Recent

% | $
Quotes you view appear here for quick access.

Netlist Inc. Message Board

  • vicl2010v vicl2010v Jan 14, 2010 4:03 PM Flag

    NLST Needham presentation Jan 14, 2010 - part1

    http://www.wsw.com/webcast/needham35/nlst/
    Needham Growth Stock Conference
    January 14, 2010 at 2:30 pm ET

    Good presentation. Lots of questions asked.

    Clearly explains cost benefit of HyperCloud at every memory chip density. So as higher density memory appears, NLST can create even higher density memory modules.

    HyperCloud - higher density, max memory increased, higher speech achieved (1333MHz vs. 1066 or 800 with other's memory), lower power consumption (memory chips are put in low power as others are used). Need to buy less servers filled with max memory AND run faster. Less servers, space and power savings for data centers.


    Also mentioned NetVault - for memory backup to flash disk in case of power failure as separate product. Alternative to battery backed solutions.


    HyperCloud IP go back to 2004 - when developing memory for Apple x-serve server. Was genesis of HyperCloud. The problem that solved then, is now an industry pervasive problem i.e. ability to address more memory.

    Data centers, utility computing - computer power supplied at will - driven by GOOG, MSFT, AMZN - building out very large server farms with hundreds of thousands of server. If HyperCloud does it's job properly, less servers be used, data centers more efficient and cost effective. Power consumption biggest challenge for data centers. Worldwide data centers consume as much power as country of Argentina.

    Challenge of providing power and removing heat is big problem.


    Memory is bottleneck - I/O, CPU far outpace the progress of memory - speed and capacity.

    Underutilization of server capacity in data centers - like search very memory intensive - if have more memory, are more efficient.

    Current servers running at half their potential, so more servers purchased - memory is bottleneck.

    The more memory there is in servers, the slower they run.


    With HyperCloud - double the memory, still system run at 1333 ..

    100% more memory capacity allows
    66% more higher memory speed
    (power consumption also less)


    HyperCloud tricks system into thinking looking at 2GBits memory chips when there are 2 x 1GBits memory chips.

    Difference is ..
    1GB is $2 - so 2 x 1GB is $4
    2GB chip is $10-15

    So NLST can use lowest cost chip and emulate the higher cost per bit chip to get to higher densities.


    Can use 2GB chips to emulate 4GB (which currently not available in industry).

    To use HyperCloud - no BIOS changs required - plug and play - for OEM or end-user is same as regular memory (in terms of installation).

    SortNewest  |  Oldest  |  Most Replied Expand all replies
    • NLST Planar-X PR suggests a doubling of capacity it seems.


      There was an interesting revelation in an earlier PR (on NLST Planar-X patent award) that was not explicitly mentioned on this board.

      From the Planar-X patent award PR issued by NLST:


      http://finance.yahoo.com/news/Netlist-Awarded-Patent-for-prnews-3917240602.html?x=0&.v=1
      Netlist Awarded Patent for Proprietary Planar-X Technology
      Planar-X Technology Enables 32GB Capacities for RDIMM Modules with Standard DRAM
      Press Release Source: Netlist, Inc. On Wednesday October 13, 2010, 6:00 am EDT

      quote:
      "Planar-X will be a key component to delivering 32GB RDIMM solutions using standard 2Gb DRAM technology available today," said Steve McClure, vice president of marketing at Netlist. "This achievement, combined with our HyperCloud technology, highlights the strength of Netlist's intellectual property position supporting innovative products addressing virtualization, cloud datacenters and high performance computing memory requirements."



      This suggests NLST can now make (or can envision making) 32GB memory modules using only 2Gbit memory chip packages.

      This is about twice as good as they were doing before - which was itself twice as good as the competition.


      The Samsung 32GB memory module offerings (16GB PR in 2009, 32GB PR in 2010) are only doing so using "dual-die" memory chip packaging.

      While NLST uses 2Gbit and 4Gbit "monolithic" (one die per memory chip package), Samsung has to use (2Gbit x 2) and (4Gbit x 2) i.e. "dual-die" configuration to achieve 16GB and 32GB memory modules.

      As posted before "dual-die" is frowned upon by NLST for causing asymmeytry in line delays which creates problem for timing efficiency.


