Intel Optane Coming April 24th

"Intel’s Optane Memory could be the most revolutionary letdown in storage history. Announced Monday morning, these first consumer Optane-based devices will be available April 24 in two M.2 trims: A 16GB model for $44 and a 32GB Optane Memory device for $77. Both are rated for crazy-fast read speeds of 1.2GBps and writes of 280MBps." http://www.pcworld.com/article/3184...ission-make-hard-drives-faster-than-ssds.html

 

Agreed. 16GB/32GB cache drives for 2017 on?

Skip.

 

My comment above is especially true as the Optane memory will only work, afaik, with HDD's.

When Intel offers us real Optane memory (DIMM's), then we'll have something to talk about.

Not at 32GB or 64GB either... I want Optane DIMM's from at least 256 (2x 128GB) or more.

Still, will be interesting to see actual hands on testing with these in a few short weeks.

 

Yeah, I'm not very interested in the current incarnation of Optane.

 

I hope we get some reviews for this. I am curious about optane. But we may have to wait 2 more years before it hits mainstream.

 

See:
http://www.tomshardware.com/reviews/intel-optane-3d-xpoint-memory,5032.html

See:
http://www.anandtech.com/show/11210/the-intel-optane-memory-ssd-review-32gb-of-kaby-lake-caching


It's interesting what a few short weeks make.

Even better? Intel proved me wrong (see post #2 and #3 above...).


While I can't see Optane Memory being used in a notebook today (capacity too low and power usage is too high), it would make for a very interesting setup for a NAS or other desktop based HDD powered platform.

What is most eye opening? The fact that even at a puny 32GB size, the Intel Optane Memory SSD is still an order of magnitude better than an 850 EVO with more than 500GB of nand onboard... (at the queue depth sizes workstation users care about...). Even CRAPID is surpassed with an Optane Powered HDD in these lower queue depths - and that's with a RAM cache that is many times superior to the PCIe bus/interface can hope for, let alone the SATA3 interface/bus that makes DRAM's performance seem like it's on Mount Everest...

See:
http://www.tomshardware.com/reviews/intel-optane-3d-xpoint-memory,5032-4.html

(See '80 Percent Mixed Random Workload' graph in the link above).

What is also intriguing is that Optane seems to be an (mostly) full duplex type of storage. Very unlike anything we've had access to in the past.

Writes don't slow down the reads significantly - and vice versa. Compared to even NVMe PCIe x4 or x8 nand based SSD's, Optane 3D XPoint SSD's are what compute platforms have been waiting for since the HDD was invented (yeah; most/all current SSD's don't do anything to alleviate this particular issue).

Aside:
I was on the phone with a client who was considering purchasing new platforms and had almost settled on the specific components when I saw these reviews while I was clicking idly while he was speaking. I pointed them out to him and he has (so far) effectively cancelled his order. That's how important Intel Optane Memory SSD's are - even in their current incarnation of 'gen 1'.

Anandtech states:

Tomshardware states:

With very little performance hit as the drive fills up, no need to overprovision, 64 addressable areas on each chip and effectively full duplex vs. anything we can use right now... the direction that 3D XPoint tech is pointing at is easy to see: massive productivity improvements with any and all compatible Intel hardware.

With a mere 32GB cache, Intel has effectively 'answered' the AMD Ryzen 'threat' for me. Can't wait for a power optimized (for mobile usage) version of this tech.

If productivity is the goal... (no matter what your workstation class workflows are).

 

Kinda meh. For now.

 

I agree.

(The reviewer, right?).

 

Both. Horrid execution of testing and an even worse attempt by Intel to sell something they can't quantify just yet.

 

I don't see where Intel is doing a bad job of 'selling' this tech?

The AnandTech and Tomshardware reviews are enough to show that they hit the nail on the head (with what they promised).

As for the M/B and Kaby Lake processor requirements? BAU*... yawn.

I still need to see this kit in action for myself, of course. But if it comes anywhere close to what I've read already - every single desktop installation that can't/won't drop in a 1TB or larger SSD will have the largest Optane Memory SSD that is available, included.

What is left to quantify? A 10TB HDD (fastest HDD available today - faster (sequential) than many TLC SSD drives are capable of, sustained...) is accelerated for common tasks to above any SSD now available for the same or less.

