Comments Locked

34 Comments

Back to Article

  • TheUnhandledException - Friday, June 14, 2019 - link

    The timing seems a bit weird with the move to PCIe 4.0 underway. 8 lanes of PCIe 3 = 4 lanes of PCIe 4.
  • shabby - Friday, June 14, 2019 - link

    Also It doesn't saturate pcie3.0 x4 either.
  • jordanclock - Friday, June 14, 2019 - link

    The article addresses that:

    "Keeping in mind that we are talking about an enterprise-grade controller and appropriate drives, we mean sustained performance, not peak performance."

    So it likely can burst to much higher speeds at would exceed x4 links.
  • shabby - Saturday, June 15, 2019 - link

    The spec sheet says around 3GB/sec so not sure what that quote is talking about when I read it.
  • eek2121 - Sunday, June 16, 2019 - link

    PCIE 3.0 x4 links. It still would not match PCIE 4.0 x4 links.
  • HStewart - Friday, June 14, 2019 - link

    Majority of computers will not have PCI-4 interface and likely even when PCI 5 is out. The majority of market it would probably laptops and I would be curious if TB 3.0 (IUSB4) would have interface to PCIe 3.0 x4 This would nice adapter out there..
  • surt - Friday, June 14, 2019 - link

    And what percentage of computers have a spare/usable x8 slot?
  • Rocket321 - Saturday, June 15, 2019 - link

    Quite a good percentage I would suspect. You can plug x8 cards into x16 physical slots and lots of boards come with more than one.
  • Qasar - Saturday, June 15, 2019 - link

    would these even fit in a pci e x8 slot ( or x16 slot for that matter ) the same ones that GPUs use ? looks like it is keyed wrong to fit in a standard pcie slot...
  • Casper42 - Saturday, June 15, 2019 - link

    Its an M.2 derivative, not a traditional PCIe slot.
    So this entire branch of comments is useless because NO ONE will have a M.4 slot on their laptop or desktop either.
    You would need an adapter with either x8 PCIe to 1 of these, or maybe a x16 slot adapter that could take 2 of them.

    Realistically, no clue who the market is for these.
    Servers are moving more and more to U.2 (NVMe in a 2.5" drive on a modified SAS connector)
    Plenty of space for NAND (not limited by 2280 specs) and as mentioned above, will support PCIe 4.0 in 2020 with the launch of new Rome boards and Cooper/Ice Lake on the Intel side.

    Desktop platforms from Intel certainly don't have x8 laying around considering how few PCIe lanes even the latest 9900K has.
  • mkaibear - Saturday, June 15, 2019 - link

    Can't work out if you're giving a pro or con comment...

    These will undoubtedly be bundled with x8 PCIe adaptors for super fast workstation storage. That's the target market. They won't be used in servers except in certain niche markets - as you say the U.2 appears to be the way forwards for those.
  • Valantar - Saturday, June 15, 2019 - link

    That doesn't manke much sense - if that was the use case, why not just make it a HHHL x8 AIC, which would have much broader compatibility and could thus sell to everyone from HEDT and workstation users to servers? As the article mentions, the company behind this makes a lot of custom/bespoke solutions, so this is likely for some specialized use case where custom motherboards are the norm. Showing them off at Computex is likely just to say "Hey, we can make custom, high-performance enterprise-grade SSDs".
  • mkaibear - Monday, June 17, 2019 - link

    >why not just make it a HHHL x8 AIC

    Because if you do that then you need a different SKU for each product, whereas if you have an adaptor available you can have a single SKU for the drive plus relatively simple adaptors?

    Putting it solely on a board means it's only ever useful for something which has precisely that form factor. Putting it on an adaptor means that you can put it in a half-height board, or a full height one, or in a future workstation board, or in a laptop if the form factor takes off. Etc, etc, etc.

    The only reason you'd want to use a dedicated HHHL board is if you need the mahoosive heatsink.
  • Qasar - Saturday, June 15, 2019 - link

    ahh thats what i thought... cause HStewart seems to imply that it was going into a pce slot, then surt asked about x8 slots... :-)
  • peevee - Monday, June 17, 2019 - link

    "will support PCIe 4.0 in 2020 with the launch of new Rome boards"

    Why 2020? New EPYC is shipping shortly (if not already).
  • Irata - Monday, June 17, 2019 - link

    Wouldn't that make the other x16 slot x 8 in that case ?
  • HStewart - Saturday, June 15, 2019 - link

    This device is not x8 or x16 slot and TB 3 can handle at least x4 and likely more in something situation.