      ----
      http://www.theregister.co.uk/2009/11/11/netlist_hypercloud_memory/
      Netlist goes virtual and dense with server memory
      So much for that Cisco UCS memory advantage
      By Timothy Prickett Morgan
      Posted in Servers, 11th November 2009 18:01 GMT
      ...
      The company also developed a memory packaging technology called Planar-X, which allows for two PCBs loaded with memory chips to be packaged together relatively inexpensively to share a single memory slot. This technique is cheaper and more reliable, according to Duran, than some of the dual-die packaging techniques memory module makers use to make dense memory cards out of low density and cheaper memory chips.
      ...
      Using the Planar-X double-board designs, Netlist can take 1Gb memory chips and make an 8GB memory module that costs only 20 to 30 per cent more than a standard 4GB module using 1Gb chips; using 2Gb chips, it can make a 16GB module, something no one else can do yet.
      ----




      As posted here before, NLST can do:

      16GB memory modules using 2Gbit memory chip packages (2Gbit x 1 i.e. "monolithic" or single-die)

      32GB memory modules using 4Gbit memory chip packages (4Gbit x 1) i.e. "monolithic" or single-die)


      As analyzed here before, Samsung announcements suggested:

      16GB memory modules using "dual-die" chip packages (2Gbit x 2 i.e. "dual-die").

      32GB memory modules using "dual-die" chip packages (4Gbit x 2 i.e. "dual-die").


      The Planar-X PR suggests NLST can conceivably do 32GB using just 2Gbit (probably "monolithic" since NLST seems to favor monolithic memory and frowns upon "dual-die" usage).



      This discussion does not include the other NLST IP like load reduction etc. - just discussing memory capacity and types of memory chips each needs to create 16GB and 32GB memory modules.

      As far as load reduction etc. is concerned, the Samsung offering it seems only ran to 1066MHz at full loading - with the load efficiency only helped as a side-effect by the use of smaller lithographic process (lower power memory dies), rather than load reduction IP it seems.

    • Trying to clarify the memory chips being used in the Samsung 32GB memory module.

      We know Samsung is using dual-die "4Gbit" memory chips.

      However, to clarify whether the dual-die memory package has 4Gbit x 2 memory dies in it (or 2Gbit x 2 dies) - since articles seem to suggest different things:


      ----
      http://www.theinquirer.net/inquirer/news/1719709/samsung-launch-32gb-lrdimm-ram-modules
      Samsung will launch 32GB LRDIMM RAM modules
      Thanks for the memories
      By Spencer Dalziel
      Tue Jun 29 2010, 12:32

      quote:
      Samsung said that its prototype has 72 4Gb DDR3 chips with an extra memory buffer chip and claimed this will curtail memory subsystem power load by 75 per cent. It also said that memory capacity can be increased up to 384GB per processor and 1.5 times that of a 512GB server system equipped with standard 32GB DDR3 RDIMMs.

      ----
      http://www.samsung.com/us/aboutsamsung/news/newsIrRead.do?news_ctgry=irnewsrelease&news_seq=18399
      Samsung First to Begin Shipping 40nm-class, 32-Gigabyte Memory Module for Server Applications
      on Mar 29, 2010

      The highly-efficient 32GB RDIMM consists of 36 dual-die 40nm-class 4Gb DDR3 chips that can perform at equal or greater levels to a 40nm-based 16GB RDIMM with no increase in power consumption.

      In addition, replacing 12 DRAM modules of 16GB density with just six 32GB modules would achieve a 192GB total density, while allowing the DRAM operating speed in a two-way server system to rise by 33 percent from 800 megabit per second (Mbps) to 1,066Mbps, as power is cut by 40 percent.
      ----



      So Samsung is referring to "36 dual-die 40nm-class 4Gb DDR3 chips". While the other article talks of "72 4Gb DDR3 chips".

      Also are they using 4Gbit x 2 dual-die or is it 2Gbit x 2 dual-die ?


      72 memory dies x 4Gbit per die = 288 Gbit total
      288/8 = 36GByte

      36GByte is sufficient to make 32GB plus enough leftover for error correction (as error correction is standard for server memory - unlike much consumer memory modules).

      Which would suggest 72 memory dies (and thus 36 dual-die packages) is the right number in order to construct a 32GB memory module.

      • 4 Replies to vicl2010v
      • The economic advantage (to NLST - i.e. cheaper to manufacture):

        - to make 32GB memory modules, NLST would use 4Gbit monolithic i.e. normal memory chips (contain just one 4Gbit memory die)

        - Samsung would requires the more complicated "dual-die" (2 x 4Gbit memory dies in one package) in order to squeeze in 32GB on the memory module circuit board, since they lack the space saving NLST IP (Planar-X and "embedded passives" IP).