With more mundane HDD examples, the performance of the system increases exponentially for mere $$...

Home run for Intel. And; it gets to jab AMD in the face once more.

*BAU=Business As Usual

 

See:
https://www.pcper.com/reviews/Stora...2GB-Review-Faster-Lightning/Depth-Performance

Scroll down to the 'Latency Percentile, 4KB Random Read QD1' chart to see how much faster a CPU, L1, L2 and L3 cache are than even DRAM (the 4th slowest part of the compute chain)... making any and all current SSD's/HDD's almost inconsequential to how a computer performs most workloads (i.e. even HDD's are still viable today for many, many workloads/workflows...). This includes PCIe x4, SATA3 and the lowly mSATA options we have available now. Except of course for Optane 3D Xpoint samples...

There is a 100,000x latency range between the still relatively slow RAM and the fastest HDD (10K RPM).

3D XPoint is still ~40,000x slower than what the CPU can handle... but it's a good start. DRAM is still ~235x slower (vs. a CPU @ 4GHz) and Optane is only ~169x slower than DRAM.


When greater than 0.5TB Optane solutions are available - at much lower power requirements (at least at idle) than XPoint currently needs - then the platforms I have today will be retired.

Intel is moving very slowly with Optane 3D XPoint (and... I commend them for that). But what that should get us are bullet proof upgrades not just to our storage - but also to our overall system productivity levels and reliability/dependability aspects too. Not only will mechanical HDD's be required to do less (caching...) while providing us with increased performance - but future versions of Win10 could be made to not just 'boot' from an Optane drive - but just 'turn on' from that volume instead.

I am also glad that Intel is focusing on their latest platforms, forward. My wallet may not be; but I don't want legacy dependencies to slow down or derail where Optane is heading...

Newer M/B's, newer CPU's and current BIOS'/software/drivers - check!

I don't upgrade my systems because time and again that has proven to be a half step forward and two steps back vs. a new platform (besides that they're maxed out when they're initially spec'd...).

A quick search for a new 100% Optane platform (with the minimum processor option - for testing), an 270 series M/B with 2x M.2 and 64GB of RAM with the 32GB Optane Memory SSD is less than $1K... when those parts are locally available to me, then internal testing will begin.

Now? I thinking/dreaming what 1.5TB of Optane would do for my workflows.

See:
http://www.hw4all.com/intel-optane-ssd-900p-new-line/

(Power consumption be damned).

One thing I have no doubt about: just like building a faster car - the higher the octane/optane the higher the performance will be.

 

Optane SSD trying to accelerate an SSD...... result....... nothing a normal user would notice other than in synthetic benchmarks:

 

Phoenix, that 'reviewer' is anything but... see post#7 in this thread...

When an 850 EVO (250GB) is improved upon (see link above your post) and even 500GB SSD's (see previous links in this thread) are also enhanced (significantly) with an ~8x and ~16x smaller Optane module... that tells you something great should be expected from 3D XPoint (Optane and hopefully QuantX too, soon).

Almost nobody that puts an SSD in a system is a normal user: as almost all systems sold today are still sold with a HDD as the default config and that is what they would use...

Getting faster than SSD speeds via a software/hardware caching method hasn't worked before now - there are sites that hint that is because today's SSD's are faster; they're wrong. CRAPID doesn't work for a normal workstation even with using RAM (many times faster/lower latency than Optane) as it's media.

The fact that caching works - and works extremely well - even with a mere 32GB, 2 'nand'/channel product is making waves. And yeah; it does speed up an SSD in real world usage.

Of course, it will depend on the SSD being cached... but if/when Optane is offered in bigger than 32GB capacities and also in lower (at least; idle) power modes - it will make the current offerings obsolete, immediately - even for 'normal users', let alone power users like everyone here is.

This is the future of computing as I see things today; Optane (at larger capacities) offered as both NVMe x2, x4 and even DIMM's will be another level of caching that today's CPU's demand to keep them stuffed with instructions and data, constantly.

Whenever I feel a little disappointed with what Optane is offering today, I keep reminding myself that this is day 3 of gen 1.

Nowhere to go but up.

 

I think the reason Samsung's Rapid mode doesn't work as well is because DRAM is volatile, and the system already uses DRAM for accelerating.

