Plus, you can now deploy [MLX projects on CUDA](https://github.com/ml-explore/mlx/pull/1983)
No processor can come close to performance per watt, and performance per watt wins the laptop market. Laptop market dominates PC sales
The grass is not the same on the other side
So it's not full ECC like servers have with dimms with a multiple of 9 chips with ECC protecting everything from the dimms to the CPU.
Keep in mind the ram is inside the strix halo package, not something HP has control over.
It's not in the package, it's on the motherboard spread around the SoC package: https://www.hp.com/content/dam/sites/worldwide/personal-comp...
The 8 DRAM packages pretty clearly indicate you're not getting the extra capacity for end-to-end ECC as you would on a typical workstation or server memory module.
I now ordered a Beelink GTR9 Pro, which unlike the Framework has dual 10G Ethernet, not the weird 5G flavor. We'll see how that goes.
And thus their machines are expensive toys as well.
(I type this on a system I built, a 9800x3d with 96GB (2x 48GB) ECC RAM.)
>If your only aim is to run LLMs how much do you actually care about ECC?
That is not me, nor do I know anyone who exclusively uses computers to run LLMs.