        NLST can use the same memory chips others are using - so they benefit from improvements in memory technology that others (like Samsung) benefit from - for example moving to 40nm lithographic process, WHILE retaining advantages that NLST IP adds to that:


        - lower power (reduces power by only keeping some ranks on within a memory module)
        - lower price (by use of lower price per bit chips to emulate higher price per bit chips - an advantage that remains as higher density memory emerges)
        - and higher speed (for maxed memory systems since maxing memory slows effective speed in current systems because of electrical load etc. issues)



        So basic memory chip technology can continue to improve - as others benefit, so does NLST, WHILE retaining benefit of the NLST IP on top of that.




        So to summarize:

        So while Samsung 32GB is a good proof of principle to demo the new 40nm memory die process (smaller chips, less power etc.).

        However, in practice NLST will outperform Samsung (for instance):

        - will use cheaper "monolithic" memory chips instead of costlier and more complex "dual-die" memory chips to achieve the same memory density

        - if competitors use "monolithic" as well, NLST can use half density memory chips (much cheaper than half price - since memory prices escalate dramatically with density) - so NLST can use 2Gbit memory chips to make 16GB memory modules, while competitors would have to use 4Gbit (much more expensive) ones, OR they would have to use "dual-die" (2 x 2Gbit) ones.

        - run at higher bandwidth - NLST can run full speed 1333MHz at much higher memory loading than Samsung is able to do at 192GB. NLST runs full speed at 384GB even.

        - as base memory chips move to even smaller lithographic process than 40nm, NLST can benefit from those improvements as well (WHILE leveraging the power efficiency etc. improvements from use of NLST IP).

      • Secondly the bandwidth issue.

        Using the lower power memory chips (thanks to newer 40nm lithographic process to make the memory dies) Samsung is able to lower load so it can provide up to 1066Mbps at 192GB memory loading.

        How well does Samsung fare when memory loading is higher than 192GB ?

        NLST is able to provide the full 1333MHz at memory loading much higher than 192GB i.e. up to 384GB. And that is using just the previous generation memory chips.

        Plus NLST always has access to the power efficiency improvements that take place with newer 40nm lithographic process dies.

        So you can see that NLST has one more layer of efficiency that it can extract BEYOND what Samsung and other can do with conventional memory module design.






        In contrast it seems Samsung has not cracked the bandwidth issue. Even using lower power memory chips - which presumably lead to lesser load on the memory lines (electrically) - they are unable to exceed 1066MHz at high memory loading.

        This is from the article above:

        quote:
        In addition, replacing 12 DRAM modules of 16GB density with just six 32GB modules would achieve a 192GB total density, while allowing the DRAM operating speed in a two-way server system to rise by 33 percent from 800 megabit per second (Mbps) to 1,066Mbps, as power is cut by 40 percent.




        So summarizing the bandwidth advantage:

        - at 192GB memory loading, the Samsung 32GB memory module can run at 1066Mbps. How much can it run at 384GB - even slower than 1066MHz ?

        - NLST is able sustain full speed of 1333MHz up to 384GB.

      • With Samsung 32GB memory modules thus likely having 72 memory dies (packaged in 36 dual-die packages - each package thus containing 8Gbits).

        So why does Samsung call it "36 dual-die 40nm-class 4Gb DDR3 chips".


        The article on NLST below suggests NLST is able to achieve:
        - 8GB using 1Gbit memory chips
        - 16GB using 2Gbit memory chips
        - (and probably 32GB using 4Gbit memory chips)

        using conventional single-die memory chips (thanks to NLST's Planar-X tech):

        ----
        http://www.theregister.co.uk/2009/11/11/netlist_hypercloud_memory/
        Netlist goes virtual and dense with server memory
        So much for that Cisco UCS memory advantage
        By Timothy Prickett Morgan
        Posted in Servers, 11th November 2009 18:01 GMT
        ...
        The company also developed a memory packaging technology called Planar-X, which allows for two PCBs loaded with memory chips to be packaged together relatively inexpensively to share a single memory slot. This technique is cheaper and more reliable, according to Duran, than some of the dual-die packaging techniques memory module makers use to make dense memory cards out of low density and cheaper memory chips.
        ...
        Using the Planar-X double-board designs, Netlist can take 1Gb memory chips and make an 8GB memory module that costs only 20 to 30 per cent more than a standard 4GB module using 1Gb chips; using 2Gb chips, it can make a 16GB module, something no one else can do yet.
        ----




        Firstly - how Samsung packs 32GB on a memory module versus how NLST does it.


