Become a fan of Slashdot on Facebook

 


Forgot your password?
Close
typodupeerror
×
Intel Data Storage Hardware Technology

Intel Unveils Optane SSD DC P4800X Drive That Can Act As Cache Or Storage (hothardware.com) 63

Posted by BeauHD from the two-in-one dept.
MojoKid writes from a report via HotHardware: Intel unveiled its first SSD product that will leverage 3D Xpoint memory technology, the new Optane SSD DC P4800X. The Intel SSD DC P4800X resembles some of Intel's previous enterprise storage products, but this product is all new, from its controller to its 3D Xpoint storage media that was co-developed with Micron. The drive's sequential throughput isn't impressive versus other high-end, enterprise NVMe storage products, but the Intel Optane SSD DX P4800X shines at very low queue depths with high random 4kB IO throughput, where NAND flash-based storage products tend to falter. The drive's endurance is also exceptionally high, rated for 30 drive writes per day or 12.3 Petabytes Written. Intel provided some performance data comparing its SSD SC P3700 NAND drive to the Optane SSD DC P4800X in a few different scenarios. This test shows read IO latency with the drive under load, and not only is the P4800X's read IO latency significantly lower, but it is very consistent regardless of load. With a 70/30 mixed read write workload, the Optane SSD DC P4800X also offers between 5 and 8x better performance versus standard NVMe drives. The 375GB Intel Optane SSD DC P4800X add-in-card will be priced at $1520, which is roughly three times the cost per gigabyte of Intel's high-end SSD DC P3700. In the short term, expect Intel Optane solid state drives to command a premium. As availability ramps up, however, prices will likely come down.
This discussion has been archived. No new comments can be posted.

Intel Unveils Optane SSD DC P4800X Drive That Can Act As Cache Or Storage More

Intel Unveils Optane SSD DC P4800X Drive That Can Act As Cache Or Storage

Comments Filter:

  • Stay away from Xpoint (Optane) products for a while.
    They're far, far off their initial promises, which points to manufacturing issues.

    http://semiaccurate.com/2016/0... [semiaccurate.com]

    • Re: (Score:2)

      by haruchai ( 17472 )

      Stay away from Xpoint (Optane) products for a while.
      They're far, far off their initial promises, which points to manufacturing issues.

      http://semiaccurate.com/2016/0... [semiaccurate.com]

      I imagine OCZ must have mixed feelings about any Optane story given their Octane SSDs debuted 5 years ago

  • When are we going to get drives that are just a bunch of blocks so that we can do our own wear and use leveling? Why should I trust a tiny computer on the SSD when I've got one that can do far more on the other end of the SATA line?

    • PCI-E X2-X4 is a lot faster then sata.

    • Not soon, probably never.

      The disk controller interfaces deliberately abstract those writeable areas. You would need to toss SATA/NVMe entirely and develop a new standard.

      This leaves the SSD manufacturers free to choose a variety of controllers and flash chips, and they can optimize the firmware for their hardware without worrying about meddling from userspace.

      Even if you did develop a standard that obstensibly provides direct access to low-level sectors, there is nothing in the world that you can do to stop

    • Re: (Score:2)

      by Trogre ( 513942 )

      What you're proposing is effectively ditching Integrated Drive Electronics and going back to CPU or card-based disk controllers.

      I look forward to seeing 1701 error codes again.

  • Cache? (Score:3)

    by wonkey_monkey ( 2592601 ) on Monday March 20, 2017 @07:10PM (#54078211) Homepage

    Intel Unveils Optane SSD DC P4800X Drive That Can Act As Cache Or Storage

    I might be missing something, but if you're going to mention that it can act as cache it would be nice to include something about that bit in the summary.

    No-one reads TFA anyway, of course, but in this case it's counter-blocked because I'm using uBlock, so I'm doubly put off.

