Measured work on DGX Spark, PyTorch, distributed systems and NLP — by Roman Bartusiak.https://bartusiak.ai/enhttps://bartusiak.ai/2025/11/07/is-it-worth-2/https://bartusiak.ai/2025/11/07/is-it-worth-2/Following up with fp32, fp16 and bf16 across GEMM, convolution and TorchBench: where the NGC container actually wins, and where the DGX Spark is simply bandwidth-bound.Fri, 07 Nov 2025 00:00:00 GMTDGX SparkGPUPerformanceNVIDIAPyTorchhttps://bartusiak.ai/2025/11/04/is-it-worth/https://bartusiak.ai/2025/11/04/is-it-worth/An 8192² bfloat16 matmul run 100 times, inside NVIDIA's NGC PyTorch container and against a stock pip install, measured on the same DGX Spark.Tue, 04 Nov 2025 00:00:00 GMTDGX SparkGPUPerformanceNVIDIAPyTorchhttps://bartusiak.ai/2025/10/31/dgx-memory/https://bartusiak.ai/2025/10/31/dgx-memory/A 7B model OOMs on a 128 GB machine. The memory is not missing — the OS page cache is holding it, and unified memory hides that from the usual GPU tooling.Fri, 31 Oct 2025 00:00:00 GMTDGX SparkGPUhttps://bartusiak.ai/2025/10/28/unsloth-spark/https://bartusiak.ai/2025/10/28/unsloth-spark/Getting Unsloth to build and run on the DGX Spark's GB10: the NGC PyTorch image, uv with pinned overrides, and the TORCH_CUDA_ARCH_LIST setting that makes xformers compile.Tue, 28 Oct 2025 00:00:00 GMTDGX SparkUnslothNVIDIAPyTorchLLM Tuning