        NLST has criticized the use of "dual-die" memory chips:

        quote:
        This technique is cheaper and more reliable, according to Duran, than some of the dual-die packaging techniques memory module makers use to make dense memory cards out of low density and cheaper memory chips.



        Samsung is having to use "dual-die" memory chips (2 x 4Gbit memory dies in one package) instead of monolithic (one die to one package) so it can cram the full 32GB in one memory module.



        NLST can make 32GB memory modules using 4Gbit "monolithic" (i.e. normal and not dual-die) memory chips:

        http://messages.finance.yahoo.com/Stocks_(A_to_Z)/Stocks_N/threadview?m=te&bn=51443&tid=19383&mid=19383&tof=5&frt=2#19383
        Netlist vRank allows for 32GB Memory DIMMs? 1-Jul-10 10:07 pm



        This is probably due to NLST ability to pack greater memory chips together using NLST IP - which Samsung is unable to do - having to thus rely on "dual-die".

    • For completeness, reposting these here:

      http://messages.finance.yahoo.com/Stocks_%28A_to_Z%29/Stocks_N/threadview?m=te&bn=51443&tid=19383&mid=19390&tof=1&frt=2#19390
      Re: Netlist vRank allows for 32GB Memory DIMMs?


      moneybaloon on mention of 32GB memory module using 4GBit memory chips in slides that accompany Jan 14, 2010 Needham presentation:

      http://messages.finance.yahoo.com/Stocks_(A_to_Z)/Stocks_N/threadview?m=te&bn=51443&tid=19383&mid=19383&tof=5&frt=2#19383
      Netlist vRank allows for 32GB Memory DIMMs? 1-Jul-10 10:07 pm


      Excellent DD.

      Although it WAS well known that NLST can make 32GByte memory modules (using 4GBit memory chips, once they become cheap). This has been discussed in conference audio.

      However this maybe the first mention of 32GByte memory modules in NLST presentations/printed material - at least I don't recall it being mention on NLST website/presentations.



      The reference is to the Jan 14, 2010 Needham presentation slides:
      http://www.b2i.cc/Document/1941/103743.pdf

      From page 20 in that pdf file:

      quote:
      ----
      OEM Benefits with HyperCloud + Planar-X (288GB 18-Slot Server Case Study)

      60% lower Memory costs for 288GB

      - Utilizes 2Gb x4 DRAM chips versus 4Gb x4 DRAM chips
      - Higher Capacity: 32GB 4 vRank DIMMs possible today with 4Gb DRAM chip
      - Better Performance
      - 2 DIMMs per channel 1333 MT/s versus 1067 MT/s
      - 3 DIMMs per channel 1333 MT/s versus 800 MT/s
      ----

      • 2 Replies to vicl2010v
      • bumped again

      • For completeness, reposting these here:


        http://messages.finance.yahoo.com/Stocks_%28A_to_Z%29/Stocks_N/threadview?m=te&bn=51443&tid=19383&mid=19391&tof=1&frt=2#19391
        Re: Netlist vRank allows for 32GB Memory DIMMs?



        We know from before that NLST can make:
        8GByte memory modules using 1Gbit monolithic memory chips
        16GByte memory modules using 2GBit monolithic memory chips

        and so by extension:
        32GByte memory modules using 4GBit monolithic memory chips


        Note the distinction between "bit" and "byte" above.
        Memory chip specs are in bits above, while memory modules are in bytes (normally 8 bits make a byte).



        While competitors can do:
        8GByte memory modules using 2Gbit monolithic memory chips
        16GByte memory modules using 4GBit monolithic memory chips

        And since 8GBit memory chips are not available yet - in the future:
        32GByte memory modules using 8GBit monolithic memory chips


        8GBit memory chips will be MUCH more expensive than 2x the cost of 4GBit memory chips (just like right now 4Gbit memory chips are much more expensive than 2x the cost of 2GBit memory chips).


        However, Samsung HAS announced 32GBbyte memory modules. But they are not using 4Gbit "monolithic" memory chips (one silicon die in one package), but are having to use (2 x 4Gbit) "dual-die" memory chips (which have two 4Gbit memory dies within one package).

