Intel Optane Memory

[bigboxes]

I've read a number of reviews on the new tech. So, if I'm reading this correctly, you need a motherboard that supports this. You then need to install software from Intel to configure it. Then it's just basically used to cache your data. Am I missing anything?

 

[deustroop]

I've read a number of reviews on the new tech. So, if I'm reading this correctly, you need a motherboard that supports this. You then need to install software from Intel to configure it. Then it's just basically used to cache your data. Am I missing anything?

Yes, for now, Tom's Hardware review mentions future developments:

3D XPoint promises to alter the memory landscape because it offers unique benefits, such as exponentially higher performance during light workloads. That should provide an explosive performance gain for desktop computers. Using 3D XPoint as a supplemental layer of DRAM is already part of Intel's data center strategy, and bringing that same functionality to the desktop could reduce the amount of DRAM you need for your system. It could also enable a radical new set of capabilities, such as eliminating the operating system boot process.

Those types of advancements could take years to surface because software always lags behind hardware, but 3D XPoint opens the door to those possibilities. For now, Intel's first 3D XPoint product comes in the form of an M.2 Optane storage device designed to speed up HDDs.

 

[bigboxes]

Yes, for now, Tom's Hardware review mentions future developments:

3D XPoint promises to alter the memory landscape because it offers unique benefits, such as exponentially higher performance during light workloads. That should provide an explosive performance gain for desktop computers. Using 3D XPoint as a supplemental layer of DRAM is already part of Intel's data center strategy, and bringing that same functionality to the desktop could reduce the amount of DRAM you need for your system. It could also enable a radical new set of capabilities, such as eliminating the operating system boot process.

Those types of advancements could take years to surface because software always lags behind hardware, but 3D XPoint opens the door to those possibilities. For now, Intel's first 3D XPoint product comes in the form of an M.2 Optane storage device designed to speed up HDDs.

So, it's a cool technology, but it's going to take a bit before it benefits us power users. I use an SSD in my file server as the boot os and it just makes the access to the drives (HDDs) across the network so much more responsive and snappy. Looking forward to its adoption.

 

[deustroop]

Just saw a note about another product from Intel:

"Intel® Optane™ SSD DC P4800X Series is the most responsive data center SSD. Built with the revolutionary new 3D XPoint™ memory media, the SSD DC P4800X is the first product to combine the attributes of memory and storage."

http://www.intel.com/content/www/us...tane-solid-state-drives-dc-p4800x-series.html

I have also seen the cost of the 375GB model estimated at $1520. A server product ?

 

[BonzaiDuck]

In 2011, makers released boards with the Z68 chipset which offered the ISRT feature. It had lukewarm reception: purists wanted "real-time" speed from "real hardware" speed improvements. Others, like myself, wanted a "cheat" alternative for performance to avoid high SSD prices other than the going-price for an SSD of 32GB to 64GB.

But the everlasting theme has been one of eliminating bottlenecks in the pyramidal conception of computer storage in general -- with CPU registers and caches at the top, electro-mechanical devices at the bottom, trading off between price-per-MB, speed, and size (dictated by price).

So? Suddenly, there is "Optane" with a new 3D-XPoint storage technology, poised to do the same thing some of us continue to do with combinations of RAM, SATA and NVMe SSDs. How much do you want to spend on RAM, and how much RAM supports these bottleneck-elimination desires? How much do you want to spend on NVMe? And in my case -- having become associated with the Romex Software product -- how much do you want to spend on software, or how much do you want to rely on it?

In the meantime. A knot of folks -- two or three including me -- posted at the Romex Primo-Cache forums, with common symptoms and difficulties emerging through the Build 1703 ordeal. The problem -- per se -- wasn't Primo: it seems to be a handful of factors related to NVMe-M.2 drives, their drivers, their proprietary software, Primo and Build 1703. Of those, I had the most difficult hurdle with a Win7/Win10 dual-boot configuration I wanted to save.

I got through it. I have "been to the mountain." I just wish I could solve the problems of others.

Works great, without Optane -- but with NVMe.

By the way. Can you use an Optane device with a Z170 chipset? Somebody must know already, to save me some poking around.

 

[ksec]

I Still could not could grasp the advantage of it on consumer's land. ( Its benefits on Server is very substantial )

Any "HOT" data will be cached on DRAM, I wish someone would test the difference in performance when you have 8GB / 16 GB Memory vs 4GB DRAM + 16GB Optane / 8GB DRAM + 32GB Optane. Both with the fastest possible SSD OR a slow TLC SSD.

