OPEN BOX - Asus Ascent GX10-GG0010BN - AI Desktop - ARM Cortex X925-128GB - 1TB PCI Express NVMe 4.0 x4 SSD - Stellar Gray - NVIDIA GB10 Grace Blackwell Chip - DGX OS - NVIDIA Blackwell

First, be aware the seller charges a 10% restocking fee. $300 plus $160 for insurance from UPS to send it back was a lot of money to find out that the DGX Spark does not live up to the hype for my use case. Overall, the hardware is good. I didn’t hear a whisper of the fan in 3 days of testing either. I have read that the Nvidia version is loud. I would recommend the Asus Ascent for its design. My problem is with the software. DGX OS is easy to set up, but it is locked into CUDA 13. I first tried to run the newest Nemotron Nano using NIM, but it has a requirement for CUDA 13.1. After upgrading, I kept getting so many errors indicating a mismatch between CUDA and the OS that I reinstalled the OS, which took an hour. Next I tried GPT-OSS:120B running under Ollama, and while it worked, I got roughly 10-15 tokens per second. NIM did work, but it was not any faster. So this supercomputer which is supposed to be able to handle 200 billion parameter models is very slow if you can get it to work. Depending on your usage, if you just want to do inference, any cloud service will take years to equal the cost of the DGX Spark. My RTX 5090 runs Nano and GPT-OSS:20B at 100-200 tokens/s, and it was the same cost. It has the added benefit of running Cyberpunk 2077 at the highest settings at 100+ fps too. This review is solely on the models mentioned. YMMV for other models or use cases. If you buy a Spark, the Asus Ascent is a good choice.

Pretty good if you need to run basic inference. And very light training jobs. Thermal limits, I think, prevents this unit for any moderate training workloads. For example a 7B mistral and 8B llama fine tuning failed with moderate data and moderate network size. Plenty of memory to spare, but the unit powers off suddenly. Until these issues are fixed I don’t recommend for LLM fine tuning.

Product is solid. I quite enjoy the compactness and the Blackwell GPU. Asus is offering this at a discounted price from DGX spark. I find Linux set up no issues. The only downside is Bluetooth connection to earphone and speakers -- I have not yet figured that out. A note to seller and future buyers: please ask the shipper to ensure the product. For me, FedEx has delivered this package to my neighbor. As a result, I wasn't able to find it for a day or two on premise. No signature was required by FedEx , no photo proof of delivery was found. Had it not for the altruism of my neighbor ,I would have completely lost this expensive package.

Excellent product. Great for everyone that wants to get a grasp of the latest technology regarding computing and data processing. Even more special was the customer attention provided by ExcaliberPC.

I have 2 of these in a cluster for ai workloads and its the best bang for buck system that supports the latest in nvidias software ecosystem. My NAS is louder than this thing. Whisper quiet. Sips power. Looks great.