        Thus, Samsung can make 32GByte memory modules, however they:

        - do not use monolithic memory chips and instead use the more complicated "dual-die" packages)

        - run at only 1066MHz (compared to NLST's 1333MHz at full memory loading)



        NLST has previously indicated how "dual-die" is inferior and causes problems:

        ----
        http://www.theregister.co.uk/2009/11/11/netlist_hypercloud_memory/
        Netlist goes virtual and dense with server memory
        So much for that Cisco UCS memory advantage
        By Timothy Prickett Morgan
        Posted in Servers, 11th November 2009 18:01 GMT
        ...
        The company also developed a memory packaging technology called Planar-X, which allows for two PCBs loaded with memory chips to be packaged together relatively inexpensively to share a single memory slot. This technique is cheaper and more reliable, according to Duran, than some of the dual-die packaging techniques memory module makers use to make dense memory cards out of low density and cheaper memory chips.
        ...
        Using the Planar-X double-board designs, Netlist can take 1Gb memory chips and make an 8GB memory module that costs only 20 to 30 per cent more than a standard 4GB module using 1Gb chips; using 2Gb chips, it can make a 16GB module, something no one else can do yet.
        ----




        Reference for Samsung 32GB memory module discussion:

        http://messages.finance.yahoo.com/Stocks_%28A_to_Z%29/Stocks_N/threadview?m=te&bn=51443&tid=12087&mid=15218&tof=1&frt=2#15218
        Re: NLST Needham presentation Jan 14, 2010 - Samsung clarification 2






        Here is an explanation of "planar-x" from that Needham conference:

        http://www.wsw.com/webcast/needham35/nlst/
        Needham Growth Stock Conference
        January 14 at 2:30 pm ET (Jan 14, 2010)
        Corporate Presentation
        Needham Conference January 2010
        http://www.b2i.cc/Document/1941/Netlist_Needham_Conf_Jan_2010.pdf

        quote:
        planar-x is a way to eliminate chip stacking and pack more chips onto a memory module using flex board technology

    • bump!! 7

    • Revisiting the Samsung 32GB memory module thanks to this post:

      http://messages.finance.yahoo.com/Stocks_%28A_to_Z%29/Stocks_N/threadview?m=te&bn=51443&tid=15196&mid=15210&tof=3&frt=2#15210
      Re: Their NetVault-NV Tier-one OEM is Dell
      quote:
      Why not use the Samsung 32GB RDIMMS? Is a cost issue?
      http://www.tomshardware.com/news/samsung-ddr3-dram-4gb-32gb,6933.html

      Some clarification of terminology:

      die - refers to the silicon wafer piece that goes inside the package
      monolithic - when they have just one die in the package
      dual-die - when they package two dies in one package (2Gbit memory die x 2 or 4Gbit memory die x 2)

      memory chip - usually refer to the die

      package - the usually black package in which the die is embedded with pins coming out

      memory chip/package sizes are usually in "bits" i.e. 2Gbit, or 4Gbit

      memory module (circuit board with ASIC and memory packages) is in "bytes" i.e. 16Gbyte or 32Gbyte




      Proceeding to the arguments ..

      Earlier I did not understand from the articles if the 4Gbit memory chip that Samsung announced this year is a monolithic 4Gbit or is a dual-die memory package (made using 2Gbit x 2 memory dies) - I assumed the latter. On rereading, it now seems it is a 4Gbit x 2 (dual-die).

      This is because on rereading last year's announcement (16GB using 2Gbit dual-die) it became clear that they were already doing 2Gbit x 2 (dual-die) last year to make 16GB memory modules.


      ----
      Last year's Samsung announcement:

      http://techreport.com/discussions.x/16600
      Samsung ships 16GB DDR3 memory modules
      by Cyril Kowaliski — 6:00 AM on March 19, 2009
      ...
      Samsung claims these are the "world's first and smallest high-density memory modules" to use 2Gb, 50nm DRAM chips, which are packed in dual-die configurations to allow for 16GB registered DIMMs. The modules themselves are rated for operation at 1066Mbps (or 1066MHz), and Samsung says it's the first to offer 16GB modules that can operate at 1.35V. It's not entirely clear whether the modules can actually hit their top speed at 1.35V, though.
      ----

      • 1 Reply to vicl2010v
      • part 2 ..

