Discussion Optane Client product current and future

[IntelUser2000]

Optane SSD 905P specs.

Sequential Read/Write: 2600/2200MB/s
Random IOPS up to: 575/550K

Slight improvement in reads and 10% improvement in writes.

Power usage specs are different between U.2 480GB and HHHL 960GB. Compared to 900P, idle power is a nice reduction to 3.3W from 5W, and slightly higher at 6W for 960GB. Load power is slightly lower for the 480GB compared to 900P but noticeably higher for 960GB.

SSD 900P: https://www.mouser.com/pdfdocs/Optane900PSeriesDatasheet.pdf
SSD 905P: https://digitallibrary.intel.com/co...9.3XaeoqV3A--yAZDsbYXgyI5O6ECn0QASt7Mz05Yy_vI

Small, but noticeable update. Also updated the first two posts.

 

[Dayman1225]

Intel Optane SSD 905P Coming Soon

The AnandTech article does a direct comparison of the 905P and 900P and of course thinks they have moved to the smaller controller, however interestingly they stated that Intel has indicated that M10 Optane Drives review samples will be coming soon.

The Optane Memory M10 is already listed by at least one third-party seller on both Amazon and Newegg, and Intel has indicated they will be providing review samples soon.

 

[Dayman1225]

Intel Optane SSD 905P 960GB NVMe HHHL SSD Review - Bigger XPoint - PCPer

Looks like decent gains all around, drives were received like 24HRs ago!

 

[Brahmzy]

Very, very small improvements over the 900p. Certainly completely imperceptible to the user.

 

[IntelUser2000]

Tomshardware preview is showing better gains. It does gain in areas where it was behind competition, which is sequential.

So for the 905P, the ARK page still calls it Mansion Beach, not Mansion Beach Refresh. It shouldn't matter, because it has all the signs for a refresh part, slightly faster speed, and available in more form factors and capacities.

By the way, I think the Optane Memory devices may be a great pair with dirt cheap SSDs. I've got a Silicon Power Slim S60 TLC SSD with SLC cache, and very basic scenarios slow it down. They should enable caching to work with NVMe SSDs, because we have really low end devices coming out now. QLC generation is going to make new records in low performance and price.

 

[Brahmzy]

Looks like Toms will have a full review later in the week or so...

 

[IntelUser2000]

ARK page shows a 280GB U.2 version also exists. Same performance specifications as the bigger brothers. Interestingly, the 905P drives have 20x greater shock tolerance(50G to 1000G). I wonder why that is, and whether there was a reason for doing so. Maybe it was being used in embedded, industrial applications quite often?

I have to conclude these drives are indeed Mansion Beach Refresh, without the Refresh moniker.

 

[cbn]

Specifically does the Optane Software handling the 32GB load the page file onto the Optane? If so, I am thinking there could be substantial difference between the two for a system used for browsing if the primary drive (being cached) is HDD.

There probably isn't a big difference in browsing. I'm using the 16GB so I can vouch for that. What 32GB is good for is scenarios where data swapping can occur because there's lot of activity going on. The 16GB doesn't know the difference so it just keeps the most recent data. The 32GB can optimize over time. I can tell you on my system when I need to reinstall drivers or reboot the OS after being on for many days it slows down for some time.

Do you page out often when browsing?

If the page file is not already allocated on the Optane I am concerned the bottleneck would be the HDD (If the 16GB Optane is already full).

 

[IntelUser2000]

Do you page out often when browsing?

If the page file is not already allocated on the Optane I am concerned the bottleneck would be the HDD (If the 16GB Optane is already full).

I don't believe so. I have a pretty predictable usage pattern. My system also has 8GB of RAM. This is why I believe I'll benefit from the 32GB model, since it tries to learn, rather than just flushing the older data out.

Update: I think I know why. The mining application I run uses 11GB for commit. The commit charge is nearly 15GB. Since the actual size of Optane Memory is listed as 13.4GB, I wonder if it means it only has 2.4GB space leftover to do everything else. I figure Browsing should use no more than 1.5GB in most cases. AMD video driver is 850MB. Surprizing facts says while 16GB's Block caching can recognize different types of files, it doesn't have an algorithm to recognize which are most commonly used.

32GB version has usable capacity of nearly 27GB, meaning leftover space would increase 7x to 16GB, and the added file caching has an algorithm which optimizes it over time.

