This is a great product - I have tested it for more than 20 hours continuously training a model and - unlike the founder’s Edition - this doesn’t thermal throttle Besides the 1 tb storage the other difference one will notice is the nvidia software updates are a bit delayed as they are offered through the OEM partners - in this case ASUS TPM - comes disabled - needs to be manually enabled For inference, don’t expect high throughput - but you will be able to run models that a normal GPU would not be able to - slightly slower. For extra storage - add external usb storage. Overall amazing product

I bought one of these to run local LLMs for my OpenClaw adventures. I was having so much fun that I bought a second one and set up a cluster. That process was very straightforward, and the hardware has been great. The cluster has been up 20+ days so far without a hiccup. I can’t promise I won’t buy another one (or two) of these.

Same basic form factor as all the DGX Spark variants, but the cooling setup is different from the majority. They draw cool air over the NVME directly from the bottom of the case. This keeps the drive cooler, but they should be placed on the bottom row of a cluster to keep them from sucking in the radiant heat from other systems. They draw slightly less power at idle than my other two systems, but do reach slightly higher temps under load. Nothing crazy, just a few degrees. The one negative I have with these is software based... neither of the GX10s I own will fully update through the DGX Dashboard. There is some misconfiguration that prevents updates from being hidden if not needed. I did some digging and found out it was trying to install an older version and the system was refusing. None of my other GB10 systems do this. It can be safely ignored until they fix it, but it's still annoying... and I never know if there are real updates to install. I bought all my systems before the 64gb versions were announced... so I paid $4k or less. These are capable systems, especially when clustered... but the current prices make it less of a value than they were even 3 months ago. If you can justify the expense... and have a workload to make use of them... it's still worth it.

Fantastic! Does exactly what I need. I found it works best if you do not try to work directly on the device but have it host an AI api. then you can have a workstation with all your tools reach out to it without bogging down the system with a load of software that interferes with its AI performance. I look forward to getting a couple more and building a cluster. So far I have not encountered anything negative on this device. it has simply done everything I asked it to.

Steady local AI running 24/7. Set up was a bit of a problem because a driver was corrupted but after the system was reset it was smooth