And idling at 1W doesn't sit well with me, its power usage is also not that impressive.

 

[BonzaiDuck]

I was going to post some Anvil Bench results for my various-configured Skylake Z170 system with 16GB RAM and caching of HDDs and SATA SSDs to an NVMe volume of my boot drive. To be more precise, the NVMe is itself cached to RAM. But ksec is looking for something more rigorously analytical and comparative. Add my name to the list: I'd like to see that, too . . .

There are several different possibilities for caching that yield you "good performance" feelings as well as bench results. Some caching possibilities at the moment don't give any indication of bench results, because they occur in a stealth mode with certain software at times the system is idle -- not during a bench test. Yet the caching to SSD (NVMe) with no RAM-caching for an HDD obviously works well -- you can tell from the hit-rate.

I have my own thoughts about this. The gains may not be as great as you would see for a server, and they vary widely by user profile. I crossed paths with one guy who analyzed HDDs forensically in criminal investigations. He was using a 1TB 960 Pro solely to cache drives with files that were 500GB in size. That was my understanding of it. It's hard to wrap your brain around something like that. He also obviously NEEDED to have 128GB of configured RAM.

Now . . . about 3D XPoint and Optane. I haven't devoted the time and attention to sources I could read about it to catch up. Is the technology for Optane only feasible for Z270 or Kaby Lake generation motherboards? What about Z170? Suppose it doesn't matter for a built-in Intel "SRT" feature, like they introduced back in 2011 with Z68? If you put the Optane "drive" in the "right" socket, would it show its true capability? Because -- you can use any SATA or NVMe drive for caching.

 

[dlerious]

Linus has a video. https://www.youtube.com/watch?v=xgTYSsaNU_A . He says z270 and Kaby Lake or better. I think he makes a good point about people buying that, probably already have a boot SSD - Intel is marketing it as boost for spinners.

 

[BonzaiDuck]

You mean . . . you can't avail of the advantages with a Z170 of recent BIOS and a Kaby Lake processor? Tell me it isn't so!

 

[dlerious]

Apparently, that's a negative. See http://www.intel.com/content/www/us/en/architecture-and-technology/optane-memory.html for requirements.

 

[bigboxes]

So, does Optane memory improve performance with systems that already have an SSD as a boot drive?

 

[BonzaiDuck]

So, does Optane memory improve performance with systems that already have an SSD as a boot drive?

From the Intel spec pages posted by dlerious in his post # 10, a qualified "yes." But unless the NVMe M.2 performance of Optane exceeds that of non-3D-Xpoint NVMe drives by a significant amount, what you can do with Optane and the Z270 (or other same-gen) chipset and the ISRT feature built into it should not offer much more advantage than my use of a 960 Pro caching volume and the PrimoCache or similar software. I can accelerate either an SATA SSD or an HDD. In benchmark terms, the same configuration for accelerating an HDD with NVMe is much more stunning than it would be accelerating an SATA SSD.

I'm speculating based on dated experience with ISRT and benchmarks of the past, but if you could expect either configuration to offer speed improvement around 80% +/- of the faster caching volume ( or Optane in our topic at issue), the HDD acceleration is relatively more stunning. No doubt, an accelerated SATA SSD will still show even better improvement.

And like the criticisms imply in our forums I've seen about caching strategies, it depends on your usage pattern and YMMV.

I haven't read up on Optane with any voracious intensity so far. But it's an NVMe device. Period. The highlighted factor is instead the 3D Xpoint technology. And what they have in mind for the mere 32GB size of the device is otherwise no different than using caching software with an NVMe volume at the top of the persistent storage slice in the pyramid.

I think they're touting other things, like improvements in boot time. Unless you cold-boot your system at least a few times a day or several, that's not the essential improvement. And you could call it "memory," but as the Intel promotional pages insinuate, it is fast but persistent storage -- not volatile RAM.

I wager that with my own Z170 rig and its storage strategy, I'm only missing the advantage of 3D Xpoint and I still reap a good part of the performance improvement with Sammy Pro NVMe.

 

[Flapdrol1337]

In real world tests like booting or loading an application regular ssd's perform pretty much identical to nvme ones, seems the bottleneck is elsewhere. I doubt this will do much if you already use an ssd.

But could this be useful on systems that don't have a lot of ram? Like maybe make an 8GB system perform like a 16GB one in tasks that want lots of it?

