Hawker News

New CRAM method offers giant boost to compressed memory reads

tomshardware.com

59 pointsby danny0025 comments

https://lpc.events/event/20/contributions/2424/

https://www.phoronix.com/news/Linux-CRAM-Compressed-RAM

Kevin_Flynn[hidden]
Ram Doubler, for Mac and Win 3.x, 1996

https://winworldpc.com/product/connectix-ram-double/windows-...

Nice to see its back, with better performance.

History repeated, with refinement.

eglintondust[4 comments hidden]
So can I finally download more RAM?
smallmancontrov[hidden]
https://downloadmoreram.com/

Whoa, new options! I'm downloading the Quantum RAM with Haptic Feedback.

swed420[hidden]
You wouldn't download a car
skavi[hidden]
probably not to your current computer. this works on supporting memory types that do compression in hardware.
cyberclimb[3 comments hidden]
the recording of the presentation has got to be on this YT channel but I only scrubbed through 2 videos (8h each) and it's not easy to find

https://www.youtube.com/@LinuxPlumbersConference

not sure where it falls in the schedule here either https://lpc.events/event/20/timetable/#all

tux3[2 comments hidden]
Unfortunately the recording is missing audio for half the talk. A frequent issue for the plumbers livestreams this year.

https://www.youtube.com/live/OPRciCSsdS4?t=2031

cyberclimb[hidden]
Legend! Thanks for digging it up
vfosnar[3 comments hidden]
https://www.phoronix.com/news/Linux-CRAM-Compressed-RAM
x0z[hidden]
Thank you <3
dang[hidden]
Added above. Thanks!
cwillu[4 comments hidden]
The article appears to misunderstand the point of the project, and mostly just describes what zswap already does (writing compressed pages back into ram), rather than talking about cram's hardware-offloaded compression that allows cacheline-level access rather than page-level. (And they appear to be aware that they don't really understand it: “My explanation of CRAM might not be completely correct”)

The phoronix article is better, and I say that as someone who usually detests the quality of phoronix's technical writing.

tancop[2 comments hidden]
It's not really about hardware offload, the biggest problem with zswap is that it's swap. The rest of the kernel treats it like a fast SSD (which is still incredibly slow compared to RAM) instead of slower memory that needs a bit of special handling on writes.

NUMA maps a lot closer to what compressed RAM actually is. The subsystem is more aware of CRAMs specifics so it can make better decisions about where to put allocations and everything gets faster. And it's less overhead because swap is not really optimized for frequent direct access but for NUMA it's the most basic function.

ahartmetz[hidden]
Whoa, building it on top of the NUMA logic is pretty clever - also, finally all that complicated code does something useful on normal computers!
ragall[hidden]
The article was written by an LLM. No understanding was involved, no brain cells harmed in the process.
madduci[9 comments hidden]
Finally we can run frontier models locally
throawayonthe[2 comments hidden]
one would hope model weights are already high entropy enough and would not be improved by a general-purpose memory compression algo?
ranger_danger[hidden]
It turns out that there is still lots of room for improvement: https://huggingface.co/prism-ml/Ternary-Bonsai-2-27B-gguf
sroussey[3 comments hidden]
We brute force AI models right now because a) we don’t know better, and b) it’s premature optimization.

I beg to differ on point b, but no one is delaying their next model just so they can concentrate on optimization.

It’s coming though.

One example: https://siliconangle.com/2026/07/28/ai-model-compression-sta...

Another is separating the the intelligence part of the model from the known facts part of the model (which can be better compressed)

djmips[2 comments hidden]
The Chinese breakthroughs are often around optimization. And I gather it's because they are constrained and very bright so it's a natural progression.
ranger_danger[hidden]
IMO Restrictions are exactly what sparks creativity in people... whereas too many options leads to choice paralysis, which is what I fear is a big problem in the FOSS world.

Don't get me wrong... I think having choices is still good, I just think that having too many choices is not equally as good, or necessarily better.

tintor[3 comments hidden]
Compression doesn't really work for model weights.

Model quantization and model distillation are two techniques to reduce model size.

madduci[hidden]
It was sarcastic
variadix[hidden]
https://blog.cloudflare.com/unweight-tensor-compression/
yonatan8070[hidden]
Is the hardware support this needs common? Can it be used on regular x86-64 desktops/laptops or is it for specialized hardware?