Dell sent me two of their GB10 mini workstations to test:
In this blog post, I'll cover the base system, just one of the two nodes. Cluster testing is ongoing, and I'll cover things like AI model training and networking more in depth next year, likely with comparisons to the Framework Desktop cluster and Mac Studio cluster I've also been testing.
But many of the same caveats of the DGX Spark (namely, price to performance is not great if you just want to run LLMs on a small desktop) apply to Dell's GB10 box as well.
It costs a little more than the DGX Spark, but does solve a couple pain points people experienced on the DGX Spark:
It has a power LED (seriously, why does the DGX Spark not have one?!)
The included power supply is 280W instead of 240W for a little more headroom
The thermal design (front-to-back airflow) seems less restricted, so is quieter and capable of keeping the GB10 'AI Superchip' from thermal throttling
But if this isn't a mini PC to compete with a Mac mini, nor a good value for huge LLMs like a Mac Studio, or AMD's Ryzen AI Max+ 395 machines, what is it and who is it for?
Well, it's a $4,000+ box built specifically for developers in Nvidia's ecosystem, deploying code to Nvidia servers that cost half a million dollars each. A major part of the selling point are these built-in 200 gigabit QSFP ports, which would cost $1,500 or so to add on to another system, assuming you have the available PCIe bandwidth:
Those ports can't achieve 400 Gbps, but they do hit over 200 Gbps in the right conditions, configured for Infiniband / RDMA. And they hit over 100 Gbps for Ethernet (though only when running multiple TCP streams).
... continue reading