Optane SSDs available at Newegg - $2 per GB!

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

nosirrahx

Senior member
Mar 24, 2018
255
24
61
#51
Tiered caching, it works, not anything more complicated than that.

With RAM prices its not really cost effective but on a high end build its just the performance that matters.

The system I built has 12GB of RAM cache, 260GB of Optane cache and 4TB of SATA SSD. I also use a UPS so I can do deferred writes.

The system presents as a single drive that is both huge and astonishingly fast in all use cases, booting, installing and loading games.

Check post #5 over here, RAM cache alone wont do this, not even close. This is the power of Optane:

https://forums.guru3d.com/threads/tried-something-interesting-with-a-u-2-900p-and-it-worked.419655/
 
Jun 30, 2004
13,804
285
126
#52
Mmmm . . . . Can anybody say when the Optane Memory motherboard features were added? Certainly, the Z170 systems would not have them. I don't see any. What about the Z270 chipset? I see the Z370 / Coffee Lake specs showing Optane Memory.

But I had speculated that I could use a 900P for a caching volume with the Primo software. Or -- I could find a Pro or EVO of maybe 500GB, and get similar results to what they found in "the best undocumented feature . . . . ever come across."

The 4K random performance of the Intel drives is more significant for exclusively SSD-caching. Throwing RAM into the equation boosts those scores. While mine may seem a bit weak or second-rate compared to the benchies others have shown here, it's nothing to sneer at; it's in the same magnitude of scores.

Now I have to come up to speed about these massive SSDs that are mentioned. You'd do about as well to find a 4TB HDD and cache that to SSD and RAM.

I keep a lot of business and media files on my server, so I don't need drive capacity of that much on my Skylake. Truth be told, I already have too much -- just for my NVME and SATA SSDs.

The Optane system with a 900P 500GB caching volume -- or even a comparable 250GB -- may outperform my system to some degree, but I don't think it's going to be that much of a difference with RAM in the mix. Without the RAM, that's another story.

Interesting to see folks go goo-gah over these configurations, though. I suppose if I wanted to double my cache size, I could investigate a 900P. Right now, I like looking at my Quicken balances . . . .

[LATER] AnandTech did a review of the 900P drives that was very revealing. It does more to verify the CrystalDiskMark scores at the Guru 3D forum. The 4K bench results are stunning, and just blow away the 960 Pro and EVO results. It just blows them away.

Replacing my EVO caching drive with a 250GB 900P would make using RAM-caching seem . . . almost unnecessary if one were already doing it.

But like I said . . . my Quicken balances . . . . I should be satisfied for now. It's too much techno-lust obsession . . .
 
Last edited:

nosirrahx

Senior member
Mar 24, 2018
255
24
61
#53
A lot of this is use case dependent. On the system I built with the tiered cache setup the Primo hit rate fluctuates between 15% and 35% so clearly RAM cache alone is not enough to avoid hitting the SATA SSD.

If you use Primo and the hit rate is near 100% then your use case fits well with how RAM cache works. You only really miss out on blazing boots and you have to wait a bit after boot for the RAM cache to fully populate.

The other main issue with today's prices is that installing enough RAM to make RAM cache worth it so is so expensive that you might as well just buy a good SSD. The only reason I want the way I did on this system was intentional overkill.
 
Aug 4, 2015
166
27
101
#54
Can anybody say when the Optane Memory motherboard features were added?
All of the hardware features necessary for Optane Memory caching and NVMe RAID are present starting with Skylake-era 100-series chipsets. Firmware support for booting from Optane Memory-cached volumes is only included on Kaby Lake-era 200-series motherboards.
 
Oct 14, 2003
6,257
357
126
#55
I would get a small Optane drive, like 128GB in size, well actually, the 118GB 800P would do it. Only if it was cheaper.

I'd split the drive into 3 portions.
-64GB for OS install
-32GB for paging
-32GB for caching with HDD.

Ideally the caching with HDD would be done with the Optane Memory application since its the fastest, but the software is made so simple it has no option other than turn on or off.

