I would have thought its far too early days to make calls like that. XPoint will come down in price through volume and nailing the manufacturing process down tighter. And speeds will improve as they iterate on the designs.
Wikibon drops bomb, says Intel's Optane could be a flop...tane
Intel's Optane 3D XPoint drives could be doomed to fail in the mass market because their performance and endurance advantages over 3D NAND SSDs are "nominal". So claims Wikibon CTO David Floyer in a report that looks at Optane's performance and endurance advantages over 3D Flash by comparing Intel's 32GB Optane Series PCIe M.2 …
COMMENTS
-
-
Tuesday 11th July 2017 06:38 GMT Mikel
A terabyte of XPoint will be as big as a toaster and almost as hot. They've got about six generations of package shrinking to catch up on density with flash at the package level. Also that much will cost as much as a small house.
It's just not worth it yet for mainstream datacenter use. Maybe in a few years it will be worth checking on their progress. If it hasn't been cancelled yet.
-
-
Monday 10th July 2017 15:07 GMT Anonymous Coward
"Nearly as good"
It's apples and oranges.
Xpoint is not an SSD, it's not RAM. It's inbetween. If the PC/system does not need it, it does not need it.
The article also makes mistakes, while a RAM disk/cache is faster it is often more expensive (even with Xpoint high priced entry to the market).
If Xpoint does not drop in price and does not increase in speed/size then yep, it will fail. But if it keeps up, it will have a few very specific use cases (where instant on/power failure redundancy is needed and an SSD does not have the write cycles/speed).
-
Monday 10th July 2017 16:20 GMT yet_another_wumpus
Re: "Nearly as good"
@TechnicalBen: "It's apples and oranges.
Xpoint is not an SSD, it's not RAM. It's inbetween. If the PC/system does not need it, it does not need it."
That's the catch. 3DNAND is killing it in storage, and the use of psuedoSLC is pretty much killing the "inbetween needs" as well. It would make all kinds of sense if 3dNAND was always "slow" (relative to xpoint).
@bldrco: "While 3DXP is more expensive than 3D NAND, it's an order of magnitude cheaper than the DRAM that it's actually aimed at replacing."
Really? Microcenter spammed me yesterday with a sale on xpoint: $80/32GB. Pcpartpicker show 32G for $209. So less than a factor of 3 (newegg had a similar price, amazon had 0 hits on "optane"). The big kicker is that they have to get the endurance up. Endurance is believed to be a few times higher than SLC, not nearly enough to replace DRAM.
If they get that endurance up, I suspect they can indeed replace DRAM. I also think that is what is holding up the DDR4-slot stuff. That and not providing *any* information to provide compatibility.
-
-
-
-
Monday 10th July 2017 15:10 GMT Anonymous Coward
Re: Wrong comparison
Yep. Just as early SSDs were there for OS only, as the cost for the entire system being 500gb flash was prohibitive, we get the same with Xpoint. Would would have dreamed of 1 or 2 tb flash drives when they first released? Now would you want to go back to booting off a platter?
But in 5 years time? A 500gb Xpoint PC, with no ram, just a "SSD", or would it be a "XSSD"? Would boot instantly, power off instantly, resume instantly... it's just seeing where they price it, between the cost of flash + DRAM, or greater than Flash + DRAM to milk the profit. I'd assume they will milk the profit, and that will kill uptake. But if they balance production and demand, it will be ok for now.
-
-
-
Tuesday 11th July 2017 08:41 GMT ilmari
Re: Wrong comparison
They already do, pretty much. Their sleep states would give around 2-4 weeks battery life, It's just that they power up constantly to check farcebook, twitters, instagrams, snapchats, whatsapps, Skype, telegram, google+, gmail, oemaccount, google play, weather widgets, location, etc...
-
-
-
-
Monday 10th July 2017 15:07 GMT wjchan
Current offering is not compelling...
After testing the Intel P4800X 375GB Optane SSD drive, I believe they will have a hard time selling that. Real-world performance improvement is marginal, and they haver to cherry-pick benchmarks to make their case. That's probably why they have held back the availability of that drive. They need to bump the 7-channel controller to 12 or 14. Better yet, the DIMM version will eliminate a bunch of bottlenecks and truly show off the underlying performance advantage of the memory technology.
-
Monday 10th July 2017 15:12 GMT Anonymous Coward
Re: Current offering is not compelling...
IMO the only place this tech makes sense is as a total replacement for SSD (but we have to wait for the cost to drop for that, as we did for platters vs SSD), or as a total replacement for DRAM.
So it's a waiting game. Remember, even the first petrol/diesel cars were slower than horses! :D
-
-
Monday 10th July 2017 15:27 GMT Anonymous Coward
Just seen a good idea else where...
Someone posted about the future of PCs being the RAM as HBM on the CPU, super quick, but smaller than now (8gb and 16gb vs the 24, 32 and 64 options) but with Xpoint in the DRAM slots (or as an option to). That could work, giving the current ram a boost in speed being so closely integrated into the CPU pipeline, and giving a nice power off, speed, size and price friendly "cache" of storage for an SSD or spinning rust.
-
Monday 10th July 2017 16:13 GMT Anonymous Coward
Thunderbolt?
This screams Thunderbolt 2. Sure, it might be superior on paper, but when the sole source is Intel, you need an Intel controller, Intel CPU, Intel driver, at 4x the price... what's the draw? The performance differential surely isn't enough for anything but niche cases.
Don't worry guys, USB is going to be dead any day now...
-
Monday 10th July 2017 17:12 GMT talk_is_cheap
We are still waiting for the 'real' product
Currently, all Intel seems to have to show is a 32GByte stick that is aimed at caching. Something of a dead end product considering the performance caches of many U.2. sticks are now larger.
The original pitch for this tech was 256GByte memory modules for installation into large fat servers where each Intel CPU could address 8+ memory sticks. The possible resulting I/O was meant to interest the big data and VM markets.
Today I wonder how many designers are more focussed on what they can do with 2 AMD top end processors and 128 PCI 3 lanes - that's a lot of 2TB U.2. sticks without paying premium rates to Intel for a lot less storage capacity.
-
Monday 10th July 2017 19:41 GMT bldrco
Re: We are still waiting for the 'real' product
I think the P4800X is the 3DXP Intel currently has available for servers: https://www.intel.com/content/www/us/en/products/memory-storage/solid-state-drives/data-center-ssds/optane-dc-p4800x-series.html?wapkw=p4800x
The DIMM version doesn't appear to be available, though.
-
-
Monday 17th July 2017 21:33 GMT skoenigsberg
I work for Infinio - a leading provider of high performance I/O through a RAM-centric server-side cache.
We work with customers every day who are looking for high performance, (performance they just can't get any other way) and a few things come up repeatedly:
1. It's all about workload: The customers who care most about microsecond latency (i.e. going 20X faster than all-flash storage) rely on data processing jobs whose value to the business increases as their time to completion decreases: by hours, minutes, and seconds. Validating or dismissing a technology outside the context of workload is largely irrelevant.
2. For the times when a DRAM-only cache can't deliver the necessary performance (because of working set size), there's a huge benefit to having Optane over 3D NAND as the tier after DRAM.
3. It's not always about relative cost. For customers whose business results can be impacted by these technologies, absolute cost is a far more relevant metric. When you look at the impact of data reduction technologies (which are, admittedly, relevant across all media types), the difference in cost from an absolute perspective is often dwarfed by the overall ROI from the business results.
You can see our more complete POV blog post here: http://bit.ly/2u2yLcO