"But the machine is far from the fastest GPU in Nvidia's lineup. It's not going to beat out an RTX 5090 in large language model (LLM) inference, fine tuning, or even image generation - never mind gaming. What the DGX Spark, and the slew of GB10-based systems hitting the market tomorrow, can do is run models the 5090 or any other consumer graphics card on the market today simply can't."
"When it comes to local AI development, all the FLOPS and memory bandwidth in the world won't do you much good if you don't have enough VRAM to get the job done. Anyone who has tried machine learning workloads on consumer graphics will have run into CUDA out of memory errors on more than one occasion. The Spark is equipped with 128 GB of memory, the most of any workstation GPU in Nvidia's portfolio."
Nvidia's DGX Spark is an Arm-based mini‑PC priced around $3,000–$4,000 that emphasizes memory capacity rather than peak GPU speed. The unit is not as fast as high‑end consumer GPUs like the RTX 5090 for LLM inference, fine‑tuning, image generation, or gaming. The Spark ships with 128 GB of LPDDR5x memory, enabling inference on models up to 200 billion parameters and fine‑tuning of models up to 70 billion parameters at 4‑bit precision. The LPDDR5x memory reduces bandwidth compared with GDDR7, but provides the VRAM needed to avoid out‑of‑memory errors on local AI workloads. Comparable high‑capacity designs exist from Apple and AMD.
Read at Theregister
Unable to calculate read time
Collection
[
|
...
]