    • Re: (Score:3)

      by Sarusa ( 104047 )

      Yeah, because it's not true. It doesn't work well as cache, and if you tried it you'd burn it out fast. It was SUPPOSED to, they promised a lot that you could use this as both storage as main memory, but they couldn't make it work and even Intel has stopped talking about Apache Pass.

      Basically, this is a evolutionary step in flash, giving you much faster random write and maybe 2x the endurance for 2x the price, so not bad as a disk.

      But it's not the revolutionary step Intel was promising.

      • The endurance listed is 30 drive writes per day for the 375GB model. That's 11.25TB/day, or about 130MB/s sustained 24/7 writes. A cache device should be spending 90% of its time being read, rather than written (or it's not doing that good a job as a cache[1]), which means that even if you're reading from it at about half of its peak rate every second of every day and getting then it's going to last its rated lifetime. Additionally, for a cache device, we really don't care if it burns out. When it does,

        • Re: (Score:2)

          by Sarusa ( 104047 )

          It's not great for reads though - Even the consumer m.2 SSD in my PC is faster for that. It's just really fast on small random writes, or small reads under LOW read/write load (small queue depth) because it really wins for latency there. So you have to be using it on something with lots of writes and low load small reads to get amazing benefit over current flash - and then it burns out fast.

          I'm sure there will be some very targeted applications where this is perfect, like maybe mostly data acquisition an

  • Its rather exaggerated (Score:5, Interesting)

    by m.dillon ( 147925 ) on Monday March 20, 2017 @07:12PM (#54078227) Homepage

    Intels claims are rather exaggerated. Their claims have already been torn apart on numerous tech forums. At best we're talking only a ~3-5x reduction in QD1 latency and the intentionally omit vital information in the specs to force everyone to guess what the actual durability of the XPoint devices is. They say '12PB' of durability for the 375GB part but refuse to tell us how much overprovisioning they do. They say '30 drive writes per day' without tellling us what the warrenty will be.

    In fact, over the last 6 months Intel has walked back their claims by orders of magnitude, to the point now where they don't even claim to be bandwidth competitive. They focus on low queue depths and and play fast and loose with the stats they supply.

    For example, their QOS guarantee is only 60uS 4KB (99.999%) random access latency and in the same breath they talk about being orders of magnitude faster than NAND NVMe devices. They fail to mention that, for example, the Samsung NVMe devices also typically run around ~60-70uS QD1 latencies. Then Intel mumbles about 10uS latencies but bandies about large factors of improvement over NAND NVMe devices, far larger than the 6:1 one gets simply assuming 10uS vs 60uS.

    Then they go on to say that they will have a NVDIMM form for the device later this year, with much faster access times (since in the NVMe form factor access times are constricted by the PCIe bus and block I/O protocol). But with potentially only 33,000 rewrite cycles per cell to failure that's seriously problematic. (And that's the best guess, since Intel won't actually tell us what the cell durability is).

    --

    The price point is way too high for what XPoint in the NVMe format appears to actually be capable of doing. The metrics look impossible for a NVDIMM form later this year. Literally we are supposed to actually buy the thing to get actual performance metrics for it? I don't think so.

    Its insane. This is probably the biggest marketing failure Intel has ever had. Don't they realize that nobody is being fooled by their crap specs?

    -Matt

    • Don't worry about the claims. Earlier articles said that all the xpoint memory was already spoken for by systems integrators for use in server class machines for at least the next year or so. Although Intel announces the SSD, it's still not available for purchase by the general public.
      • ummm these are enterprise cards for server class machines, they aren't meant for the general public.

    • They say '12PB' of durability for the 375GB part but refuse to tell us how much overprovisioning they do. They say '30 drive writes per day' without tellling us what the warrenty will be.

      Those numbers (12.3PB) work out to be very nearly 3 years [wolframalpha.com], for what it's worth -- perhaps (???) there's a 3-year warranty or something (or expected lifetime).

    • But with potentially only 33,000 rewrite cycles per cell to failure that's seriously problematic. (And that's the best guess, since Intel won't actually tell us what the cell durability is).

