Slashdot Mirror
Intel Launches Optane Memory That Makes Standard Hard Drives Perform Like SSDs (hothardware.com)
MojoKid writes: Intel has officially launched its Optane Memory line of Solid State Drives today, lifting embargo on performance benchmark results as well. Optane Memory is designed to accelerate the storage subsystem on compatible machines, to improve transfer speeds, and reduce latency. It is among the first products to leverage 3D XPoint memory technology that was co-developed by Intel and Micron, offering many of the same properties as NAND flash memory, but with higher endurance and certain performance characteristics that are similar to DRAM. The SSD can be paired to the boot drive in a system, regardless of the capacity or drive type, though Optane Memory will most commonly be linked to slower hard drives. Optane Memory is used as a high-speed repository, as usage patterns on the hard drive are monitored and the most frequently accessed bits of data are copied from the boot drive to the Optane SSD. Since the SSD is used as a cache, it is not presented to the end-user as a separate volume and works transparently in the background. Paired with an inexpensive SATA hard drive, general system performance is more in line with an NVMe SSD. In benchmark testing, Intel Optane Memory delivers a dramatic lift in overall system performance. Boot times, application load time, file searches, and overall system responsiveness are improved significantly. Setting up Intel Optane Memory is also quick and easy with "set it and forget it" type of solution. Optane Memory modules will hit retail this week in 16GB and 32GB capacities, at $44 and $77, respectively.
nobody bought it.
Hybrid drive stuff has been around a while. It works OK up to a very limited point, then it performs like a regular drive. No voodoo magic is going to cache an entire multi-terabyte drive on a tiny expensive SSD. You might boot your OS quicker and have some limited applications perform well but it is strictly limited.
The benefit this provides over the existing hybrid drives, where the flash is in the drive, is the cache is larger and faster, however still tiny compared to a standard modern big hard disk.
Or to the non-buzzword community "use".
Who cares?
So they invented modular hybrid disks or Apple Fusion.
Newsflash, expensive high speed non volatile memory is always best used as a cache till the price comes down. News from 1960.
Some drink at the fountain of knowledge. Others just gargle.
Yup, this is no different to "fusion drives" that have been on the market for years - a small SSD acting as a cache for a large spinning disk.
What is different is that all Kaby Lake Intel chipsets come with support for setting this up in the bios, easily and quickly, so long as you are using an Optane PCIe stick as the cache device.
Once the DIMM packaged versions become available, thats when Optane will really start to take off - slightly slower than DRAM, but not much, but considerably cheaper than DRAM for the same capacity - so you get slightly slower, much much cheaper RAM, meaning large RAM setups (like 1TB plus) are no longer out of many peoples budgets...
Sounds interesting but does this increase hard drive longevity?
This isn't news, it's an advertising for Intel.
There are already many ways to do this without using Intels expensive SSDs.
For instance get an SSHD which basically does the same thing in hardware.
Or use ZFS with the relevant ssd arc cache setup
Or use one of many windows programs that do the same thing
Or use the 10$ SSD/HD cards that are out there that do the same thing
Or use a couple of the linux filesystem modules, that aren't as difficult as ZFS, that do the same thing
Don't see why Intel get a headline for something that's been out for years in many different forms, to suit many different operating environments.
Take a look at the ATTO Disk Benchmark graphs and you'll notice that optane comes in at dead last on both read and write performance. Sure, it'll beat Intel's SSD for the first few milliseconds but it gets absolutely destroyed by all the Samsung SSDs. Though, for all we know, the memory controller made the system retarded. Either way, it's not a winner.
The upside of this is that I learned the Samsung SSD 960 Pro M.2 has excellent performance characteristics.
Anons need not reply. Questions end with a question mark.
And I've got this bridge between Manhattan and Brooklyn that's for sale.
Cheap.
Hanging out for a variant that moves from the HD to the SSD to avoid virus scans, and back. I'm not really hanging out for it, but I expect it, because if it's even remotely possible, some twit will do it.
I was working on a very similar system in 1985. Finally someone made it ready to market...
Perhaps should mention that this overpriced, not-that-impressive cache solution is only compatible with the very latest Kaby Lake systems.
