1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

News Intel announces Atom C3000 16-core Denverton parts

Discussion in 'Article Discussion' started by Gareth Halfacree, 22 Feb 2017.

  1. Gareth Halfacree

    Gareth Halfacree WIIGII! Staff Administrator Super Moderator Moderator

    Joined:
    4 Dec 2007
    Posts:
    11,980
    Likes Received:
    1,581
  2. Vault-Tec

    Vault-Tec Green Plastic Watering Can

    Joined:
    30 Aug 2015
    Posts:
    8,859
    Likes Received:
    821
    Naturally :rolleyes:
     
  3. edzieba

    edzieba Virtual Realist

    Joined:
    14 Jan 2009
    Posts:
    3,013
    Likes Received:
    241
    Xeon-D with weaker cores (didn't work out well for ARM when they tried that, although the Denverton cores beat the pants off any core ARM has in SP performance), or tiny cut-down Xeon Phi?
     
  4. Taua

    Taua Member

    Joined:
    20 Sep 2014
    Posts:
    87
    Likes Received:
    0
    :D
     
  5. IamSoulRider

    IamSoulRider Member

    Joined:
    24 Aug 2016
    Posts:
    89
    Likes Received:
    4
    Yeah, this is the only fix they could find for the C2000 problems...
     
  6. David

    David RIP Tel

    Joined:
    7 Apr 2009
    Posts:
    13,046
    Likes Received:
    2,115
    Followed by an unlocked version that costs more than a quad?

    Bitter? Me? No, of course not. :sigh:
     
  7. Chicken76

    Chicken76 Member

    Joined:
    10 Nov 2009
    Posts:
    907
    Likes Received:
    23
    How exactly is it going to address 64GB of RAM on a single channel? Can it use RDIMMs or FBDIMMs?
     
  8. Gareth Halfacree

    Gareth Halfacree WIIGII! Staff Administrator Super Moderator Moderator

    Joined:
    4 Dec 2007
    Posts:
    11,980
    Likes Received:
    1,581
    You're not limited to a single DIMM slot just because you have a single memory channel. Think of how many dual-channel architecture motherboards have four slots; there's nothing to stop a single-channel chip addressing two or even four DIMMs, it just can't address them all at the same time.

    We had multiple memory chips in computers long before multiple channel memory buses were a thing. The Z80-based Sinclair ZX81 had either one 1KB or two 512B chips depending on what Sinclair could buy cheap at the time. Intel didn't get dual-channel support until the 865 chipset, and you had multiple SIMMs and DIMMs in PCs long before that: my old 8088 had 1MB of RAM spread across four 256KB SIMMs.

    Dual, triple, and quad-channel memory architecture isn't about addressing more memory than single-channel; it's about addressing the memory you've got faster.
     
  9. jb0

    jb0 Active Member

    Joined:
    8 Apr 2012
    Posts:
    433
    Likes Received:
    50
    Given it is targeting embedded systems, I'd be surprised if most implementations can use ANY DIMMs. RAM soldered directly onto the board shall rule the day.
     
  10. Gareth Halfacree

    Gareth Halfacree WIIGII! Staff Administrator Super Moderator Moderator

    Joined:
    4 Dec 2007
    Posts:
    11,980
    Likes Received:
    1,581
    It's targeting "high-throughput yet low-power devices like network attached storage and smart network switch systems as well as microservers" - and most of the NASes and all of the microservers that would have a 16-core Atom in them will be taking DIMMs for their RAM, guaranteed. Just take a look at Synology's current higher-end Atom NASes, like the RS815+, which has two DDR3 DIMM slots in it.
     
