News SanDisk and SK hynix join forces to standardize High Bandwidth Flash memory, a NAND-based alternative to HBM for AI GPUs — Move could enable 8-16x...

[Admin]

SanDisk and SK hynix have announced they're collaborating on High Bandwidth Flash (HBF)—a potential alternative to traditional, DRAM-based HBM for AI GPUs. Instead of only relying on DRAM, HBF adds NAND to the memory stack, allowing for non-volatility that can cut energy (and thermal) costs, along with upping the storage capacity.

SanDisk and SK hynix join forces to standardize High Bandwidth Flash memory, a NAND-based alternative to HBM for AI GPUs — Move could enable 8-16x... : Read more

 

[we_are_theBorg]

Oh great.... I'm sure all the big AI hardware vendors are salivating at the idea of inserting a NAND write endurance based death clock in to all of their new products... good luck picking these up on the secondary market,

Fantastic.

Ironically this is where something like Optane would be perfect. Higher density than DRAM, lower latency than NAND, massive write endurance. Cost premium insensitive. Hello Intel, get your butt off the floor and figure out the licensing!

 

[deesider]

Oh great.... I'm sure all the big AI hardware vendors are salivating at the idea of inserting a NAND write endurance based death clock in to all of their new products... good luck picking these up on the secondary market,

Fantastic.

Ironically this is where something like Optane would be perfect. Higher density than DRAM, lower latency than NAND, massive write endurance. Cost premium insensitive. Hello Intel, get your butt off the floor and figure out the licensing!

I think it's all too easy to view Optane through rose-coloured glasses, and ignore the issues that prevented it from becoming successful - namely high power usage, and lack of scalability rather than miss-management.

- The high power usage prevented it from being used in most consumer devices, which are largely power-sensitive laptops and phones.

- The lack of scalability seems to be due to either fabrication or design limitations that prevent it from being layered like conventional NAND. Being able to fabricate dozens of layers of NAND at once was a major breakthrough that finally provided relatively cheap SSDs. It seems that wasn't possible with Optane, either due to its design (which may also be why it operates so much faster) or it could be impractical to have that many layers of a temperature sensitive material crammed together.

A good comparison would be plasma TVs. They were still the benchmark for picture quality for at least 5 years after production was ceased, but ultimately, despite the advantages, were a dead-end technology - more expensive to manufacture than LCDs, and the cells couldn't be made small enough for 4k resolution - which apparently was what the market demanded.

 

[KryptonQuark]

Hardware Vendors are surprised ! ! !