What makes the GB10 and Strix Halo interesting to me is that the integrated GPU has access to 128GB of unified RAM. This provides a very different runtime environment than a B580 GPU with 12GB of device memory.
I would like to see further comparisons between just the GB10 and Strix Halo for AI applications that need the larger RAM capacity. More investigation into the extent to which CPU and GPU compute can be combined would also be interesting.
Thanks Chester! Another great review! I appreciate that you put the GB10 into proper perspective, especially as its sometimes referred to as a "desktop supercomputer", which (IMHO) it both is and isn't. It does look like a great development platform for CUDA applications and tweaking AI distills.
More on the much lower end of AI accelerators, I would be interested in your take on the Pi AI HAT+ 2 NPU add-on for the Raspberry 5. Would be nice if they would send you one for review 😄.
I have a RTX 5070 that I'm not currently using. I could loan it to you for a couple months, if you're interested in comparing it to the GB10's iGPU (or doing any other sorts of tests with it).
What makes the GB10 and Strix Halo interesting to me is that the integrated GPU has access to 128GB of unified RAM. This provides a very different runtime environment than a B580 GPU with 12GB of device memory.
I would like to see further comparisons between just the GB10 and Strix Halo for AI applications that need the larger RAM capacity. More investigation into the extent to which CPU and GPU compute can be combined would also be interesting.
There are lots of long paragraphs.
It will be better if you put line break (html br tag) for each sentence to make it easier to be read, especially in mobile browser.
Using sans serif font also improves readability in screen.
Thanks Chester! Another great review! I appreciate that you put the GB10 into proper perspective, especially as its sometimes referred to as a "desktop supercomputer", which (IMHO) it both is and isn't. It does look like a great development platform for CUDA applications and tweaking AI distills.
More on the much lower end of AI accelerators, I would be interested in your take on the Pi AI HAT+ 2 NPU add-on for the Raspberry 5. Would be nice if they would send you one for review 😄.
I have a RTX 5070 that I'm not currently using. I could loan it to you for a couple months, if you're interested in comparing it to the GB10's iGPU (or doing any other sorts of tests with it).
Was testing for Strix Halo done on linux?
DGX spark runs Linux (ubuntu like variant)
any reviews on intel 18a panther lake laptops coming up and with it's Arc igpu.
I think what would be interesting are those Panther Lake SKU with Intel 3 process igpu chiplets.
Just to gauge how the High Density Cell libraries for intel 3 performs.