  11. Chicken76

    Chicken76 Member

    Joined:
    10 Nov 2009
    Posts:
    907
    Likes Received:
    23
    As you put more DIMMs on a channel, the signals degrade. That's why we don't see more than two slots per channel for unbuffered DIMMs. Going to three would require putting a 'register' chip on each DIMM to receive the serialized addresses from the memory controller and then access the actual SDRAM chips (thus RDIMM). Of course, you can put more than two DIMMs on a channel, but you'd have to lower the frequency significantly to maintain signal integrity, and possibly use higher voltage which would increase power consumption, neither of which are desirable.
    Going to four DIMMS per channel usually requires buffering both address and data lines (FBDIMM).

    Anyway, my point was this: with 16GB being the biggest capacity DIMMS available for DDR4, how is the new Atom going to address 64GB? Four 16GB DIMMS on a single unbuffered channel? No way! Both RDIMM and FBDIMM need specific silicon in the memory controller. So the question is really, if the chip has the capability to use (besides UDIMM) RDIMM or FBDIMM?

    If you're going to say: "there will be 32 GB DDR4 DIMMS in the future", then you may only be technically right, but practically wrong. No one is going to bother to put out such huge memory modules for the "client market" (Intel jargon).
    Remember how DDR3 theoretically had 16GB unbuffered modules available? (from a single company and with a $300+ pricetag) Have you actually seen one? Hell, do you know anyone who knows someone who has seen one? The same thing will happen to 32GB unbuffered DDR4 DIMMs.

    It kind of makes you wonder how they validated the chips to work with 64GB RAM, doesn't it?

    Didn't the ZX have 48KB of RAM? I had a compatible in the early nineties that had 56 or 64 KB (can't remember the exact value now)

    In light of what I wrote above, it kinda' is about both, wouldn't you agree?
     
  12. Gareth Halfacree

    Gareth Halfacree WIIGII! Staff Administrator Super Moderator Moderator

    Joined:
    4 Dec 2007
    Posts:
    11,980
    Likes Received:
    1,581
    That'd be a question for Intel - I'm just telling you what the Ark entry says!

    That's the ZX Spectrum. Sinclair Research's first computer was the ZX80, which had 1KB and was the first fully-functional computer in the world to sell for under £100; the ZX81 followed a year later with a massive reduction in chip count thanks to a Ferranti Uncommitted Logic Array (ULA) and a price drop, but still 1KB of RAM. The ZX Spectrum launched a year later as an attempt to beat Acorn, bringing colour, basic sound, and a choice of 16KB or 48KB of RAM, but at a higher price (£129 for the 16KB version) and without the option of buying it as a self-assembly kit.

    (Technically, Sinclair had a computer before the ZX80: the MK14. It launched under the Science of Cambridge banner, though, and was more of a calculator-on-steroids than anything you'd recognise as a modern computer.)
    It can certainly help, to be sure, but that's not why the first dual-channel architectures were developed (to my knowledge, anyway - happy to be corrected!)
     
  13. Chicken76

    Chicken76 Member

    Joined:
    10 Nov 2009
    Posts:
    907
    Likes Received:
    23
    True, I wasn't questioning your report, just throwing the question out into the ether.

    Ah, I see. I had no knowledge of the prior models. In fact I have never actually seen an original Sinclair unit, only compatibles. (compatibles, pfff..., who am I kidding, more like rip-offs. The Spectrum was not open source hardware, was it?)
     
  14. Gareth Halfacree

    Gareth Halfacree WIIGII! Staff Administrator Super Moderator Moderator

    Joined:
    4 Dec 2007
    Posts:
    11,980
    Likes Received:
    1,581
    No, but a basic enough design to easily churn out copies of. Back in those days computer PCBs were double-layer at most - you can copy one of those with a photocopier!

    I've never seen a proper foreign Spec-clone in person. There's a guy selling ex-USSR clones, boxed and unused, on FleaBay, but I can't justify the asking price. Sadface.
     
  15. Chicken76

    Chicken76 Member

    Joined:
    10 Nov 2009
    Posts:
    907
    Likes Received:
    23
    Unfortunately I don't have mine anymore, or I would send it to you.
     

Share This Page