Kaby Lake Systems with Intel Optane SSDs Coming Soon | Hacker News Reader

Kaby Lake Systems with Intel Optane SSDs Coming Soon

anandtech.com

71 pointsgbrown_35 comments

The Optane details seem like a false start, at least for consumers. It may have use in enterprise just due to the endurance for WALs and ZIL etc.

Performance of the NVMe Samsung 960 Pro is to the point where enterprise HHHL cards were not much more than a year ago. These are affordable for mid-high end notebooks, and the 960 Evo delivers close to the same amount of performance for an actual consumer workload not running benchmarks that will easily hit the rest of the mid range and down into low end at lower capacity.

According to Micron, there isn't much demand for large consumer SSDs. 1TB will be held for a while as the sweet spot.

So, where do you need 16GB of cache but for HDDs? And where is an HDD actually useful in a modern notebook? I can see people wanting a few TB in a portable USB enclosure. But as a boot/OS drive, you really want all flash.

I'd rather a big, fast SSD than an Optane+HDD combo. An HDD in a laptop is just yuck. But Optane does seem to out-perform the top-end PCIe enterprise SSDs by a factor of 5-7x, at least judging by the benchmarks Intel has published against it's own top-end SSDs.

It seems like they are putting those drives on x2 PCIE slots there is no way it's going to outpace any NVME drive.

Their capacity is also very small this looks to me like 2005 when Lenovo had SSD cache addin cards for the pc/express card slots.

Give me a 2tb drive or go him until the the 960/960 pro is all the rage.

Well: the industry estimates that SSD for consumer laptops will have a marketshare of ~50% in 201-7-2018.

HDD are $45/TB while SSD at $170/TB(17 cents/GB). if we assume optane is 10x more expensive than SSD , it's at $1.7/GB or $27/16GB. So we're comparing HDD+optane at ~$72 with a $170 SSD. For the lower end of the market or for people who want large capacity , this seems attractive.

Also if the performance is nice enough, it might change the buying decision for some people - why settle for 1TB ?

Most Intel technologies are not for consumers initially. Eventually, consumers may wind up with a product that incorporates the technology after a while. For example, the laptops in the Lenovo announcement that triggered the Anand piece are all in the SMB to Enterprise segment. A person can't walk into BestBuy and walk out with a ThinkPad T4xx.

So I agree that to a first approximation, the critical factor in the consumer space is how well the browser performs. Hitting the hard disk doesn't matter and that's why smartphones and tablets can ship with 16 or 32GB of total mass storage.

What this will might eventually do in the consumer space is to allow competition on total storage size in a way that's cost effective for manufacturers, even a 256GB SSD is a significant fraction of a $350-$500 typical consumer laptop.

I think that you misunderstand the usage of the Optane, it isn't meant to be a cache for an HDD. It will serve as a cache for the NVMe SSD.

In hardware terms even an NVMe SSD's random access is slow compared to DRAM memory.

From the article: Intel's SRT already has caching algorithms tuned for 16GB to 64GB caches in front of hard drives, so now that they have a solution for mediating access to NVMe SSDs it is simple to enable using both features simultaneously. The changes do need to be added to both the RST driver and to the motherboard firmware if booting from a cached volume is to be supported. Backporting and deploying the firmware changes to Skylake motherboards should be possible but is unlikely to happen.

One of the criticisms of the Microsoft Surface Studio a few months ago was that it used a hybrid drive rather than a big SSD. This suggests why Microsoft might have chosen that direction for its hardware's architectural roadmap. It's the kind of customer that would have early access to Intel's pipeline via quid pro quo.

The use of a hard disk instead of ssd to back the Optane suggests the performance differential between SSDs and Optane is not great enough to justify another level in the cache hierarchy (i.e., sticking the optane in front of an ssd doesn't measurably improve performance, or improves performance less than adding RAM for the same cost).

Maybe they will have something more compelling (more pcie lanes or more capacity) next year, and then they'll start backing these with cheap TLC, or start using them for primary storage.

Optane SSD should have latency 10 to 100x lower then even the best SSD. And much higher QD1 throughput.

The problem is how much will this translate into real world experience. Are we still limited by I/O?

And why HDD? It is a waste of space in laptop.

There are still reasonable use cases for mass storage in a laptop: photographs, video, music, etc. And there's a tradeoff of overhead and risk with putting stuff 'in the cloud' in the form of management overhead, network connectivity, and the evaporation of a service upon which someone depends.

Laptop minimalism is just a set of tradeoffs. A Thinkpad X-series or Chromebook isn't for everyone. Neither are Xeons nor Atoms.

Yes, but the market for Mass Storage in a Laptop? That is even more of a niche. My understanding is that even PROs want 1TB SSD rather then 4TB HDD.

I used 'mass storage' to refer to the largest element of the local persistent storage stack. In the end, what matters is not the technology but the performance: what demanding users require is a performance level that is currently provided by SSD's, how that is achieved doesn't matter (marketing and faster alternatives not withstanding).

If Optane achieves the latency benchmarks claimed in [1], I would be happy to use it as swap for large-scale computing. For IO-heavy tasks, it could start to blur the decision to scale up vs scale out.

