Today, we are allowed to share only the unboxing of the Dell Pro Max with GB10. I know folks will want to see the unit powered on, performance, features, quirks, and more. For today, NVIDIA told Dell, who told us, we can only show an unboxing. Still, this will be one of the hottest tiny AI systems of 2025.
Dell Pro Max with GB10 Unboxing a NVIDIA GB10
Here is a quick unboxing video that Sam and I made for this:
In the unboxing video, we go into the system and its external connectivity, which you can look up on the Dell or NVIDIA websites. Here are the key specs from the Dell website:
- NVIDIA GB10 Grace Blackwell Superchip
- 128GB LPDDR5x Unified memory
- Supports up to 200B parameter models
- One Petaflop (1000 TFLOPS) of FP4 computing power
- ConnectX-7 SmartNIC
- NVIDIA DGX OS
That blog post also shows the internals:
There is a lot going on in here. What really sets this apart is the 200Gb NVIDIA ConnectX-7. I know many of our readers may wonder if this would connect to the MikroTik CRS812 DDQ 400GbE Switch Launched CRS812-8DS-2DQ-2DDQ and might be why we have featured that switch on the STH main site. We cannot comment on that at this time.
Final Words
I think most folks know I do not like unboxing videos. If I were tuning into STH, I would want to see this turned on and running. We hope to have a lot of GB10 content, so stay tuned.
What I can say for now, and I have said since GTC 2025 earlier this year, the GB10 is going to be one of the hottest AI products this year. If you are a Dell AI shop, this is what you want. Also, if you are a developer or an executive and want to take an awesome AI development workstation with you on the road, you probably want to get a reservation in on these sooner rather than later.
Any chance this dell flavor comes with idrac or some kind of BMC?
As tested by SGLang running on Ollama, not faster than M1 Max.
I can’t wait to see these benchmarked against the max 395 systems. While the compute performance will have a large gap the model size will be about the same.
Also it seems the comment system is not capable of using email addresses outside of the major providers.
These are for developers for nvidia’s ecosystem. The customers are completely different from max 395 systems.
Did you have the change to compare noise and thermal performance compared to the “founder” DGX Spark? It’s still a good option or the founders is better?
@Chad – Unfortunately not. I pushed Dell on it, but it’s plain to see that it’s Nvidia’s reference board, so no BMC.
Is this only for AI/KI development or are there other use cases on the horizon, like e.g.
– low DPC latency for audio-related tasks/near real-time processing of audio,
– video processing/rendering, photo processing with KI functions,
– gaming?
How about
– noise and energy consumption when working with office applications,
– DPC latencies,
– energy consumption in such different use cases?
Are we ever going to get a review on this thing?
Hey Rano – Dell just sent two production units so we are splicing that in today and Friday. The article/ video will go live the first half of December once this is all in.