinix (Sean Li) · GitHub](https://github.com/seli-equinix) Sounds very interesting, but seems that repo is not available anymore. ---
**Author:** ![kyrylo.gorbachov](https://sea2.discourse-cdn.com/nvidia/user_avatar/forums.developer.nvidia.com/kyrylo.gorbachov/32/465121_2.png) [@kyrylo.gorbachov](https://forums.developer.nvidia.com/u/kyrylo.gorbachov)\ **Post date:** [February 2, 2026, 4:42pm UTC](https://forums.developer.nvidia.com/t/some-new-development-work-for-qwen3-on-the-spark/359100/5 "2026-02-02T16:42:52Z")
> [@christopher\_owen](#): > > `hellohal2064/vllm-dgx-spark-gb10:latest` I think is that one: > **[GitHub - seli-equinix/vllm at feature/sm121-gb10-support](https://github.com/seli-equinix/vllm/tree/feature/sm121-gb10-support)** > > A high-throughput and memory-efficient inference and serving engine for LLMs - seli-equinix/vllm and > **[hellohal2064/vllm-dgx-spark-gb10 - Docker Image](https://hub.docker.com/r/hellohal2064/vllm-dgx-spark-gb10)** ---
**Author:** ![hellohal2064](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@hellohal2064](https://forums.developer.nvidia.com/u/hellohal2064)\ **Post date:** [February 3, 2026, 4:02pm UTC](https://forums.developer.nvidia.com/t/some-new-development-work-for-qwen3-on-the-spark/359100/6 "2026-02-03T16:02:45Z")
hello - am the one working on getting this into the main build for vllm - I updated it yesterday. It should all be there. [feat: Add SM121/GB10 (DGX Spark) Blackwell-class GPU support by seli-equinix · Pull Request #31740 · vllm-project/vllm · GitHub](https://github.com/vllm-project/vllm/pull/31740) - [feature/sm121-gb10-support](https://github.com/seli-equinix/vllm/tree/feature/sm121-gb10-support) is the correct fork. I just pushed a new container to dockerhub, it is running the latest 0.16 vllm and i have gotten it to load the qwen models much faster. I now have it running with YaRN and a functioning 1M context window. I also updated all the docs on dockerhub.