Intel delays the introduction of 3D XPoint memory modules

by Mark Tyson on 24 October 2016, 14:01

Tags: Intel (NASDAQ:INTC)

Quick Link: HEXUS.net/qadacy

Add to My Vault: x

Intel's highly anticipated next generation 3D XPoint memory was first revealed over a year ago. This 'revolutionary' new type of computer memory, a new class of memory sitting nicely between DRAM and NAND, started sampling early in 2016. Then we saw some impressive technology demos by Intel in April, and heard that Micron was making QuantX SSDs using 3D XPoint memory in August. Now it looks like we had better sit on our hands and wait another year or two before memory modules using 2D XPoint tech are leveraged alongside a future Intel Xeon processor.

'Future' not next generation

Finance orientated site Motley Fool reports that Intel has 'quietly' delayed the introduction of 3D XPoint memory modules. Intel's next server processors Skylake EP, which are due in H1 2017, won't support 3D XPoint memory modules in the Purley Server platform. Instead such support will debut in the second generation Purley servers, powered by Cannonlake-EP. In Intel CEO Brian Krzanich's words following the recent results statement "There will be a second generation of Purley that includes 3DXPoint."

According to what is known of Intel's roadmaps and release timings, the above outlined intention would mean that 3D XPoint memory modules won't debut until Cannonlake-EP does, which is estimated to be arriving anywhere in or around Q4 2018 to H1 2019.

Unfortunately Intel has previously talked-up 3D XPoint memory modules like a technology that is just about to hit the market. At a 2015 investor meeting after the initial introduction of the tech, Intel was claiming the total addressable market of 3D XPoint as a substitute for DRAM could be worth up to $34 billion by 2020. Disappointing your investors and slipping schedules are always bad news but Intel isn't in any financial difficulties right now, as the record quarterly revenue announced last week revealed.



HEXUS Forums :: 4 Comments

Login with Forum Account

Don't have an account? Register today!
FWIW, I believe that Intel would be kinder to its “fan club” if:

(a) it recognized that the DMI 3.0 link has the exact same upstream bandwidth
as a single M.2 NVMe port, and, therefore …
(b) it announced roadmap plans to expand that link to x8 or x16 PCIe 3.0 lanes;
© it announced its intentions to do (b) when it implements PCe 4.0's 16 GHz clock;
(d) it committed to future chipsets that support at least 4 integrated U.2 ports
with full support for all modern RAID modes;
(e) it demonstrated the leadership to build storage subsystems with variable transmission
speeds, perhaps with pre-sets like 6G, 8G, 12G and 16G, with full Plug-and-Play compatibility;
(f) it solicited customer opinions of proposals to populate 2.5“ SSDs and SO-DIMMs with Optane;
(g) if it can't build its own, then OEM an NVMe RAID controller like Highpoint's model 3840A;
(h) if Optane is not meeting expectations, Intel has enough cash to acquire better solutions
being developed by other manufacturers e.g. Everspin comes to mind (there are others);
(i) instead of ”locking in“ their chosen NVRAM e.g. with proprietary solutions,
a philosophy change would ”open up“ the Non-volatile ecosystem to exciting experimentation
by DIYs, prosumers and enthusiasts, who originated CPU and DRAM ”overclocking" years ago;
(j) recognized that the world has now entered an Era in which mass storage
is capable of operating at speeds comparable to DDR3-1333 and DDR3-1600 DRAM:
x16 lanes @ 8 GHz / 8.125 bits per byte = 15.75 GB per second!
x16 lanes @ 16GHz / 8.125 bits per byte = 31.50 GB per second!!

I could go on, but this latest news is a serious disappointment.
Wow… wow… hang on a minute! If sales are steady and profit is acceptable then there's ZERO reason to release anything new. /s
excalibur1814
Wow… wow… hang on a minute! If sales are steady and profit is acceptable then there's ZERO reason to release anything new. /s

And there, right there, is why the current quasi-monopoly sucks bad. You've really got to hope Zen is worth the hype, but it won't be big enough to more than gently wet the side of the oil-tanker that is Intel right now.
Charlie over at SemiAccurate has an interesting write up as to why XPoint gets less and less amazing (especially in Intel presentations) as time goes by.
http://semiaccurate.com/2016/09/12/intels-xpoint-pretty-much-broken/