      I can guarantee you it's not 33.000 R/W cycles - the only tech that would allow that is SLC, and practically nobody sells SSD based on SLC anymore. A few manufacturers sell highly overpriced SLC-based SD and microSD cards. Hell, nowadays you'll struggle to even find MLC-based SSDs (~10.000 rewrite cycles AT BEST). Every SSD manufacturer today uses TLC, which means 1000 R/W cycles per cell.

      • Is there any way we can banish xx.xxx (or even worse, xx xxx) notation?

        It is incredible that the world doesn't have a single standard for numbers (not counting scientific notation, which is something different, and btw rarely used).

        How are we supposed to differentiate 33.000 (i.e. 33 thousand) from 33.000 (33, to an accuracy of 3 decimals)?

        We can go to the moon, but can't state numbers unambiguously. Mind officially blown...

      • I can guarantee you it's not 33.000 R/W cycles - the only tech that would allow that is SLC, and practically nobody sells SSD based on SLC anymore. A few manufacturers sell highly overpriced SLC-based SD and microSD cards. Hell, nowadays you'll struggle to even find MLC-based SSDs (~10.000 rewrite cycles AT BEST). Every SSD manufacturer today uses TLC, which means 1000 R/W cycles per cell.

        You're still talking about flash technology, as far as I can see... the point of Optane is that it's using this new phase-change stuff. Supposedly it would have a thousand times the write endurance, and be a thousand times faster than NAND flash (though I don't think they ever said whether it was SLC/MLC/TLC they were comparing to). It doesn't look like either promise has come true, though.

    • Matt I respect your work on both Freebsd and Dragonfly but have you ever seen latency spikes from lots of small writes? My home PC has 2 raid 0 Samsung pros. Latency spikes go well over 1000 when running several Hyper-V or VMware workstation sessions easily. I am sure a server grade solution would be a little bit better.

      From what I have read is low latencies is the key and writes. Xpoint doesn't use blocks and does have endurance issues. 550,000 iops and lower latency is no joke on a SQL database or virtual

    • This is probably the biggest marketing failure Intel has ever had.

      This is a marketing failure. Not a market failure. Currently all we have to go buy is what Intel say. It won't be a market failure until a) the product is released and b) the product fails in the market.

  • This seems like a particularly dumb idea. Wouldn't it be better to just add more RAM?

    • Re: (Score:2)

      by darkain ( 749283 )

      SSDs for cache is extremely common in the storage server market. Check the section on L2ARC https://en.wikipedia.org/wiki/... [wikipedia.org]

    • Re: (Score:2)

      by AHuxley ( 892839 )
      Need to find a product thats much faster than SSD but not as expensive as a lot of RAM.

    • Go run a server? For a corporate file shared drive using this as a tier in server 2012 R2 or later will put the frequently accessed files on this greatly improving performance over accessing the slower mechanical disks in the raid. SQL databases can use this for stored procedures and frequently accessed data too as a cache from the slower SAN and latency

    • RAM is expensive and a lot of it requires quite a bit on energy to keep refreshing,. Also it's not persistent so files in memory don't survive a reboot. This is especially true if you are working with data sets in the tens of gigabytes.

  • XPoint is well... pointless. It can't compete with MRAM and by the time it matures enough (If it matures enough) to substitute flash in any kind of significant way MRAM is likely to have already taken over.

    MRAM has effectively infinite read/write endurance, high density and performance characteristics of static ram.

    XPoint even if executed perfectly has only a narrow window in which it can hope to remain relevant.

    • The first article I read that promised MRAM real soon now was around 2008. I'm not holding my breath.

      • There's MRAM available for sale right now. Everspin sells it:
        https://www.everspin.com/ [everspin.com]

        The problem is it has less density than DRAM and it's a lot more expensive. It does have better latency though, so it could be used as a kind of persistent last level cache.

Slashdot Top Deals

Our OS who art in CPU, UNIX be thy name. Thy programs run, thy syscalls done, In kernel as it is in user!

Close