News for hackers. Stuff that matters.
Re:I'm returning my Framework 16 by wqaatwt on Thursday December 25, 2025 @05:54AM (#46383639)
> soldered-on RAM modules

That can have significant performance advantages, though. Which might be hard to overcome due to physics

Re:I'm returning my Framework 16 by swiftcoder on Thursday December 25, 2025 @09:08AM (#46384489)
Worth noting that even Framework's own desktop has soldered-on RAM, for exactly this reason
Re:I'm returning my Framework 16 by jdujfkhllhl on Thursday December 25, 2025 @04:49PM (#46387265)
And that was even after constructive cooperation with amd. Now think about the more common, hostile, interaction many suppliers provide.
Re:I'm returning my Framework 16 by dngray on Friday December 26, 2025 @07:16AM (#46391358)
It's also worth noting the CAMM2 ram gets about the same performance.
Re:I'm returning my Framework 16 by cosmic_cheese on Friday December 26, 2025 @08:56AM (#46392012)
Framework has stated that it asked AMD if there were any way to make the RAM on Ryzen AI Max APUs (like used in the Framework Desktop) socketed, and AMD said no due to the stability hit that’d entail — the physical distance from the CPU that’d be required with RAM sockets reduces signal integrity too much for it to function.
Re:I'm returning my Framework 16 by Dylan16807 on Saturday December 27, 2025 @03:10PM (#46404805)
Which is weird. The entire point of [LP]CAMM[2] is to be able to make that work.

The Framework desktop clocks the memory at 8000MHz. That's well within the limits of the interface. Something is flawed or omitted in those CPUs if they can't handle it.

Re:I'm returning my Framework 16 by goku12 on Friday December 26, 2025 @08:47AM (#46391930)
If you're gaining advantage by changing RAM from sockets to soldered joints, it's probably time to change the system design altogether. It's better to put the DRAM on the same IC/SoC as the processor - on a dedicated die if necessary. Any additional memory requirement can be added as socket based RAM modules. They sure will be slower. But they can be treated as another memory layer, kind of like the optane memory (without persistence) or NUMA. You'd still get significant speed up because a portion of the DRAM is colocated with the CPU now.

This also adds to the core philosophy that I'm trying to push. Modularity and serviceability doesn't necessarily mean sacrificing performance, compactness or security. That's a myth that's too prevalent in the industry.

Re:I'm returning my Framework 16 by Dylan16807 on Friday December 26, 2025 @09:33AM (#46392306)
There is a new system design, it's called LPCAMM. And framework would have used it in the desktop but those CPUs have some flaw that make them not compatible with full speed LPCAMM.

Moving the memory even closer doesn't have all that much advantage. And having super close RAM and sockets is a waste of die space on all those I/O channels. One or other can fit all the needs of any particular CPU.