I've seen some 1st run tests that show some gain for Optane Memory. It doesn't sound possible, but let's think for a bit. There are small size files, usually system files that are universally used by various applications. Some reviews show after the first reboot, the system does some Optane specific optimizations. The small system files might be cached even before the application is loaded for the first time.

1st run tests
http://www.digitaltrends.com/hard-drive-reviews/intel-optane-32gb-review/
http://hothardware.com/reviews/intel-optane-memory-with-3d-xpoint-review-and-performance?page=3
https://www.gamespot.com/articles/intel-optane-memory-review/1100-6449530/

Looking back at Intel's SRT 1st run tests, the benefits are essentially zero. The algorithm and the type of files it caches are indeed different. Of course back in the SRT days SSDs were much more expensive, and its far cheaper now so it has to be much more competitive. I also theorize the reason NAND SSD caching wasn't as effective is because NAND has inherent limits like having significantly different performance when the drive is dirty versus nearly empty and clean.

Take a look at this preconditioning test with its big brother, P4800X: http://www.tomshardware.com/reviews/intel-optane-3d-xpoint-p4800x,5030-5.html

There's no such thing as "steady-state" performance with Optane! That makes it a far better cache. How can you accelerate it properly when it has to deal with its own TRIM and massively fluctuating performance?

http://techreport.com/review/30813/samsung-960-pro-2tb-ssd-reviewed/5

Loading tests show there's almost no difference between the first good consumer SSD, the Intel X25-M, versus the best-in-class NVMe SSD, the Samsung 960 Pro. In fact, the difference between the slowest and the fastest device is seperated by hundreds of milliseconds. We complain about lack of CPU speed improvements. Well, SSDs are even worse. Big picture wise, Optane Memory is "there".

The difference between a 2009 SSD and 2017 one is much more minute. Support for TRIM, better sustained performance, better software.

I can see why Optane Memory is so polarizing. Intel could have done better too. Implementing a idle power state that brings it down to 0.1W would have made it much more acceptable to laptops(even though we'll see laptops using it). Perhaps they could have included Core-based Pentiums among acceptable CPUs, if they view Celeron as being too low. The entire reason they limited it to Core i3/i5/i7 is because they want to move people to those CPUs. But I think its sort of short sighted. I think Pentium/i3/i5/i7 would have significantly increase adoption and help with volume regarding 3D XPoint products(which will ultimately bring prices down faster and improve yield learning).

I have a Desktop with X25-M 80GB and 300GB HDD(well, it was free). Even as sort of an enthusiast, I see SSDs being too pricey for the capacity. I would move to Optane Memory setup, even if its just for fun. Dual drive setups suck.

 

I personally can't tolerate a HDD inside a notebook nowadays. And thin&light notebooks have m.2 slot anyway, no place for HDD there. What I want is cheaper high-capacity SSDs (which won't crawl to HDD speeds, of course)... yet industry gives everything but. This Optane caching crap is just another disappointment, honestly.

 

I agree.

It is borderline impossible to get a laptop or a desktop machine without an SSD or M.2 drive in Scandinavia nowadays.

The amount of time wasted having an HDD as your main drive is absurd.

EDIT: Just wanted to say that I cannot phantom why someone would purchase an HDD to then purchase the Intel Optane drive to boost their speeds. Additionally, most desktops/laptops that actually NEED the Intel Optane technology because their HDD is bottlenecking their system, do NOT even have the capability to install such technology, it is too new and the WS is too outdated! Arghhh - sorry for the rant.

 

The reason CRAPID doesn't work (at least for any real world workflow I've tried...) is because the driver is slower than the boost in performance it promises; hence, 'CRAPID' mode. Optane works and the system already uses DRAM too, so that argument is shot down.

The reason Optane works from first run is because when you set up the system to use Optane Memory, it pre-fills the cache.

It also works because the (read, mostly) latency is an order of magnitude better than any SSD currently on the market. Dirty/full has nothing to do with it. SRT with 70% OP'ing on a clean/brand new SSD was just as slow and useless and any further testing I did back then. Today's SSD's won't fair much better (keep in mind; on 'my' workflows) - sure; they're faster - but they're just as laggy...

The Intel X-25M was good - today's SSD's are definitely better though. Consistency is not enough: high(er) burst speeds are important too.

