@shantanugoel: The config I arrived at after 2 days of sweeping through a bunch of hyper parameters and patches
Summary
Shantanu Goel shared a configuration recipe for optimizing Qwen 3.8 Flash Next on a single DGX Spark, tested for practical tasks and plans to benchmark it further.
View Cached Full Text
Cached at: 08/29/26, 05:59 AM
The config I arrived at after 2 days of sweeping through a bunch of hyper parameters and patches https://t.co/j5KH4WSDyX
Shantanu Goel (@shantanugoel): My Qwen 3.8 Flash Next NVFP4 recipe for a single dgx spark is working well now. Instead of reporting absurd tok/s number with lots of conditions, I tested it in configs that will work for actual tasks. Going to benchmark it with more known benchmarks too.
Similar Articles
Qwen3.8-Flash-Next NVFP4 2xDGX Spark config: 50t/s decode, 2,900t/s prefill
A user shares their optimized configuration for deploying the Qwen3.8-Flash-Next model on dual DGX Spark hardware, achieving up to 50t/s decode and 2,900t/s prefill speeds with technical patches and setup details.
Built this yesterday with Qwen3.8-Flash-Next (NVFP4, 262K context) on a single NVIDIA DGX Spark
A developer built a project in 8 hours using the Qwen3.8-Flash-Next model on a single NVIDIA DGX Spark, generating around 10k lines of code and consuming 800k tokens.
40+tok/s - optimized recipe for Qwen 3.5 122B Int4 on a single DGX Spark with vLLM
User shares an optimized recipe for running Qwen 3.5 122B Int4 on a single DGX Spark with vLLM, achieving over 40 tokens per second. They invite others to try and further optimize it.
DGX Spark agentic usage numbers
A user shares benchmark results and configuration for running Qwen3.6 models on NVIDIA DGX Spark using vLLM, focusing on agentic workloads with concurrent requests and tool calling.
Today I hit 181 toks/s (aggregate) on Qwen3.8-Flash-Next on 2x DGX Sparks
Achieved 181 tokens per second aggregate throughput on the Qwen3.8-Flash-Next model using a 2x DGX Spark cluster with optimizations like NVMe mapping and speculative decoding.