New 'Gen-Z Consortium' Announces Interconnect Technology Optimized For Next-Generation 'Storage Class Memory'

Status
Not open for further replies.

jasonf2

Reputable
Oct 11, 2015
126
24
4,695
2
I applaud this development. Intel with it's xpoint and chip-set marriage was setting the stage for a high margin nightmare. Remember RIMMs. This will set the stage for a universal standard that will allow for reasonably low cost access to nv memory. This will probably be the death blow for PCI though.
 

Xajel

Distinguished
Oct 22, 2006
135
1
18,685
0
So They're creating a standardized version of NVLink, I think Intel might enter later... and maybe NV later, much later...
 

Xajel

Distinguished
Oct 22, 2006
135
1
18,685
0
So They're working on a standardized competitor to NVLink, expect Intel to join later... and NVIDIA might join much, much later...
 

josejones

Distinguished
Oct 27, 2010
901
0
18,990
2
"Intel doesn't appear to be part of the consortium as of right now"

Will Intel join at some point or is that a definite no? We need to know. And when will this be available on motherboards and which ones? I hope this doesn't take 10 years to be released.
 

memadmax

Distinguished
Mar 25, 2011
2,492
0
19,960
95
Intel and microsoft have teamed up to corral us into a closed system.
Microsoft says Win10 will only support kaby lake...
That means intel has to be the one that puts in the hardware to close it...

Total BS.

Gen-Z is the savior of the PC industry.
 

bit_user

Splendid
Ambassador
CAPI, perhaps. But I don't see PCIe going away, anytime soon. I expect to see v4.0 devices, next year. CAPI is only about 50% faster than that, from what they've said. I actually like PCIe, except that v4.0 seems to have gotten slow-walked.

OMG. You have it completely backwards. They said they'd only add Kabylake support to Win 10. Win 10 supports lots of older CPUs, and even ARM-based Raspberry Pi!

As for Gen-Z, this is aimed mostly at the datacenter. Perhaps it'll trickle down to the PC, but that's not where they're aiming.

If you read the article, it talks about rack-level aggregation of storage-class memory. What they're after is the ability to pool nonvolatile memory and have various chips in all the boxes within the rack gain access to it as if it were local.

And even though they say the software can treat it naively, I think 1 microsecond is a long time to wait for a cache miss. I could see the OS swapping to non-nonvolatile memory, or snapshotting application/VM state, for instant recovery in the event of hardware failure.
 
Status
Not open for further replies.

ASK THE COMMUNITY