    But realistic - most people don't need any of these slots - HEDT market is a small part of actual market.
  • Qasar - Saturday, June 15, 2019 - link

    these are NOT for the HEDT market HStewart, these are aimed at server and workstation.
  • peevee - Monday, June 17, 2019 - link

    "and TB 3 can handle at least x4"

    I hope by "at least" you mean "at most", and it is just PCIe3.0. There is no higher-speed controllers, and the advertised 40Gb/s data rate is obvious marketing BS.
  • Korguz - Saturday, June 15, 2019 - link

    surt, hstewart is talking about the speed, pcie 4 or 5, not the slots.. .and i think the amount of computers with pcie 4 will go up.. staring next month :-)
  • shabby - Saturday, June 15, 2019 - link

    When intel finally comes out with pcie4 you'll be licking your wounds.
  • Korguz - Sunday, June 16, 2019 - link

    shabby... and when will that be?? mid 2020 ???
  • brakdoo - Friday, June 14, 2019 - link

    Max read 3.2 GB/s? You don't need more than pcie 3.0 x4 for that. And 16 channels? 8 channel drives have the same speed. At least IOPS are good.
  • DanNeely - Friday, June 14, 2019 - link

    something's off on that specsheet. The performance is listed for U.2 which is only an x4 interface. Combined with the numbers being dead on for an x4 3.0 bus I suspect someone in marketing may have screwed up the copy/paste when printing them. Either that or "2 m.2 glued together" doesn't just mean the physical form factor; but that it actually appears to the US as two drives each with 3.2/2.8gbps of io. The latter seems rather silly since in that case your low production volumes probably trump any savings from the marginally lower part count in the system.
  • jordanclock - Friday, June 14, 2019 - link

    The two U.2 could appear as one device. It would just take the x4 lanes of both interfaces. As the article says, this company is aiming for specialized systems and their numbers are for sustained throughput, not burst. So x4 is fine for sustained, but they're not that far off from saturating x4 and thus x8 makes sense.
  • azfacea - Friday, June 14, 2019 - link

    PCI-SIG should just trademark m.3 m.4 m.5 .... m.20 to stop every fart being branded as m.x
    seriously WTF? i have no problem with this product existing but imagine if logitech and razer started inventing USB 4, USB 5, WIFI 8 and what not
  • Santoval - Friday, June 14, 2019 - link

    This looks like it only allows for more flash capacity, rather than more performance. Its specs (save perhaps the random 4KB read IOPS) are beaten by the cutting edge M.2 PCIe 3.0 x4 SSDs, while the first M.2 PCIe 4.0 x4 SSDs that just cropped up are already well beyond this in performance.

    On the other hand if these numbers indeed refer to fully sustained performance then the comparison might be difficult, since commercial SSD vendors only report peak performance numbers. That would still suggest higher sustained performance than any M.2 PCIe 3.0 x4 SSD though, but not necessarily than the newest M.2 PCIe 4.0 x SSDs,
  • erinadreno - Friday, June 14, 2019 - link

    I don't recall any commercial drive begin able to achieve 800K random read ops, especially given E16 is just E12 with pcie 4.0 PHY. Although they might benefit from higher bus bandwidth, but they only implements 8 channel thus have inferior random IOPS.
  • peevee - Monday, June 17, 2019 - link

    It could only be 800K with something like 128+ queue length. You cannot get around PCIe + flash latencies by making the buses wider.
  • jordanclock - Friday, June 14, 2019 - link

    The sustained vs peak numbers are likely because this is aimed at industries where there will be large amounts of continuous bandwidth usage. They likely don't care much about peak speeds, only that they're high enough for occasional bursts.
  • CheapSushi - Friday, June 14, 2019 - link

    I think M.3 (NF1) did good enough in terms of allowing more NAND per PCB, Weren't Samsung showing 16TB QLC M.3's? They even showed off hotswap variants with SuperMicro. I think that was fine.
  • peterfares - Saturday, June 15, 2019 - link

    Would a slot capable of taking this be able to instead take two M.2 drives? The screw holes could potentially be an issue. Would be cool to have that flexibility.
  • invasmani - Saturday, June 15, 2019 - link

    I'd like a PCIe 3.0 x16 interface M.8 SSD I'd swap my GTX980 into the PCIE 3.0 x16 slot that's x4 link speed. I mean a 1% performance loss at 4K in exchange for a drastically better SSD that's a more general purpose system improvement seems like a fair trade off.
  • Skeptical123 - Tuesday, June 18, 2019 - link

    Cool, this is the current (PCIe gen 4 is not here just yet) generation of standardized tech pushed to the max with out doing too much crazy stuff. Ie standard M.2 PCI interface, just two, same z height, same NAND, just 2 or 3 controls daze changed together.

Log in

Don't have an account? Sign up now