News Intel's Annual Report Indicates Second-Gen 3D XPoint Might be Delayed Until 2021

[Admin]

Among other information, Intel's annual report indicates that Intel might not be expecting a broad launch of its second-generation 3D XPoint products, consisting of the Alder Stream SSD and Barlow Pass DIMM, until 2021.

Intel's Annual Report Indicates Second-Gen 3D XPoint Might be Delayed Until 2021 : Read more

 

[InvalidError]

It wouldn't be modern-day Intel without things getting delayed by a couple of years.

 

[bit_user]

It wouldn't be modern-day Intel without things getting delayed by a couple of years.

Well, their 14 nm products seem to launch on time. In that case, the delays don't come in to the picture until you actually try to order them.

Let's see if their dGPUs launch on schedule.

 

[InvalidError]

Well, their 14 nm products seem to launch on time.

Broadwell launched fashionably late and everything after Kaby Lake only exists because Intel needed something on the market while 10nm was "coming soon" like a horde of creeper zombies.

 

[bit_user]

Intel explained why the 25% increase in 14nm wafer capacity merely resulted in a low double-digit (10-15%) increase in PC supply. Intel attributed this to higher model volume, larger die sizes and more14nm chipsets (which continued to move from 22nm)

That's what I've been saying. They had to add cores to keep up with AMD, which in turn hurts their supply crunch even more.

 

[bit_user]

Broadwell launched fashionably late

Yes, I didn't think you were going back that far, because all of their Skylake-branded products were basically on-time.

and everything after Kaby Lake only exists because Intel needed something on the market while 10nm was "coming soon" like a horde of creeper zombies.

Be that as it may, they've been executing to their roadmaps, on launches of 14 nm products.

The delays are all coming from 10 nm.

 

[InvalidError]

Be that as it may, they've been executing to their roadmaps, on launches of 14 nm products.

14nm products that were not meant to exist and are fundamentally the same Skylake as four years ago with some hardware fixes and more mature fab process. It would be highly problematic if Intel ran into any problems executing that.

 

[bit_user]

14nm products that were not meant to exist

Yeah, I get it. They represent a strategic fallback position that wouldn't have been needed if Intel could've executed on their 10 nm plans.

However, you seem to be spinning a story where Intel literally can't execute on anything, whereas the reality is really just that specifically their 10 nm has been a disaster.

It would be highly problematic if Intel ran into any problems executing that.

Intel has been rolling out new low-power cores, and their Gen 11 UHD iGPUs launched on schedule. The place hasn't fallen utterly to pieces. Not yet, at least.

 

[InvalidError]

However, you seem to be spinning a story where Intel literally can't execute on anything, whereas the reality is really just that specifically their 10 nm has been a disaster.

Successfully executing steppings/re-spins of Skylake for four "generations" isn't much of a technical or engineering feat. That's basement level expectations compared to how Intel used to successfully execute simultaneous node and architecture upgrades. It hasn't done either in a particularly meaningful way in four years. Yes, it added extra cores in the last two (soon to be three) generations, but that is little more than a copy-paste exercise in a ring-bus architecture.

Intel has been rolling out new low-power cores, and their Gen 11 UHD iGPUs launched on schedule.

Ice Lake isn't on schedule, the first time Intel put Ice Lake on a road map, it was supposed to launch two years ago.

 

[bit_user]

Ice Lake isn't on schedule, the first time Intel put Ice Lake on a road map, it was supposed to launch two years ago.

If you want to undermine my point, try citing something that slipped which isn't made on their 10 nm process.

 

[spongiemaster]

Successfully executing steppings/re-spins of Skylake for four "generations" isn't much of a technical or engineering feat. That's basement level expectations compared to how Intel used to successfully execute simultaneous node and architecture upgrades. It hasn't done either in a particularly meaningful way in four years. Yes, it added extra cores in the last two (soon to be three) generations, but that is little more than a copy-paste exercise in a ring-bus architecture.

The architecture hasn't changed much, but the node optimizations weren't trivial. The Intel engineers have really accomplished something getting 8 core mainstream (ie high volume) CPU's up to 5GHz on 14nm. Rumors are that they'll make it to 10 cores at 5GHz before they get to 10nm.

