|
DeepSeek-V4-Flash on 2x DGX Spark throughput collapses at large prompt+concurrency. Would a 4x stack + Vision-Exp fix it?
|
|
1
|
400
|
November 1, 2026
|
|
Qwen3.8-27B on one DGX Spark: 38 tok/s with vLLM + DFlash2, why MTP wins under load, and why your Ollama comparison is probably skewed
|
|
0
|
379
|
October 29, 2026
|
|
nvidia/GLM-5.3-Flash-FP4-Local with 2M tokens of FP8 KV Cache
|
|
20
|
2555
|
October 24, 2026
|
|
My DGX Spark extra cooling solution
|
|
16
|
9606
|
October 19, 2026
|
|
Three times ( VoiceClone | VoiceDesign | CustomVoice ) - Faster-Qwen3-TTS for NVIDIA DGX Spark (GB10)
|
|
63
|
4892
|
October 11, 2026
|
|
HOW-TO: setup-dgx-spark docker inference - A "Sane" Inference Stack for GB10 (Need Contributors!)
|
|
40
|
4248
|
October 10, 2026
|
|
Recipe: NOOA on DGX Spark (VSCode)
|
|
0
|
176
|
October 3, 2026
|
|
ConnectX-7 bricked — stuck in pre-init / static_config_not_done after unsolicited mlnx-fw-updater firmware flash — ASUS GX10 [error -110]
|
|
25
|
2241
|
September 16, 2026
|
|
ConnectX-7 network cards disappear after DGX Spark system update due to cx7-pcie-hotplug driver issue
|
|
3
|
751
|
September 2, 2026
|
|
Qwen3.8-27B on single DGX Spark: 1.9x stock NVFP4 (2.5x FP8) at 16 concurrent, ~50 tok/s on a single request
|
|
7
|
1674
|
September 1, 2026
|
|
A Faster (maybe better?) Qwen 3.5 122B
|
|
9
|
1401
|
August 31, 2026
|
|
Qwen3.8-Flash-Next NVFP4 on 4x DGX Spark — vLLM TP4 serving, 4.7M-token KV pool, and the three fixes you WILL need
|
|
0
|
722
|
August 31, 2026
|
|
DGX Spark Cluster Dashboard
|
|
5
|
918
|
August 29, 2026
|
|
Single-Spark setups — which models do you actually run for coding, and how? (sharing mine + a test prompt)
|
|
60
|
11208
|
August 19, 2026
|
|
Running Kimi K3 on 24 DGX Spark Systems — TP24 vs TP8/PP3 Advice
|
|
39
|
4777
|
August 14, 2026
|
|
Deepseekv4 flash with QWEN for vision
|
|
2
|
862
|
August 12, 2026
|
|
Lowest KLD DeepSeek-V4-Flash-0731 on a single GB10 (drafter converted, 31 t/s on code), 256k context recipe
|
|
0
|
527
|
August 8, 2026
|
|
Sparkup - DGX Spark as code
|
|
0
|
406
|
August 5, 2026
|
|
Deepseek 4 0731 Single Spark - 40t/s decode - 131k ctx - Dwarfstar CUDA custom engine
|
|
3
|
825
|
August 5, 2026
|
|
1x DGX Spark: DeepSeek-V4-Flash-0731 on llama.cpp: 131K ctx/slot, 19.7 tok/s single-stream, 52 tok/s @ 4 concurrent
|
|
3
|
1516
|
August 5, 2026
|
|
Resetting a single GPU in a DGX node with gpu-operator running
|
|
2
|
443
|
August 4, 2026
|
|
6× DGX Spark, 200G RoCE fabric — My learning Journey
|
|
1
|
698
|
July 30, 2026
|
|
Best Daily Single Spark Driver - King Model - Still Qwen 3.5 122b?
|
|
15
|
2282
|
July 27, 2026
|
|
Super Idol Master: Multi-Agent Driven 3D Character Asset Pipeline on DGX Spark
|
|
0
|
436
|
July 22, 2026
|
|
My DGX Spark Hangs ... is this normal?
|
|
5
|
1433
|
July 15, 2026
|
|
Soft lockup in nvidia_modeset DisplayPort path during Xorg logout (DGX Spark, kernel 6.17.0-1018-nvidia)
|
|
3
|
509
|
July 2, 2026
|
|
Quick Test Results Regarding DGX SPARK Temperatures
|
|
2
|
1427
|
June 30, 2026
|
|
Docker Image: NVIDIA vLLM 0.23.0 with Claude Code 2.1.195+ Compatibility
|
|
6
|
654
|
June 29, 2026
|
|
2 node spark vs 3 or 4 node spark
|
|
18
|
2709
|
June 29, 2026
|
|
New model (one of the best i ever used, in my humble opinion): Qwen Agent World for local programming
|
|
0
|
761
|
June 29, 2026
|