Can't really compare an X-25M to today's SSD's though... unless you have a brand new one that you install today's O/S's on... Many, many SSD's of that period felt slower than my vRaptors from that time (yeah; laggy...) Even the X-25M, while being the best in class, still exhibited that 'lag' vs. a properly tuned HDD storage subsystem.

Will Optane get more power efficient? Yeah. It has to. But here, in the beginning, if you want more performance than you need to be less efficient with the power... for DT's, it is an easy choice (1W idle won't make a difference either way). For mobile systems? 1W additional power at idle is huge...

Wait for gen 2...

For DT's, 10 drive setups are my preferred setup. 2 to 4 SSD's for O/S, Programs and Temp/Scratch drive space and the rest of the drive bays filled with the biggest and fastest 7200RPM HDD's available at purchase setup with Win10's Storage Spaces.

If Optane Memory works for the above; in my next desktop platforms I'll be sure to include it too for $77.


Both the X-25M and the 300GB HDD are slowing down your system if it is a Sandy Bridge i5 or better/newer platform. Both have been greatly surpassed (especially if OP'ing and shortstroking, respectively).

 

Agreed!

Here's a scenario: someone that wants a fast responsive system and they're buying today; but they also need the biggest storage capacity possible - on a budget.

Hello Optane!

 

@tilleroftheearth please make sure to check if Optane works as a storage drive on Skylake once you lay your hands on one. (=

 

Don't have to check; others already have and yeah; it does.

 

The volatile nature of DRAM plays a part too. On the next boot, it won't have anything to pull data from. It strictly works after you load something.

I think you are missing the point. You can't pre-fill applications when you have never loaded them. I'm saying the benefits are likely because its caching Windows system files that are commonly used by the same applications. And Intel engineers would have known to program it to put those files in the beginning.

Sure it does. Caching setups are very small. It'll take far longer to fill up even a 128GB SSD compared to a 16GB one, or even 8GB used for caching. I'd have to assume that dirty state is reached in the first few hours of usage. The problem I'm saying is with NAND caching, the SSD controller is trying to manage garbage collection, the big drop in performance when dirty, while the caching algorithm is trying to do something with the HDD. We had a big talk about that in the X25-M days. TRIM support and whether to do idle garbage collection was a big deal in the enthusiast space. It's like a teacher trying to teach students when he has occasional serious depression and bursts of memory loss.

I moved from a Raptor setup to the X25-M. Actually Tech Report and other tests show in terms of loading times, the modern NVMe drives have nearly no gain. The 4K random read QD1 tests show the X25-M doing nearly 6K IOPS while 960 Pro does 14.7K. Optane Memory is at 70K(when tested solo). If you translate that into latency, X25-M was at 160us while 960 Pro was at 68us. Optane Memory would be at 14us. Responsiveness is highly dependent on random reads. I'm not saying Optane Memory + HDD is faster. I'm saying the gains are very small for latest SSDs.

If you have a version with TRIM support, like with Gen 2 X25-M, the difference in responsiveness is minimal compared to latest NVMe drives. I know, I put Intel SSD 540 into my parents system and once it got really slow. Web browsing became slow. I ran TRIM and the difference was quite noticeable.

 

I disagree on every point you make below. Sorry.

CRAPID uses disk space to save it's last state. This is just crappy/inefficient programming that Windows does better (and has for a very long time).

Yes, you can pre-fill a cache before a program is loaded. They're called PF (pre fetch) files since the days of XP.

You're ignoring what I wrote; "Even with 70% OP'ing...". No nand was 'dirty'. Speed up still non-existent.

Comparing loading times to see the value/benefits of SSD's is like saying a horse is faster to 'wake up' than a car is at -40C... You only do that once per session... Unless your 'workflow' is to just load programs, continuously, but never actually use them.

That Intel SSD 540 is also a very slow SSD vs. anything current/higher end. (Sorry). Sure, after running TRIM on it the difference would be noticeable - but just you comparing it to the circa 2008 X-25M is telling... The '540 may be almost 8 years newer; but that doesn't mean it is better than a top of the line 2008 model though...


I agree the gains are smaller and smaller for the latest (and 'best') SSD's - but they are still a far cry from circa 2008 tech - even if it was 'gen 2'...

There are some (otherwise 'horrible' eMMC) systems I've tried (i.e. NOT an SSD inside) that are considerably faster than the original X-25M and X-25M gen 2's I've had experience with.