OS installation determines most of the responsiveness. Paging is for when it runs out of RAM. Caching for accelerating applications.
 

nosirrahx

Senior member
Mar 24, 2018
255
24
61
#56
I kind of wish they sold a hybrid M.2 SSD where 1 side was 1TB of NAND and the other was 64GB of Optane that cached the NAND.

An all in one solution like that could address capacity, 4KQ1T1 speed and ease of install at the same time.
 
Mar 27, 2009
12,968
36
106
#57
Here is a typical task manager screenshot for Browsing with Firefox using 2GB RAM, 800p, Windows 10 Pro:



Browsing performance is still very good at this level of page out....I attribute this to having enough free DRAM available for writes.

However, If I go quickly (ie, open tabs very fast) to 8.0 GB commit charge here is what happens to the available DRAM:



Performance at this level is very slow.

......but this does improve as the CPU settles down and some DRAM simultaneously gets freed up:

 
Last edited:
Mar 27, 2009
12,968
36
106
#58
Currently Testing with 2GB RAM, 2TB 2.5" Firecuda and the 58GB Optane 800p (as dedicated pagefile device).

Performance for browsing seems almost the same. I get a huge boost when I disable pagefile on the 2.5" Firecuda and move it (manual set to 15GB) to the Optane Drive.

.
 
Jun 30, 2004
13,804
285
126
#59
Currently Testing with 2GB RAM, 2TB 2.5" Firecuda and the 58GB Optane 800p (as dedicated pagefile device).

Performance for browsing seems almost the same. I get a huge boost when I disable pagefile on the 2.5" Firecuda and move it (manual set to 15GB) to the Optane Drive.

.
Well -- you could also put the pagefile on the Firecuda, and use the Optane to cache that HDD. The pagefile would be cached. If you put the pagefile exclusively on the Optane, then you can't accelerate the Firecuda.

How big of a pagefile set do you need?
 
Mar 27, 2009
12,968
36
106
#60
Well -- you could also put the pagefile on the Firecuda, and use the Optane to cache that HDD. The pagefile would be cached. If you put the pagefile exclusively on the Optane, then you can't accelerate the Firecuda.
At some point in the future I will do some testing with the Intel Optane caching software and compare to having a fixed size page file that is exclusively reserved and cannot be affected by other files.
 
Jun 30, 2004
13,804
285
126
#61
All of the hardware features necessary for Optane Memory caching and NVMe RAID are present starting with Skylake-era 100-series chipsets. Firmware support for booting from Optane Memory-cached volumes is only included on Kaby Lake-era 200-series motherboards.
Just revisiting Billy's post. "Firmware support." I'd be interested to find out if you couldn't use the same hardware and circumvent the limitations of the proprietary firmware. I'm still a bit confused by "Optane Memory," and I can more easily see what can be done with an Optane NVME device.
 
Mar 27, 2009
12,968
36
106
#62
Something else to consider for the 800p Optane SSD would the be the AMD Ryzen Fuzedrive software by Enmotus.

I only wonder how well it would work for paged out browsing with a hard drive? (Reason: when the page file expands other data on the 800p SSD would have to be relocated to the second fuzed drive....and if this is a hard drive the process would predictably be slow and rate limiting.)

With this noted, a SATA SSD could be used for the second drive in an Optane 800p fuzed volume.
 
Last edited:

IntelUser2000

Elite Member
Oct 14, 2003
6,257
357
126
#63
Then, what are those "special features" IntelUser2000 cites?
I am not aware of everything. Though reviews say it pre-caches white-listed OS files, so you get the benefit immediately.

Also the differences between RST and Optane Memory are quite clear: http://www.tomshardware.com/reviews/intel-optane-3d-xpoint-memory,5032-4.html

Optane+HDD is often better than Optane+600p. The former is using the Optane Memory application, while latter is using Optane Memory but using RST.