So unless you replaced your PC in the past couple of months, this product is not compatible. Not that I'd recommend anyone to touch it with a ten foot pole anyway, with perhaps an exception for very specific professional applications.
But now we see it is not all that, latency is really good but the endurance is barely over flash, so bad that for enterprise product they had to actually use several times the apparent capacity, otherwise it would die really fast due to wear, and for the general consumer the only product they could come up with was an expensive hard drive accelerator, which sincerely, nobody in their right mind should buy, there are already hybrid HD out there with integrated flash that do the same and do not depend on the motherboard chipset/BIOS to operate, and if you are cheap enough to not by a small (and yet much bigger) SSD for your OS for almost the same price, you are not going to buy this.
Hope they some day can live up to the initial hype, but this is not looking to good.
32 GB of Optane for $77 is $2.40 per GB, Samsung 850 Pro 1 TB is $0.50 per GB. Intel is nearly 5x more expensive.
Hybrid storage systems are common in the enterprise SAN market, but generally to be useful they need something like 20% of capacity to be flash. At ratios of 1-3% of HDD capacity, I don't see the Intel use case as being especially useful.
I had a Seagate 2.5" years ago that was 32 GB flash plus 512GB and it only felt marginally faster than a standard disk drive. You didn't notice serious performance boosts until you went completely flash.
So does Intel have a yield problem or are they still ramping up production facilities to make these in quantity? It's hard to see a system more convoluted than straight SATA or NVMe flash disk being that big of a deal. I think in order to make this product competitive it has to be offered at $/GB competitive with ordinary flash disks or only a small premium.
Yes this is basically the same as a hybrid SSD/HDD or Fusion drive companies are already selling. The key difference is that this cache sits on the PCIE slot directly rather than SAT so it's 5x-6x faster max throughput than SATA interfaced devices will ever be for your most used files. I would assume it works with multiple hard drives as well to speed up what you got.
Still with NVMe PCIE SSDs hitting the consumer market, and prices dropping, this thing has limited lifespan when for a little more money most users would be better suited with a small PCIE SSD and a regular HDD for mass storage.
Promise Ex-8350 128mb ECC ram caching raid sata 1/2 controller (SATA 1/2)
GigaByte IRAM 4gb DDR2-Ram based SSD (SATA I) - strictly for PageFile placement
8gb Kingston DDR-3 RAM (1gb for 64-bit NTFS Compressed Software RamDrive = browser caches, hosts file location, print spooler, %TEMP% ops, + %COMSPEC% location)
I place my hosts file on software ramdisk redirecting it by registry (for performance + security):
HKLM\system\CurrentControlSet\services\Tcpip\Parameters
(Via "DataBasePath" parm)
I also increase hosts' priority assigned its load/read:
[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\services\Tcpip\ServiceProvider]
"HostsPriority"=dword:00000005
"DnsPriority"=dword:00000006
"LocalPriority"=dword:00000007
"NetbtPriority"=dword:00000008
ALL disks = NTFS timestamp = off + perf counters off + excess services off.
APK
P.S.=> Less work done on OS/Programs SSD bootdisk = faster doing less bs vs. REAL work + less fragmentation on it ... apk
Intel Turbo Memory take 2?
https://en.wikipedia.org/wiki/Intel_Turbo_Memory
What's the difference between this and the Fusion Drive Apple's been shipping for, like, 5 or 6 years?
Bleh,think I'll just stick to 10000rpm raptor drives and lots of ram..
you can get real PCI-E SSD for about $1/gig or less and you don't have to deal with any of the fake raid bs.
Non "standard" memory won't make me switch from my SSD back to a mechanical one. I'm not into speed like I was in the late 80's through late 90's. With the SSD, photoshop boots in about 10 seconds, fast enough for me.
You mean caching slow storage with faster storage improves performance?
Shut. Down. Everything.
we have seen these type of quirky hybrid storage solutions before. I think at this point in time it's clear that the best solution is a pure SSD, best options being those from Samsung and Toshiba, given their price, performance, and reliability.
HGST (formerly owned by Hitachi) performs very well too, but they will eventually shift from Hitachi engineers and R&D in Japan, to Western Digital's own engineers and manufacturing, and it will likely mean that reliability and endurance will drop significantly and match that of WD's other products.
