Micron's memory cube supposedly in 2013

[Infinisearch]

Not sure if this is the right forum for this, so feel free to move.

Micron supposedly has a memory cube coming in 2013. 320GB/s bandwidth.

http://www.extremetech.com/computin...t-in-2013-threatens-to-finally-kill-ddr-sdram

To bad none of the new consoles will have it.

 

[almighty]

8Gb at 320Gb/s... Mmmmm..

 

[Infinisearch]

And they're comparing it to DDR3 so i think that bandwidth is for a single 64bit channel. It also says 70% less power consumption.

 

[XpiderMX]

Anyone know why MS quit the hmc consortium?

 

[SlickShoesRUCrazy]

Anyone know why MS quit the hmc consortium?

probably won't be ready in time for the Durango to be rtm'd

 

[MrFox]

They joined mid-2012, that's very late for anything related to the xbox (the SoC needs to be designed to use this technology). Also I can't believe they could completely redesigned Durango to use something else just 6 months later when they left either. My guess is that it was considered only for tablet and stuff, not Durango.

The weird thing is that most companies that could use this for GPU cards and console SoC, are completely missing from the list of partners. No Intel, no AMD, no Nvidia. I'm hoping it's because they all decided to go the HBM route in the future.

 

[dobwal]

They joined mid-2012, that's very late for anything related to the xbox (the SoC needs to be designed to use this technology). Also I can't believe they could completely redesigned Durango to use something else just 6 months later when they left either. My guess is that it was considered only for tablet and stuff, not Durango.

The weird thing is that most companies that could use this for GPU cards and console SoC, are completely missing from the list of partners. No Intel, no AMD, no Nvidia. I'm hoping it's because they all decided to go the HBM route in the future.

Well AMD had been sponsoring the research of stack memory on their own a few years back. What I find amusing is that the researcher used a Nvidia gpu paired with a stack memory configuration of 4GB of GDDR5 (600 GBs).

The researcher even has a paper of a hybrid memory system for a gpu with varying small amounts of NVRAM (1MB...1GB) interfaced with VRAM and both are interfaced with a memory controller. Look similar to Durango's setup but the NVRAM is off chip and could only read from the VRAM at 10 GBs.

 

[MfA]

Meh, colour me not impressed. I was hoping for order of magnitude gains with DRAM connected to logic through wide-IO, either through TSV or silicon interposer, this is incremental ... the interface is relatively narrow, connected through bond wires and designed for normal substrates and PCBs.

 

[MrFox]

Micron opened up the Hybrid Memory Cube to other companies, though initial implementations are aimed at the server market. Future generations will target mobile applications, but no one knows when those will begin showing up.

That explains why the companies on-board are only server and networking equipment makers (must be important for high-end routers, I see my favorite company in the list), and there are also cellphone SoC manufacturers, so portable is a future target, or at least it got these companies interested, that doesn't mean they have anything planned though.

For GPU and Consoles, an AMD presentation have shown slides about using HBM in the future, which makes sense to me because it's a JEDEC standard. I'm partial to it, because of painful memories of Rambus versus JEDEC.