 

[dsplover]

Boards with multiple M2 slots where 3 x 32GB Optane devices, each with their own 4TB 7200 HDDs sounds like a nifty way to get premium storage at a consumer price.
Is this possible or are you limited to one device?

I'm building a Thunderbolt 3 Hackintosh.
Would love to fill all 3 x M.2 slots with 32GB Optanes attached to WD Black 4TB HDDs.

Thanks for any help.
I grew weary after reading a few articles, none of which mentioned multiple device use.

 

[BonzaiDuck]

Boards with multiple M2 slots where 3 x 32GB Optane devices, each with their own 4TB 7200 HDDs sounds like a nifty way to get premium storage at a consumer price.
Is this possible or are you limited to one device?

I'm building a Thunderbolt 3 Hackintosh.
Would love to fill all 3 x M.2 slots with 32GB Optanes attached to WD Black 4TB HDDs.

Thanks for any help.
I grew weary after reading a few articles, none of which mentioned multiple device use.

That answer to that question I wouldn't know for sure, but the original ISRT feature only provided for a single pairing of an SSD and HDD. The agnostic software solution I use is more flexible than that. Also, I'm thinking I could simply fold Optane into my own solution and circumvent using the Intel chipset features. And -- yes -- at this point, I'm wondering if somehow Optane acceleration doesn't again require a BIOS storage setting of RAID mode.

As I choose, I can create three caching volumes on a 250GB NVMe M.2 and assign them to different slower devices -- either SATA SSD or HDD. I can cache across AHCI and RAID-mode devices. No less, I could create a single 100GB or greater caching volume on that M.2 and cache several disks into the same cache space. One user who does forensic hard disk analysis in criminal investigations used an entire 1TB 960 Pro with 128GB of RAM to cache files of 500GB each.

With the Intel approach, once set up, you may not need to give it any attention. In my case, working with the Romex software complicated by Windows upgrades has been a learning process unto itself. I have described it as a Swiss Army knife of caching options. But as I guessed beforehand, you see people who went all goo-gah about the various aspects of it in piling up its application, ignoring simplicity and common sense.

For instance, why would you want to cache an NVMe system-volume to a caching volume on the same disk? Or -- why would you bother caching an HDD to RAM if it's already cached to NVMe? It's sort of like a story of someone who visited a smorgasbord -- Eric Cartman at Casa Bonita in the South Park episode. They want to fill their plate with everything, and then spend the evening with a bottle of Pepto-Bismol or an hour on that special seat in the small, warm room.

If your time to purchase new hardware in your personal cycle of PC turnover coincides with the new improved technology, I say go for it -- get the motherboard and the Optane drive/device. I would still investigate the possibility of more flexible alternatives to Intel's patent configuration options. PUt it another way. With the old ISRT acceleration feature, you had to have RAID mode. So far, so good. Then along came Samsung with its SSD "RAPID" feature. For that, you had to have AHCI mode. So you couldn't use both technologies at once unless you used a second HDD controller. With what I have now for a mere $30 per PC, I can do it all and all-at-once with a single piece of software. I'm just not going to stick Butters in a fallout shelter so I can go to Casa Bonita and make myself sick.

Just as afterthought, and only with my prior experience using ISRT. It doesn't as much matter that Intel builds their solution into the chipset firmware as opposed to using hardware-agnostic software. You will still have instances where maintenance is required. Even some sporadic hardware failure -- a power outage -- may trigger the equivalent of CHKDSK repair. You would see that happen with ISRT, as the firmware would run a check and fix operation when something was out of kilter.

That's why some among us favor simply fast hardware without bridging gaps between storage levels -- it's going to be simpler, more reliable with something less in maintenance. I chose to go my own way, because I thought I could get to a point where I'm not spending time "fixing" stuff. I may have reached that point, actually. But with Windows 10 ability to check mounted disks including system volumes, I use it frequently just to be sure nothing "hinky" is going on with storage. And -- so far, so good.

 

[dlerious]

The specs say the interface is PCIe NVMe 3.0 x2 . http://www.intel.com/content/www/us...orage/optane-memory/optane-32gb-m-2-80mm.html
My 960 Evo looks like it has faster read and much faster write speeds. The Optane has lower latency.

Watching Linus video, it appears there's a software component as well (for caching). Optane sounds like it would be a good choice for older systems with spinners, but I can't see much use for a newer system that's more likely to have SSD and/or NVMe.