Worse... It's just a cache....
"File to fit, pound to insert, paint to match" - Aircraft Maintenance 101
I think Intel either has a yield problem, or simply that X-Point is a lot more expensive to manufacture than they pretend. NAND and DRAM have very mature manufacturing processes that are hard to beat in cost.
I think in fact that it costs a lot more than $77 to manufacture, that's why they enforce all these artificial restrictions (only Kaby lake only 200 series motherboards) - because they are selling below cost and don't want to hurt their margins too much.
Now that 3D-xpoint is finally available, does anyone have hard numbers for it's read life (how many times it can be read) and/or shelf life (how long it will last without being turned on)?
This is just like those bite-sized freebies in your local grocery store. By themselves, for practical use these drives make little practical sense. Review sites shouldn't have bothered with that Intel's cache bullshit. Instead just treat them as teeny tiny SSD drives serving as a technology preview.
Bottom line - latency on these things is awesome. Write granularity is good too - will be awesome with proper abstraction level in OS. Write endurance - we don't know - waiting for somebody to write the shit out of them.
We'll see if Intel/Micron can get a market foothold (being sandwiched between DRAM and flash SSDs is no picnic).
The era of primary and secondary storage is coming to a close, and when unified memory is widely available we will often wonder why we mucked around with disk drives, SSD, and even filesystems. Bubble memory tried to be that promise, but failed to keep up with Moore's law. But the revolution is guaranteed to happen and we have some good theories on what form it will take.
Prepare for all operating systems and applications to be obsolete.
The point everyone is making is that the new technology has to be competitive somehow, either price, performance, capacity, something, but this one is pretty much pointless, it is more expensive, has greater performance but not really impactfull (my spreadsheet now opens in 0.003s with Xpoint instead of 0.005s with flash SSD, yey!?), for now capacity is very limited and endurance is a far cry from the promised during the first announcements, in the order of 30 fold less.
The strong criticism is about a pointless product sold as great with a technology that should be a big leap forward that is not that much better.
God, marketers are a plague. It's a cache, just call it that.
Unless there is a pretty good reason to still use a HDD, why not just get an SSD? Would take up less space than a HDD and one of these things.
This is my sig, there are many like it but this one is mine
I just wanted to remind everyone that hybrid drives with flash memory mini-SSDs have always been inconsistent, overpriced, underperforming crap ever since they were invented. My company has tested dozens of them and they're all completely unpredictable. Also a 16GB buffer? Great, my game's sound and data files are 19GB. It's just not practical and the firmware isn't nearly smart enough nor are they big enough. A 64GB one could just barely get by on loading common operating instructions but that's not how big any of them are.
Answer the question in my subject UNIDENTIFIABLE anonymous TRUE coward: Go for it - I'll rip your ass in 1/2, easily... mirroring is your clue & WHY you'll "Run, Forrest: RUN!!!" & it happens 2 ways here (IF you're smart enough to get that - somehow, despite your below bacterium IQ, you will - I give part of how above in what you replied to SO stupidly on your end).
* To quote Will Smith from 1 of my fav. films? "6 redundant drives" (OR the data on them).
APK
P.S.=> Dumbshit... apk
Hate to tell you all this... but there is NO SUBSTITUTE for REAL PHYSICAL RAM in your machines.
This has been true since the dawn of computing.
From RAMDRV.sys all the way through ZFS on FreeBSD.
If you want awesome disk performance, you cache or run one in RAM.
If you can't wait for the data to be primed on boot, that's your problem.
But once it gets to RAM, you're golden.
And it doesn't cost any more than a pile of DIMMs.
For that matter, most of you don't even know that using a simple mirror doubles your read performance at no more than the cheap cost of HDD/$/GB.
Far fucking lower than this $$$$$ unencrypted Optane shit.
There's so many alternatives available, as other people point (RAM disks, SSD to HDD adapters for caching purpose, etc)
And they work on pretty much any system built in the last decade.. as opposed to the Intel Optane memory which works only on some specific high-end configurations (which don't need that cache anyways, since their OS would be running off a modern SSD most likely)
There's almost no viable scenario where this product would be used, where we wouldn't have better alternatives.
X-Point has the potential to eclipse traditional flash. Sure it's new and expensive but if, at launch, it's already faster and more dense than traditional flash that's a big sign of things to come.