1 = https://www.extremetech.com/computing/237817-first-intel-opt...

so you would prefer 16GB Optane drive as swap instead of ... you know, actual 16GB DIMM?

At boot time or waking up from hibernation, probably. Or when I'm on low power mode.

In which universe is that a serious question?

In the same one where Lenovo announces brand new Kaby Lake laptops with '16GB Optane cache'.

To be clear, then, I would not use a 16GB DIMM of RAM for swap. I would use all the RAM I had as RAM.

Having something faster than an SSD for swap would be a win; hence my original point. Specifically, if the latency is only 10x higher than RAM (as shown in the image I linked to), then it would allow for trade offs to be made from an engineering perspective regarding scaling up vs scaling out.

We should wait and see what they release before being too critical.

I have little idea of how 3D XPoint works but if it is that fast and non-volatile what prevents it from replacing RAM+Disk? Cost I presume or are there other reasons?

This isn't necessarily holding 3DXpoint back, but one place where RAM is more useful is security. You don't want TLS encryption keys on a non-volatile medium. Who knew that RAM's quickly lose contents when off property would turn out to be a feature?

DRAM is still significantly faster than 3D XPoint.

For now

16GB SSDs are way too small to be usable in that case, especially when paired with an HDD... They should be paired with an slower SSD.

A white ThinkPad. Kill me now. Or at least Lenovo should release the Retro to atone for their sins.

me thinks it's time to start making fused CPU/RAM monster chips. Monster CPURAM chips and PCIe SSDs! nothing else makes sense! then make it smaller - use crystals for a 3D process! START USING BLOOD TO COOL IT AND FEED IT. CALL IT MITTENS. ehm.. got carried a way there.

the days of an Octacore sitting at the helm of 128GB of RAM on a single die are soon upon us, me brethren/ettes.

for the daily joe, all of this doesn't mean much really. if anything, there's some serious improvements coming to our backbone infrastructure and laptop industries. considering improvement in lithium, this might kill off the desktop entirely, if done right.

Consumer based SSD's should be avoided. They are all a big lie when it comes to write performance. Some SSD's are hardly faster than regular hard drives for write operations.

https://blogs.technet.microsoft.com/filecab/2016/11/18/dont-...

So IOPS drops to HDD-like levels after several minutes of 4K mixed reads/writes with write-through. Why is that a reason to avoid consumer grade SSDs? The original article is talking about ThinkPads, which are highly unlikely to ever see this kind of workload. Endurance isn't an issue with the kind of writes you'll typically get on a laptop either. Consumer-grade SSDs are more than fast enough for even heavy consumer-grade workloads.

Why should they be avoided? You should reread you link:

>"Once its reserve runs out, the device runs slower until a break in the action lets it catch back up. With web browsing and other consumer scenarios. the chances of noticing this are small."

Also note that in the last test they are doing "write through" and not using the disk controller cache at all. Something that would not at all be common to consumer level workloads. Additionally most consumers don't run benchmark utilities on their laptops throughout the day.

While write amplification is well known. How about reads? There is no read cliff to fall off of. There is no such thing as seeks on an SSD or does that not count for anything in your assessment that they should be voided?

I can't remember the last time I had I bought a laptop and there was an option to forgo NAND flash in favor of an HDD.

The link talks about consumer SSDs and server workloads, makes a valid point regarding those two things (don't mix them), but doesn't say anything with regards to your comment. Also: Optane is a completely different technology. So, what's your point?

Server workloads is a bad excuse. Most browsers and a lot of applications use SQLite. Firefox and Chrome is known to write a lot more than they read.

http://www.phoronix.com/scan.php?page=article&item=samsung-9...

That's still ten to twenty times faster than a hard drive, and browsers are not exactly slamming their history database. It's a trickle of writes, and a consumer SSD can handle it fine.

You are absolutely right that a consumer SSD can handle it just fine, just like you can surf the web fine with 2GB ram and a single core CPU. But some people pay more bucks because they want better overall performance, and it sucks to pay for a top consumer SSD model that underperform the enterprise models by a huge margin for some basic tasks. Almost everyone who review consumer SSD, with a few rare exceptions, never benchmark sync write performance / latency. Those number matters a lot more than many believe.

I've already seen to many people who use consumer drives with VMware, ZFS Intent log, SQL servers and so on, that gets frustrated because those applications are behaving a lot slower than what they expected and they feel really bad when I tell them that they have been fooled to believe that they bought "fast" SSD drives.

And this is relevant to Optane because?

the author of this article refuses to state if they used a Samsung EVO or PRO (or something like the Crucial MX series.)

Those TLC drives have SLC emulation and will slow down after a certain sustained write time. That is by design. If you dont want that characteristic, dont buy a TLC drive, stick to the PRO etc.

>Second, oh my does the performance drop off. This is no longer an “SSD”: especially as it goes over the write cliff – which is still there – it’s merely a fast HDD, at about 220 IOPS. Writing NAND is slow!

That is a pretty disingenuous conclusion if you dont specify whether the NAND is TLC or MLC.

Reply on news.ycombinator.com