Most caching implementations operate at the block level, but Intel’s Optane Memory Technology is apparently aware of the file system. The software has a white list of frequently-accessed operating system files that it preloads into Optane Memory during installation, so you will notice accelerated performance as soon as you boot into Windows.
The block level vs file system difference is due to the way 3D XPoint works compared to NAND. Hence, Optane-specific optimizations.
 

nosirrahx

Senior member
Mar 24, 2018
255
24
61
#64
I am not aware of everything. Though reviews say it pre-caches white-listed OS files, so you get the benefit immediately.
This is absolutely true. On the Optane cache system I built the bootup seems faster than should even be possible and the OS drive is a SATA SSD. A SATA SSD is fast but nothing like what I actually observed. Check post #5 here, I posted a clip of it booting:

https://forums.guru3d.com/threads/tried-something-interesting-with-a-u-2-900p-and-it-worked.419655/
 

nosirrahx

Senior member
Mar 24, 2018
255
24
61
#65
The only annoying thing I have observed about Optane cache so far is that it makes the OS drive being cached invisible to any procedure that would update its firmware.

Its a lot like RAID in that respect. You see the Optane cached volume as its own entity, not the 2 drives that it consists of.
 
Aug 4, 2015
166
27
101
#66
The block level vs file system difference is due to the way 3D XPoint works compared to NAND. Hence, Optane-specific optimizations.
That's a pretty vague, hand-waving sort of explanation. Can you provide any description of the mechanism by which you think a NAND vs 3D XPoint media difference makes one more suitable for block-level caching and the other more suitable for file-level caching?
 

IntelUser2000

Elite Member
Oct 14, 2003
6,257
357
126
#67
That's a pretty vague, hand-waving sort of explanation. Can you provide any description of the mechanism by which you think a NAND vs 3D XPoint media difference makes one more suitable for block-level caching and the other more suitable for file-level caching?
So this, I was going off of Tomshardware's explanation. This site explains it better. http://surprizingfacts.com/what-is-intel-optane-part-1-optane-memory-blog-of-intel-surprizingfacts/

Initially I was thinking it could base caching off byte-level addressing(rather than block level with SSDs) to improve performance. This was probably a mistake because it still needs to be routed through NVMe.

on a 16GB device, only block level caching is supported, on a 32GB device, block level caching and file leveling (both work simultaneously)
This is interesting. So this is why they market the 32GB for enthusiast usage and 16GB for general users. I also get it from the reviews the 32GB has more information present on the Optane Memory application. I'd like to read a review that compares between the 16GB and the 32GB version.
 
Last edited:

nosirrahx

Senior member
Mar 24, 2018
255
24
61
#68
This is interesting. So this is why they market the 32GB for enthusiast usage and 16GB for general users. I also get it from the reviews the 32GB has more information present on the Optane Memory application. I'd like to read a review that compares between the 16GB and the 32GB version.
There is also a difference between the 16 and 32GB versions when it comes to the optimize task, only the 32GB version (and above) have this.
 

IntelUser2000

Elite Member
Oct 14, 2003
6,257
357
126
#69
There is also a difference between the 16 and 32GB versions when it comes to the optimize task, only the 32GB version (and above) have this.
Yea that's what I meant by the information. So there's a seperate button then?
 

nosirrahx

Senior member
Mar 24, 2018
255
24
61
#70
Yea that's what I meant by the information. So there's a seperate button then?
In scheduled tasks there is an Intel one that runs once a day to Optimize what I assume are files. In the Optane control panel there is a section that tells you when this last ran.

For the 16GB version I do not believe either of these exist.
 
Mar 27, 2009
12,968
36
106
#72
In scheduled tasks there is an Intel one that runs once a day to Optimize what I assume are files. In the Optane control panel there is a section that tells you when this last ran.

For the 16GB version I do not believe either of these exist.
Here is a screen shot of that:

 
Mar 27, 2009
12,968
36
106
#73
This is interesting. So this is why they market the 32GB for enthusiast usage and 16GB for general users. I also get it from the reviews the 32GB has more information present on the Optane Memory application. I'd like to read a review that compares between the 16GB and the 32GB version.
The Intel forums confirm the same thing:

https://communities.intel.com/thread/120086

We would like to inform you that this option is only available in the 32GB version of the Optane™ Memory because it has Block cache and File cache. The 16GB only has Block cache. The file cache is the one that allows you to schedule the task.
And more info from the Surprising facts link you gave:

Secondly, while the RST system is running, the driver will continuously generate caching. And here there is one important difference between Optane Memory modules of different capacities: on a 16GB device, only block level caching is supported, on a 32GB device, block level caching and file leveling (both work simultaneously). In the case of block caching, the decision to cache a block occurs instantly at the time of the I / O request. In the case of file caching, the driver monitors the frequency of access to files and puts it all in a special table, which then (at the time of system downtime or according to the user's schedule) is used to determine which files are left in the cache, which are deleted and which are added.
Both types of caching use rather clever algorithms for caching decision making – I can not describe them deeply, but for general understanding I note that, for example, video files are not cached (yes, the driver looks at File extension), the size of the file is taken into account, the type of load is determined – caching preferences are given to random access rather than sequential, which makes sense due to extremely slow operation of hard disks in random access operations, etc. On the Internet, I met some negative comments on the topic that "the cache will be immediately overwritten with data," "16GB capacity is not enough for anything" and the like – usually reviews from people who have never tested Optane Memory. I have not yet heard negative feedback about the performance of this solution from any of our partners I work with.
 
Last edited:
Mar 27, 2009
12,968
36
106
#74
cbn said:
I still like the idea of systems using the Optane memory with Hard drive (particularly if Intel ever allows Pentium and Celeron to do this....20th Anniversary Celeron?).
I know that I've commented about this before, but it seems to me, that "Optane Memory" (cache) is most suitable to budget rigs, that would still utilize a HDD for storage purposes, rather than go all-SSD like a more expensive system. I really think that Intel missed their target market completely with Optane.
With SO-DIMMs being 30mm x 69.6mm (DIMMs being even larger still) and the Optane memory having so much extra room on 22mm x 80mm (see below)....I am hoping we are not that far away from Intel releasing some Baby Optane SO- DIMMs (or DIMMs).



Then enable on Celeron for mobile and the low end desktop chipset H310* so a person could run 1 DDR4 DIMM and 1 Baby Optane DIMM. (With full size/high capacity Optane DIMMs coming 2H 2018 for Intel Cascade Lake maybe they still have time to surprise us for the second generation Coffee Lake? And if they don't maybe in 2019 we could see the small Optanes finally transition from NVMe to DIMM)

P.S. One advantage to having Optane on a DIMM (rather than NVMe) is that it not only is the latency a lot less but it also becomes byte addressable (so no write amplification).

EDIT: NVMe Optane does not suffer from write amplification. See this article (and quote from it below):

Understanding why the Intel Optane is a disruptive technology is a bit more complicated…but not really. When you erased information from a hard drive, what would typical happen was that the index to that information was destroyed, and then the information was simply ‘run over’ as new information was added to that spot on the hard drive eventually. SSDs don’t work like that. Information is stored on an SSD in pages which are typically 4kb, 8kb, or 16kb in size. Pages are stored in a block and may consist of 128 or 256 pages per.
The difficulty with NAND storage is that erasing data in a SSD could only occur in blocks so, theoretically, a 4K page deletion could require the movement of as much as 4mb in data to clean (or store a file) on that specific block. This is write amplification and limits the SSD lifespan in result. The Intel Optane SSD 900P, and specifically Intel 3D XPoint memory changes this, in that, this memory can add, delete or replace information at the most basic level without requiring the large movement of data to accommodate such. This results in significant performance and endurance increase and is very much considered disruptive as it has the potential to change the storage industry as we see it today.
*One concern I have with 1 DRAM DIMM and 1 Optane DIMM is that it maybe need some special software to run properly...and if this ends up being a variation of the Optane memory driver/application then H310 will probably not be able to use since it already has this exclusion.
 
Last edited:
Mar 27, 2009
12,968
36
106
#75
Maybe 1 DRAM DIMM and 1 Optane DIMM (which I mentioned in the post above) could work without special software if one memory channel on the processor was disabled (but two DIMMs per channel were kept). In this way perhaps the two DIMMs would essential act as one NVDIMM-P?

However, if this were done I believe the processor have to be BGA or used in a motherboard other than the H310 (because the two slots in a H310 board meant for dual channel).

P.S. If Board was BGA perhaps the Optane would not to exist as a DIMM....it could just be soldered onto the motherboard. Then the motherboard could just have one DIMM slot (used, of course, by a conventional DRAM based DIMM).
 
Last edited:


ASK THE COMMUNITY