Density is king. Pushes your physical package size, manufacturing, and BOM costs down. Speed is also obviously good.
If Intel/Micro can ramp up production, improve their process, and/or expand their production they could turn the solid state storage market on it's head.
The traditional flash giants are rightly scared shirtless.. Which is why you see so much aggressive trash talking of X-Point in online forums. If X-point improves substantially Intel/Micron will take big chunks of their business... Starting with the very highest margin high performance markets.
> are SANs that are designed for backups
High speed cache is good for data that is accessed, then accessed again a few seconds later. Web servers are a good example - the same page may be loaded many thousands of times per hour, or even thousands of times per minute.
For backup, each sector of data is accessed no more than about once per day. In my experience, backup is where you want sustained throughout, caching doesn't help. We use wide arrays.
Just what I would have said, but better, thanks.
I know ZFS's implementation best, and to be honest it's not as good as it could have been. The L2ARC (read cache) is a simple Least Recently Used list, but with the caveat that re-used blocks are not moved to the front of the list. I expected a lot more from the implementors (and inventors?) of the ARC (their RAM cache). It has some logic to prevent caching streaming reads though, which is good. Additionally, it wastes a significant amount of RAM per L2ARC size on overhead, so using a TB of flash with 32GB of RAM is out of the question. It limits the write rate filling the L2ARC (pretty low by default), which doesn't seem to correlate to whether one would want to cache the data.
Their SSD write cache (log devices) perhaps has less problems. It does require data to stay in RAM until written to slow disk, but at least ZFS is good at writing sequentially, so random writes will be quite fast anyway.
I won't go changing to this new tech from Intel, but I wish ZFS could optimize the L2ARC a bit more.
you can get real PCI-E SSD for about $1/gig or less and you don't have to deal with any of the fake raid bs.
How do you define "real PCI-E SSD"? Would you include a $9,000 enterprise-grade PCI-E SSD from Intel with up to 850,000 IOPS 4K random reads?
Intel DC P3608
Guess what? It consists of two SSD's that are configured as RAID0 by Intel's RSTe driver software.
... that you need to buy an expensive Intel-only memory.
Intel chipsets had the feature to use an ssd as a cache for an hard disk for many years (it was called "Intel Smart Response"). Until now, you could use any ssd, optimal for pairing an high-capacity low-power HDD with a mSATA SSD on a laptop to gain a great boost in I/O performance and ssd-like boot times.
but if you buy the premise that, you have a brand new system with a new chipset and cpu but for some reason decided on older spinning disks without built in cache (because you like your 1990's drive) then it all makes sense.
It looks to me like Intel/Micron wants to get their new product in the hands of early adopters first instead of trying to appeal to the mass market off the bat. This technology is radically different than what we are used to. They are looking for people who see the vision of what they are trying to accomplish and are willing to throw time, effort, and money at moving that technology forward. People who buy this first product are not the ones who think 'I am only interested if this is much better, cheaper, and faster than what I am now using!' but rather 'How can I help prove this technology by putting it through the paces because I like where this can lead to?' That is the mentality of an early adopter as opposed to the masses.
Could you go in to more detail regarding the "10$ SSD/HD cards" and "couple of linux filesystem modules"?
They sound very interesting?
"Traditional" NAND flash was much more expensive than spinning rust but came in sizes useful at least for boot disk applications *and* delivered overwhelmingly better performance from the same bus/connection as spinning rust.
IMHO, Intel can't pimp this out as faster than NAND flash for more money. Like CPUs, flash storage has more or less hit the speed levels where more speed simply isn't that useful outside of very narrow use cases.
The angle they needed to work was density and write endurance. There's still a fair use case for spinning rust at certain scales, driven mostly by slot limits in server and storage chassis. If you want 40 TB but only have 10 slots, you have to use spinning rust. Providing a solid state disk at this density with superior write endurance would really be a market disruption.
So pair a $55 2Tb HDD with $77 32GB Optane drive = effectively a $132 2Tb HDD w/ near SSD access speeds?
SSD street prices for 1TB SSDs are about $300.
Sounds like a deal to me.
Self-importance and self-indulgence is the root of ALL evil.