KAIST startup Panmnesia, named after an ability one can only dream of having in exam week, has unfurled a fab new tech trick to massively expand GPU memory. Hitching their breakthrough to the cost-savvy NAND-based SSDs, they’ve managed to push memory into the terabyte territory without trading off much in the way of performance. The twist in the tale? It’s all hitched to the still-green Compute Express Link (CXL) standard, a tech prom date that’s hard to predict – will it be a fit or a flop?
Technical babble time – CXL is not your everyday tech chatter. This open standard interconnect plays nice by allowing CPUs, GPUs, and other siblings to share memory in a manner that reminds one of happy, cooperative toddlers. Panmnesia claims their CXL-GPU prototype can dance circles around traditional methods with speed that triples anything we’ve seen before. To make it happen, they’ve got this whole setup connected with high-speed MCIO cables, proving once again that in tech, as in fashion, accessorizing is key.
Yet, integrating this wonder into everyday gadgets isn’t like slipping on Cinderella’s slipper; it’s more like threading a needle while riding a rollercoaster. Current GPUs will need new slots and a bunch of rework to play nice with CXL. It’s a resource-gobbling change that may deter some, but if big-scale AI brains are your game, the memory boost offered by Panmnesia’s tech could well be the fairy godmother your GPU has been waiting for.
**Hot Take**
Let’s face it – in a world where forgetting your wedding anniversary is less forgivable than misplacing your kids at the mall, Panmnesia is on to something spectacular with its memory boost bonanza. If GPUs had a social network, they’d be posting selfies with their new terabyte memories, humblebragging left and right. Just let’s hope this CXL thing sticks around longer than most internet challenges.
Original Article: https://www.techradar.com/pro/new-tech-promises-to-add-terabytes-of-memory-or-ssd-storage-to-any-gpu-but-theres-a-big-catch