 

[thecoolnessrune]

Doesn't look like the 905P is going to restore the Power Loss Protection support that the 900P had, and then didn't.

 

[cbn]

I think we will eventually see a delineation between Optane for HDD and Optane for SSD.

They already do this BTW but it is undocumented. The 900P is the perfect Optane cache for a large SATA SSD since it caches far more and is far faster than the 16/32GB offerings targeted towards HDDs.

Can you elaborate on the delineation between Optane for HDD and Optane for SSD?

 

[IntelUser2000]

Doesn't look like the 905P is going to restore the Power Loss Protection support that the 900P had, and then didn't.

It doesn't need Power Loss Protection capacitors because DRAM buffers aren't needed.

 

[nosirrahx]

Can you elaborate on the delineation between Optane for HDD and Optane for SSD?

I think the implication is that the Optane drives designed to be used as cache are slower than the 900P drives. As a result a SATA SSD paired with an 16/32GB Optane drive is not a lot better than a NVME M.2 SSD. A SATA SSD paired with a 900P on the other hand give some seriously great performance.

Intel does not support this even though it does work.

 

[cbn]

Below is quote from post #123:

There's also the question of how low power the M10s are. The 800P review on TH shows its acceptable, but nowhere near Chromebook level. The top level NVMe drives are behind SATA drives in power consumption, and 800P is middling in the field of NVMe drives.

https://www.tomshardware.com/reviews/intel-optane-ssd-800p,5497-2.html

Looking back on post #61 and #70, I wonder if the 16GB M10 (if used in a Chromebook or Ubuntu "Browser Build" Laptop) could be tuned for even lower power consumption while retaining enough performance to be used for a paging out during browsing? Is there such a thing as combining a PS2 read with a PS0 write?

 

[thecoolnessrune]

It doesn't need Power Loss Protection capacitors because DRAM buffers aren't needed.

Capacitors are a large part of Power Loss Data Protection on DRAM based SSDs, yes, but it's hardly the only component. The controller and data paths themselves still have to be able to flush in-flight data before loss of power in a predictable and reliable manner. You'll note that the highest end Optane, the P4800X, still has this certification. The Optane 900P originally had this certification, but it was changed on Intel's site roughly a month after release to not having Enhanced Power Loss Data Protection.

 

[Brahmzy]

Intel does not classify the 900p as an 'Enterprise' SSD, but an Enthusiast SSD. This is one of many things that didn't make the cut.

 

[IntelUser2000]

Capacitors are a large part of Power Loss Data Protection on DRAM based SSDs, yes, but it's hardly the only component. The controller and data paths themselves still have to be able to flush in-flight data before loss of power in a predictable and reliable manner. You'll note that the highest end Optane, the P4800X, still has this certification. The Optane 900P originally had this certification, but it was changed on Intel's site roughly a month after release to not having Enhanced Power Loss Data Protection.

There's a post on SLOG forums(I think?), that says it likely has PLP, but not Enhanced PLP which becomes a necessary thing in a datacenter environment.

I think this would be a topic worth investigating into.

 

[thecoolnessrune]

There's a post on SLOG forums(I think?), that says it likely has PLP, but not Enhanced PLP which becomes a necessary thing in a datacenter environment.

I think this would be a topic worth investigating into.

Agreed. Lots of reviews still say the drive has Enhanced PLP because the drive had that certification on Intel's site when reviews were hitting websites and first shipments were getting in consumer's hands. I'd say it's safe to say pretty much any XPoint or similar technology has "PLP" covered, because of the lack of DRAM and the "frozen" state XPoint Memory remains in on power loss. That stated, end-to-end power loss protection has to include both the hardware and the firmware components to make sure all buffers are reliably flushed before power is lost in the controller.

Personally, I think this goes back to something like the Intel 730 SSD PLP situation several years ago. That drive was built just like an S3700, including the capacitor bank, but it didn't come with PLP Certification. It was believed that was due to the 730's Controller running at a higher clock speed, and thus consuming more power than the S3700, to the point where the design couldn't be validated to necessarily guarantee PLP under all scenarios. There's probably something just different enough about the 900P compared to the P4800X even though they're the same build, to not be able to guarantee end to end PLP functionality in all scenarios. Could be clock speed, firmware, signaling, die quality. All sorts of things really. But it was enough to have Intel retroactively remove validation for ePLP from the 900P.

