Framework's first desktop is a strange–but unique–mini ITX gaming PC

Re:Framework's first desktop is a strange–but unique–mini IT... by abdullahkhalids on Tuesday February 25, 2025 @05:54PM (#43178617)

I don't know anything about RAMs or their bus size. Is this something that will be "fixed" in the future, idk, with DDR6? Meaning we can have replacable RAM with such bus.

[ Parent | on HN ]

Re:Framework's first desktop is a strange–but unique–mini IT... by pests on Tuesday February 25, 2025 @08:03PM (#43179484)

This is getting down to speed of light and the distance to the ram. The RAM needs to be very close to the chip.

This is also being seen in the X3D CPU's where the memory is stacked on top to reduce latency and as density was maxed in 2d space.

[ Parent | on HN ]

Re:Framework's first desktop is a strange–but unique–mini IT... by SSLy on Tuesday February 25, 2025 @06:29PM (#43178864)

There isn't any standard to put that kind of RAM on DIMM-like slots.

[ Parent | on HN ]

Re:Framework's first desktop is a strange–but unique–mini IT... by TheAmazingRace on Tuesday February 25, 2025 @07:46PM (#43179365)

Honestly... I wonder why they didn't consider CAMM2? The bandwidth and signal integrity rivals that of soldiered memory. They could have had the best of both worlds, really.

[ Parent | on HN ]

Re:Framework's first desktop is a strange–but unique–mini IT... by rangestransform on Wednesday February 26, 2025 @12:08AM (#43180832)

AFAIK you would need 4 CAMM modules for that kind of bandwidth, good luck stuffing that onto a board at all, let alone for a reasonable price

[ Parent | on HN ]

Re:Framework's first desktop is a strange–but unique–mini IT... by anticensor on Wednesday February 26, 2025 @12:45AM (#43181012)

CAMM, unlike DIMM, is designed to be amenable to stacking from the get go.

[ Parent | on HN ]

Re:Framework's first desktop is a strange–but unique–mini IT... by chipsa on Thursday February 27, 2025 @08:00AM (#43193990)

A single LPCAMM2 module is supposed to be all the RAM for a laptop, so it’s 128 bit wide. You’d need two LPCAMM for this, not four. And they are currently available in 32G and 64G varieties ($180/$330). So they aren’t absurdly priced, if they were in stock.

[ Parent | on HN ]

Re:Framework's first desktop is a strange–but unique–mini IT... by rangestransform on Thursday February 27, 2025 @10:27AM (#43195214)

For dual channel 256-bit wide LPDDR5X, you would definitely need 4 CAMM modules

It’s also not that absurd pricing of the modules themselves but of the R&D effort to cram 4 CAMMs onto a board with sufficient signal integrity

[ Parent | on HN ]

Re:Framework's first desktop is a strange–but unique–mini IT... by chipsa on Friday February 28, 2025 @09:04AM (#43205692)

“Dual channel” is not a thing with CAMM. CAMM is 128 bits wide. To feed a 256bit wide memory controller, you need 2 CAMM modules. LPCAMM2 uses 4 memory packages. The observed number of packages next to the APU on this is 8. So you need 2 modules.

[ Parent | on HN ]

Re:Framework's first desktop is a strange–but unique–mini IT... by neRok on Tuesday February 25, 2025 @11:03PM (#43180475)

Apparently not. In Linus Tech Tips latest video, the Framework CEO says they talked to AMD about it and that one of their engineers ran some simulations and found the stability would degrade too far for it to be possible.

[ Parent | on HN ]

Re:Framework's first desktop is a strange–but unique–mini IT... by sliken on Tuesday February 25, 2025 @11:04PM (#43180481)

Problem is there is no socket available for thin/light/low power CPUs to have a 256 bit wide bus.

So to ensure 256 bits @ 8000 MHz works well AMD did it in the same package. In theory they could A) ship a halo without ram B) design a new socket, C) allow motherboard makers to pick 2 x CUDIMMs or 4x DDR5 dimms. Not sure that would buy them much market though.

As apple and others have proved, not many care about replaceable dimms, especially if it gives them 2-3x the performance and/or better perf/watt.

[ Parent | on HN ]