 

[bigboxes]

Sounds like the improvements would be hard to detect for SSD users. More like it would only be noticeable in benchmark tests. Doesn't mean I'm ruling anything out for the future. That Intel page list a test platform with a HDD, and not an SSD, as the base if I'm not mistaken. It's kind of hard to read since it's really more about marketing and less about tech review.

 

[BonzaiDuck]

Well, if you want a 2TB SATA SSD, count on paying ~$550 for an MX300. If you want a 2TB 960 Pro, I think you'll shell out $1,200. But you can get a 3TB HDD for less than $100.

Suppose you already have a 512GB or 1TB 960 Pro or EVO. If you can live with 400GB of a system volume for the smaller unit, you have something around 100GB left for caching. $30 for software (and there's more than one alternative to the one I favor), and $100 for the HDD. $130 to make that storage seem a lot faster. But if it seems faster -- what does that make it, really?

That's going to be a difference you can "detect." Suppose, instead of the HDD you get a 1TB SSD unit for $250 -- Not a Sammy, maybe a Crucial or ADATA. And, you cache it, too, to the NVMe 100GB volume. Howsoever you cache it to NVMe, you won't notice any difference between the HDD and the SATA SSD except for the write speeds. If you dare yourself, you can fix that too.

It's a long-established strategy going way-back-when, but your IT veterans think it makes more sense for server systems. While I don't like weighting my midtower with a handful of 3.5" drives, I not only want fast essential (system-OS) storage, but I want cheap, high-volume storage. I'm using 3x 2TB 2.5" "laptop" spinners in my desktop with the 5,400rpm spec, in addition to my NVMe drive. Do I care if they have the spec lower than 7,200? No. And anything between 120MB/s and 300MB/s of electromechanical storage can be lifted in performance high enough that it doesn't make a difference whether they're the fastest VelociRaptors around. As for those lappie drives, I cache one of them to the NVMe, a second one to non-persistent RAM-cache, and the third is just a backup drive for daily Macrium images.

But in all of this, you also have folks choosing between 2x4 8GB, 2x8 16GB and 2x16 32GB RAM kits. I find that 16GB gives me a comfort-margin, but it's not frequently utilized -- it's just "available." A lot of folks get by with 8GB and there are plenty of lappie and even desktop systems out there with 4GB. I can spare AT LEAST 4GB for RAM-caching. What could I do with a 32GB kit? It could almost make the Optane seem inconsequential, but it's not PERSISTENT cache, except for the software's feature to save the cache on restart. There -- you have to wait a little bit at boot time, except for the speed of your boot disk. Then, you would fret over the number of writes to your NVMe boot drive.

It has become such an interesting proposition and question for me that I've had my eye on a 32 GB pair of 3200 14-14-14 G.SKILLs and the 2TB MX300 for about 4 months now. I can't decide what to do with $1000 sitting in my budget and account to spend. And the performance of the 960 NVMe together with the 2TB Barracuda and a 16GB kit of RAM totally dispels any frustration I have about it. Those products will be available even if I wait through the whole year.

 

[dlerious]

Unfortunately, prices are going in the wrong direction right now for SSD. My 2TB MX300 was $516 four months ago. I think I paid $400-$410 for my 1TB 960 Evo. Afraid to look at RAM, I'd like to grab some 3600 or faster, but don't know what to expect on AM4, pretty sure my 2x16GB FlareX 2400 will work.

 

[BonzaiDuck]

. . . and meanwhile . . . Seagate is producing 2.5" 3TB, 4TB and 5TB Barracudas. Don't like the 2.5"? They certainly come in the established 3.5" flavor and weight. Of course, the bigger drives increase my last estimate by between $50 and $100.

 

[dullard]

You mean . . . you can't avail of the advantages with a Z170 of recent BIOS and a Kaby Lake processor? Tell me it isn't so!

From what I can tell, this is currently a software limitation. The hardware has not yet been tested (or maybe it failed the tests) on those motherboards. So, the Intel software prevents Optane from being used on a Z170 motherboard. Since it is a software limitation, that could change. But I wouldn't count on it.

 

[dullard]

Sounds like the improvements would be hard to detect for SSD users. More like it would only be noticeable in benchmark tests. Doesn't mean I'm ruling anything out for the future. That Intel page list a test platform with a HDD, and not an SSD, as the base if I'm not mistaken. It's kind of hard to read since it's really more about marketing and less about tech review.