 

[cbn]

They should enable caching to work with NVMe SSDs, because we have really low end devices coming out now. QLC generation is going to make new records in low performance and price.

Optane memory can cache NVMe NAND SSDs, but it is not officially supported:

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

We present the Seagate BarraCuda Pro 10TB both as a single HDD and combined with the Optane Memory SSD (Optane + HDD). This is the only officially supported Optane Memory configuration with the preloaded files in place (and any other optimizations that Intel didn’t disclose). An Intel 600p 1TB SSD also appears both as a single drive and combined with the Optane Memory drive acting as a cache (Optane + NVMe SSD). This is not an officially supported configuration, but it is possible with the Optane Memory SSD acting as a generic cache device.

(Some results found in the link above posted below. Very interesting that in Sequential read Optane Memory + HDD substantially beats Optane Memory + 600p NVMe SSD. Also for low QD Sequential write the Optane memory + HDD is faster than the Optane Memory + 600p NVMe SSD. For low QD 4K write the 600p NVMe SSD is faster than the Optane Memory +HDD which is faster than the Optane Memory + 600p NVMe SSD)

 

[IntelUser2000]

Optane memory can cache NVMe NAND SSDs, but it is not officially supported:

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

The Optane Memory + HDD has an advantage in, sequential read and write, and low queue depth read and write(though the low queue depth advantage is slight).

Tomshardware says you need to use SRT to make it work with NVMe drives. SRT is Intel's generic caching application. It's when you want to use Optane Memory application, it doesn't work with NVMe. Optane Memory application, according to Surprizing facts is the one that has the new caching schemes.

It looks like SRT doesn't work with Atom-based platforms, so no Geminilake. If they consider SRT a deprecated feature, to be replaced by Optane Memory, they should enable it on their low power platforms too. I guess I can use 3rd party ones. I might just do this when I get my J5005.

 

[nosirrahx]

It would be cool if Optane had something like NVMe support mode where only files that would be read faster from Optane would be cached.

 

[cbn]

For a Linux mobile device meant for Browsing (Firefox, etc.), Streaming (Youtube, Steam, etc.) and Cloud (Microsoft Office Online, LibreOffice Online etc.) I've been wondering if 2GB RAM and 16GB M10 Optane might be better than 4GB RAM and 32GB eMMC or 64GB eMMC.

eMMC can be pretty slow*, and if local compute is not really being done to a large extent perhaps Optane + a smaller amount of RAM is a better idea overall (most of the time)?

*Particularly the eMMC devices listed after the first one (in the linked thread).

P.S. With eMMC I could imagine 2GB RAM + Optane allowing more browser tabs to be comfortably used than a system with 4GB RAM and eMMC. (This provided the swap partition for the Optane system is large enough and better yet if the swappiness is set higher than default).

Swappiness is a Linux kernel parameter that controls the relative weight given to swapping out of runtime memory, as opposed to dropping pages from the system page cache. Swappiness can be set to values between 0 and 100 inclusive. A low value causes the kernel to avoid swapping; a higher value causes the kernel to try to use swap space. The default value is 60; setting it higher will increase performance of "hot" processes at the cost of making a return to inactive "cold" ones take a long pause, while setting it lower (even 0) may decrease response latency. Systems with more than adequate RAM for any expected task may want to drastically lower the setting.

Though to be honest I do admit is a very low configuration and I am wondering about the power consumption difference between 2GB DDR4 + 16GB M10 Optane and 4GB DDR4 + 32GB eMMC (or 64GB eMMC).

 

[biostud]

Anyone tried using primocache with a 900p?

https://www.romexsoftware.com/en-us/primo-cache/

 

[nosirrahx]

Anyone tried using primocache with a 900p?

Yes and it technically works but you can't really feel it most of the time. I tried it out on my son's high end gaming system and after a few months the cache hit rate is only around 15%.

On his system the caching is actually a step beyond that. It caches a 4TB SATA SSD to a 280GB 900P and then to 12GB of Primo cache.

Turning on the Optane cache was instantly noticeable, adding Primo on top helped a little more but the jump was nothing like adding Optane.

 

[cbn]

Thinking about how eMMC is so popular in systems used with low power SoCs.....I wonder how low Intel could get Optane power consumption when used in a DDR interface? This while still retaining good 4K read and write?

P.S. Something interesting to think about regarding lowering power consumption would be how much this might increase DWPD in a client situation?