Ice Lake isn't on schedule, the first time Intel put Ice Lake on a road map, it was supposed to launch two years ago.

I think you're confusing Cannon Lake with Ice Lake. Cannon Lake had one token CPU sku released about 18 months ago so Intel could claim they were shipping 10nm. Ice Lake was not on roadmaps 3 or 4 years ago.

 

[InvalidError]

I think you're confusing Cannon Lake with Ice Lake. Cannon Lake had one token CPU sku released about 18 months ago so Intel could claim they were shipping 10nm. Ice Lake was not on roadmaps 3 or 4 years ago.

Before Intel scrapped its PAO strategy, the roadmap was skylake(arch)->kaby(optimization)->cannon(shrink)->ice(arch).

 

[Giroro]

"we are developing products to disrupt the memory and storage hierarchy. "

Does developing "disruptive" products matter if Intel doesn't actually release them, though?
Have they updated the 905p with a modern controller with competitive sequential reads/writes to compliment their advantage in random? no. (And now they probably can't catch up with their current gen memory since they missed the boat on PCIe 4.0)
Have they updated the Optane add-in modules so their tiered storage is better positioned against current high-end SSDs? no.
Have they built an SSD that replaces the SLC cache with Xpoint? no. -except that weird pointless thing that used 2 controllers in tiered storage and was functionally the same as pairing one of those add-in cards with an SSD.
Disruptive pricing? No. But, nobody would ever expect that out of Intel.

There's a lot of disruptive things Intel could have done with Optane, but so far they are letting the consumer facing branding and product stack die on the vine in lieu of slowly grinding the server market with monopolistic Xeon-style pricing on what seems to be low-volume sales to a limited number of customers. That's what this delay reeks of. The quintessential Intel attitude of "Nobody else has made something better, so why should we?"

If they had bothered to keep their product stack up to date, they could have been using Optane as a reason somebody should buy Intel over AMD, but now high-end flash based SSDs are "good enough" for the vast majority of their potential customers for less cost. Their potential market for Optane has got to be shrinking.... Then again I've heard Intel was selling all the Optane that they can manufacture, Who knows if that is a large volume or not?

It's frustrating to watch what should have been a revolutionary technology utterly squandered.... or perhaps 3D XPoint was never as good as we were led to believe. I don't know, I just thought we would have better SSDs by now.

 

[InvalidError]

"we are developing products to disrupt the memory and storage hierarchy. "

Does developing "disruptive" products matter if Intel doesn't actually release them, though?

One of the major problems with "disruptive technology" is that a lot of it fails to become economically viable or practical for manufacturing at the time of discovery/invention and resurfaces 10-20 years later when the costs and manufacturing challenge have been overcome. Some of it gets locked behind onerous patents and mostly forgotten until applicable patents expire.

 

[Giroro]

One of the major problems with "disruptive technology" is that a lot of it fails to become economically viable or practical for manufacturing at the time of discovery/invention and resurfaces 10-20 years later when the costs and manufacturing challenge have been overcome. Some of it gets locked behind onerous patents and mostly forgotten until applicable patents expire.

If it's locked down, ignored, or not economically viable, then that' the opposite of disruptive though, isn't it?

I think that in order to be considered disruptive, the technology needs to... disrupt something, or some established market.
The problem with Intel calling themselves disrupted is that they are the big old established power that is there for other companies to try to disrupt.
Cable companies were disrupted by Netflix, publishers were disrupted by online distribution, Big Tobacco was disrupted by vaping. When something truly disruptive comes along, you will know when the big guys either buy it or try to ban it - usually both.
Intel may not own the storage market- and they have a technology that could revolutionize it, but you can see the trickle down effect of their leaders' big-business mentality. They are accustomed to being on top and therefore fear revolution. So the empty and vaguely hypocritical marketing rubs me the wrong way.

But if Intel really does want to be disruptive, they have more reach and resources than almost anybody in the world. So just do it Intel. What are you waiting for?