If the tech hadn't really advanced; SSD's would be dead water today.

 

I am just waiting for someone to have the Optane in their WS for a while so we can see some (allegedly) real performance boosts.

 

See:
http://www.computerworld.com/articl...ry-nand-flash-killer-or-dram-replacement.html


Some very good info there, with many links.

Including Samsung's Z-NAND (quote below from the link in the quote above).

Since this is the first time I've heard of Z-NAND (which should be released sometime this year; 2017...) - and it "shares the fundamental structure of V-NAND", it seems Micron and Intel are far, far ahead of anyone else... Even with being 2 years (since 2015...) late with Optane ready products...

I'd be surprised if Samsung even releases it (as stated above), tbh. With much slower latency than what Optane provides, does it even make sense to?

 

Tried to find somewhere to post this link to a Toms Hardware article, this seems to be the place!
http://www.tomshardware.co.uk/intel-optane-raid-report,review-33903.html

It's a way of using x3 Intel Optane drives in RAID 0 as boot drive. Apparently the reviewer felt that the system was clearly more repsonsive in daily use in comparison to NVMe drives, but not the big difference that you see going from an HDD to a plain old SATA SSD. The guy should have done some real world load time tests, but it was mostly synthetics it seems.

 

Yeah, read that article this morning on the US site of Tom's...

What has me very interested is the fact that the author thinks that both the 16GB and 32GB versions of current Optane Memory products are just single channel and with 3x RAID0 array we get this type of performance that puts NVMe drives in their place (i.e. with the rest of the SSD's...). While the writes are restricted so as to not burn through the warranty too fast - the reads already max out the PCIe interface.

What will be interesting is enough Optane 'nand' (1TB or more) with a controller that will take those write speeds to the limits of PCIe.

What will be worth throwing away every computer I have currently is a handful of TB's of DIMM based Optane flash that is bootable...

From some of the graphs in that link; I'd be inclined to think that HDD>SSD is the same as SSD>Optane*** - but the author is just holding his enthusiasm in check.


*** Not the as-tested RAID0 setup; I thinking a more rounded Optane offering (min 1TB Optane flash w/equal read/write speeds and at least 4 channel controller or better).

 

XPoint under an electron microscope:

See:
http://www.anandtech.com/show/11454/techinsights-publishes-preliminary-analysis-of-3d-xpoint-memory

Considering this is first gen, commercial release; it bodes very well for future iterations.

Just hope this comes soon(er) and in usable quantities for consumers (i.e. 1TB or 2TB minimum - smaller capacities need not apply).

 

I am looking forward to this too, however, when they reach that point, wouldn't RAM cache disks overtake Intel's vision? RAM is getting faster and cheaper by the minute.

 

I've never met a RAM cache disk I've liked (let alone provide me with actual productivity boost with reliability for my workflows/workloads...).

DIMM's are not that cheap... Optane is poised to overtake it (capacity wise) very soon.

In the end; we'll still need both for the medium term future - both RAM and Optane. What will be a game changer is when the O/S (WinxxxPro for me...) catches up with the new ram/storage paradigm and starts to use the latest and greatest hardware more efficiently than it does now.

RAM by itself is cheap enough. But getting the platform to allow your O/S greater than 128GB of RAM or so isn't that cheap or reliable.

 

See:
http://www.tomshardware.com/news/intel-900p-optane-ssd-computex,34558.html

Just a tidbit of what's to come, but still exciting nevertheless.

(Seems all we're getting is bits and pieces of info from anyone lately).

Above 2 quotes from:

See:
http://www.tomshardware.com/picturestory/764-best-of-computex-2017.html#s14


In summary; with just the 32GB Intel Optane Memory caching a HDD, the real world performance goes from ~18.87 seconds to load a particular game from ~46.73 seconds - a nice ~2.48x increase in responsiveness that anyone could 'feel'. A comparison with a full SSD would have been nice too - but that has already been shown to be inferior to an Optane + HDD combo.

With an 900p full of 1.5TB of Optane 'nand' - I am getting excited to see where real world performance will reach.

For most people; SSD's have increased boot times and how responsive the O/S is. Optane is set to actually increase performance (productivity...) of all programs used.

Santa! I've been good this year so far!