Intel has produced a high-end server Optane drive and a low-end Optane Memory cache. They have not yet released one for the enthusiast level. So, if you already have a fast M.2 SSD (or similar) it probably won't do too much for you. The very few benchmarks that I have seen pairing the Optane Memory cache with a fast SSD is often on the order of a few percent faster in real-world applications.

However, if you are like me and have a SATA SSD, then Optane Memory can help a lot. I see no reason to pay hundreds of dollars to ditch my current SSD for a faster SSD, but the SATA connection is holding it back. The current Optane Memory cache should help that situation.

From Intel, http://www.intel.com/content/www/us/en/architecture-and-technology/optane-memory-faq.html:

Yes. Intel® Optane™ memory can be used to accelerate any type of SATA-based storage media, including SATA SSDs. However, the performance benefit of adding Intel® Optane™ memory will be greater on slower storage devices like an HDD versus a relatively faster storage device like a SATA-SSD. Additional performance data is forthcoming.

 

[BonzaiDuck]

I remember my experience with ISRT using the IRST drivers and software. You could only cache a single device to SSD. You had to use RAID-mode for the entire Intel controller. Every so often -- possibly resulting from an unintended system reset, BSOD, power-outage -- you would see ISRT "do its thing" at boot-time that reminded of CHKDSK.

This is storage-mode agnostic, capable of caching across controllers, and caching multiple storage drives even including RAID arrays, with multiple caching "tasks." You don't see problems any greater than with ISRT. The worst that's happened so far in 2.5 years of my experience with it occurred this month with the Creators Update. A common symptom after upgrading a system already configured with the caching software occurs at boot-time before the pre-logon screen appears in Win 10: "Diagnosing, scanning, repairing {GUID} /volume . . " The Windows upgrade gets confused about the non-NTFS SSD caching volume on an NVMe, and even volumes on an SATA SSD. I fixed it completely by reinstalling the Sammy NVME driver and Magician 5.0. Others weren't so lucky, and Romex is working on a patch. But the problem wasn't Romex -- it was Windows. There are some other things you learn not to do, like running memory-grabbing stress-tests with active RAM-caches. But the patch will be forthcoming, and addresses -- not RAM-caching, but SSD caching.
SuperSpeed SuperCache is one other option, but it only does RAM-caching.

With the other, I feel good about the lifetime ~$30 license. If it weren't for April 11, there would be nothing to note about it except features or glitches common to ISRT. I would probably expect the same thing with Optane caching. And I'd stop using it if it weren't working properly.

But all of those solutions, including Intel, depend on a software component.

 

[dsplover]

Thanks Bonzai.
I already use apps that store RAM buffers, 64k in size.
When I press a key on an 88 note MIDI Controller, the seek gets bounced from RAM to the NVMe/SSD and streams that data until I release the key. So after reading some docs Optane probably wouldn't be useful.
There's 100s of 1000s of these, the entire template is 18GBs.

I do see how I can use a 32GB Optane with a 4TB HDD for OS+Apps.
Think I'll go with that.
Cheaper than a 2TB Samsung 850 Pro, twice as big....

 

[BonzaiDuck]

Thanks Bonzai.
I already use apps that store RAM buffers, 64k in size.
When I press a key on an 88 note MIDI Controller, the seek gets bounced from RAM to the NVMe/SSD and streams that data until I release the key. So after reading some docs Optane probably wouldn't be useful.
There's 100s of 1000s of these, the entire template is 18GBs.

I do see how I can use a 32GB Optane with a 4TB HDD for OS+Apps.
Think I'll go with that.
Cheaper than a 2TB Samsung 850 Pro, twice as big....

That would work . . . I'd just want a bigger caching drive volume out of techno-lust. So I bought the 960 Pro 1TB which contains dual-boot OS volumes plus a 120GB caching volume for a 2TB Barracuda. And I want to buy MORE stuff, with insane ideas about further improving performance when the performance doesn't matter so much for personal use.

I'd still use either an SATA or NVMe SSD for the OS. You could put more apps on a cached spinner or a cached SATA SSD.

But let's look back at things concluded last fall as everyone had an itch for the Samsung releases.

In some ways, you notice the difference. With my stock-car racing simulators, there are times when it gets almost scary.

But for the money, any practical mainstreamer might stop after buying an SATA SSD with decent capacity. In a way, I spent extra just to prove to myself that a high-capacity HDD could be integrated with NVMe in ways that make you forget HDDs are slugs.

When I build these systems, they are general-purpose. These days, it seems like tinkering with systems for systems' sake . . .