 

[bit_user]

if Intel really does want to be disruptive, they have more reach and resources than almost anybody in the world

They have a lot of resources, but they also have shareholders who don't want to see them sink too much capital into risky moonshot projects. The part you missed about disruptors is that a lot of them fail, utterly. Venture Capitalists know this, and adjust their investment portfolios to account for it. Intel's investors are vastly more conservative, and definitely don't expect Intel to be making bet-the-company gambles on a single technology.

If anything, they probably should've waited one more generation before splitting with Micron.

So just do it Intel. What are you waiting for?

You put your finger on it, though. They want to use Optane to give their expen$ive server CPUs an advantage, so they're splitting focus between that and conventional applications (i.e. NVMe drives). Not exactly the approach of a real disruptor.

In addition to that, I get the impression they encountered some manufacturing-related delays.

 

[JayNor]

Intel had problems solving the 10nm multi-patterning, but persevered.

In parallel they had a group working on 7nm with EUV, but ASML only began producing a machine that meets their high volume requirements in q3 of 2019, and has only built 9 of them to date.

In parallel, Intel also developed Sunny Cove core, wifi6, thunderbolt3, gen11 graphics, optane memory controller, optane ssds, nnp chips, dlboost avx512 instructions ... so it isn't like they did nothing while 10nm multi-patterning problems were being solved.

This performance improvement info on Optane is more interesting than the delay. I hadn't seen this before ... only speculation about the density improvement. Is this a big deal?

"... are designed to deliver three times the throughput while reducing application latency by four times ..."

The Cooper Lake and Ice Lake Server roadmap details list Barlow Pass support as a feature. Did something change in the Optane memory interface?

 

[Giroro]

By the way, Micron's first 3D XPoint drive, the x100, is promising 2.5M IOPS and 9GB/s read/write over PCIe 3.0
I don't know which generation of 3D XPoint it's built on, but it was supposed to arrive this year in time to compete with 2nd gen optane
https://www.micron.com/products/advanced-solutions/3d-xpoint-technology/x100

 

[spongiemaster]

By the way, Micron's first 3D XPoint drive, the x100, is promising 2.5M IOPS and 9GB/s read/write over PCIe 3.0
I don't know which generation of 3D XPoint it's built on, but it was supposed to arrive this year in time to compete with 2nd gen optane
https://www.micron.com/products/advanced-solutions/3d-xpoint-technology/x100

A storage "drive" that requires a 3.0 16x slot and an 8 pin power connector? Yikes. Specs are incredible though. Could even be noticeably faster than current high end NVME. Wouldn't expect a consumer product in the near future. Anandtech pointed out that 2nd gen was never mentioned by Micron, so it probably isn't.

 

[InvalidError]

A storage "drive" that requires a 3.0 16x slot and an 8 pin power connector? Yikes.

Catching up with NVMe storage is one of the core reasons why the PCI-SIG is rushing 6.0 - the final spec may arrive as soon as 2021 with the first products in 2022.

Seven years from 3.0 to 4.0, possibly so few as three years from 4.0 to 6.0!

 

[bit_user]

Catching up with NVMe storage is one of the core reasons why the PCI-SIG is rushing 6.0 - the final spec may arrive as soon as 2021 with the first products in 2022.

Just to play devil's advocate, I wonder how important it is to cut this down to x2 lanes, when it's burning so much power. I feel like power is going to be your density-limiting factor.

Clearly, storage is going to be one of the winners of faster interfaces. However, you've also got networking, GPU compute, and AI accelerators, as potential drivers for PCIe 6.0. I mean, even at such speeds, you still need an x2 slot for a single-port 100 Gbps network card.

 

[bit_user]

A storage "drive" that requires a 3.0 16x slot and an 8 pin power connector? Yikes.

Yeah, I don't know if it's more the controller or the actual memory, but I definitely recall reading about 3D XPoint being a huge power hog.

 

[InvalidError]

Just to play devil's advocate, I wonder how important it is to cut this down to x2 lanes

You'll have to ask companies that have racks worth of servers hosting TB-scale in-memory persistent databases how important they think massive IOPS and bandwidth are. Chances are they wouldn't be doing in-memory if storage could keep up