        So the new thing in the Jan 2010 PR is that they are making 4Gbit dies and then they are making dual-die packages (4Gbit x 2) now - and THIS is what is enabling them to do 32GByte memory modules.


        The main point is that Samsung is making these with 40nm process (compared to 50nm process last year) - and this enables some reduction in power consumption - this in turn reduces the load on the memory controller, thereby allowing them to have better memory bandwidth. And they maybe doing this without MetaRAM, Inphi or NLST IP.

        After doing all this, Samsung is still only able to achieve 1066MHz speeds (compared to NLST's 1333MHz).

        ----
        http://www.earthtimes.org/articles/show/samsung-first-to-begin-shipping,1223721.shtml
        Samsung First to Begin Shipping 40nm-class, 32-Gigabyte Memory Module for Server Applications
        Posted : Mon, 29 Mar 2010 02:50:40 GMT
        Author : Samsung Semiconductor, Inc.
        ...
        In addition, replacing 12 DRAM modules of 16GB density with just six 32GB modules would achieve a 192GB total density, while allowing the DRAM operating speed in a two-way server system to rise by 33 percent from 800 megabit per second (Mbps) to 1,066Mbps, as power is cut by 40 percent.
        ----




        What does NLST do - NLST is able to make 16GByte using 2Gbit monolithic and 32Gbyte using 4Gbit monolithic memory - i.e. WITHOUT using dual-die packaging.

        In addition, NLST HyperCloud can run at the full 1333MHz speed.

        Plus the power efficiency.

        The cost efficiency - doesn't require dual-die. NLST uses it's "Planar-X" technology to do this - while others may try to use "dual-die" (which NLST says has weaknesses):

        ----
        http://www.theregister.co.uk/2009/11/11/netlist_hypercloud_memory/
        Netlist goes virtual and dense with server memory
        So much for that Cisco UCS memory advantage
        By Timothy Prickett Morgan
        Posted in Servers, 11th November 2009 18:01 GMT
        ...
        The company also developed a memory packaging technology called Planar-X, which allows for two PCBs loaded with memory chips to be packaged together relatively inexpensively to share a single memory slot. This technique is cheaper and more reliable, according to Duran, than some of the dual-die packaging techniques memory module makers use to make dense memory cards out of low density and cheaper memory chips.
        ...
        Using the Planar-X double-board designs, Netlist can take 1Gb memory chips and make an 8GB memory module that costs only 20 to 30 per cent more than a standard 4GB module using 1Gb chips; using 2Gb chips, it can make a 16GB module, something no one else can do yet.
        ----




        Thus in conclusion it would seem that Samsung 32GB, even using the most advanced 4Gbit memory dies (in dual-die packaging to make 4Gbit x 2 = 8Gbit package) and using the smaller 40nm process to reduce power consumption:

        - does not achieve similar speedup (1066MHz only compared to NLST 1333MHz)
        - uses more expensive dual-die packaging, while NLST can use the cheaper monolithic packaging (single die per package) to achieve the same or better results

    • http://www.earthtimes.org/articles/show/samsung-first-to-begin-shipping,1223721.shtml
      Samsung First to Begin Shipping 40nm-class, 32-Gigabyte Memory Module for Server Applications
      Posted : Mon, 29 Mar 2010 02:50:40 GMT
      Author : Samsung Semiconductor, Inc.
      quote:
      In addition, replacing 12 DRAM modules of 16GB density with just six 32GB modules would achieve a 192GB total density, while allowing the DRAM operating speed in a two-way server system to rise by 33 percent from 800 megabit per second (Mbps) to 1,066Mbps, as power is cut by 40 percent.


      So you can see here that 12 slots x 16GB memory modules (non-NLST) only yields the slower 800MHz (speed slowdown we have been talking about).

      Article is claiming that using their 4Gbit memory chips-based modules you can get a modest improvement to 1066MHz.

      Article suggests Samsung is able to achieve this modest improvement because newer 40nm semiconductor process has better power efficiency.


      However, with NLST HyperCloud you could get the full 1333MHz speed (and that is with older and cheaper 2Gbit memory chips).


      Note how Samsung has NOT indicated what the speed would be with 12 slots x 32GB - maybe they fall down to 800MHz at that memory loading.

      This would superficially suggest that these Samsung 32GB modules may not be leveraging NLST IP, but are leveraging improvements from smaller 40nm lithographic process.



      So NLST HyperCloud would retain the advantage of higher speed (1333MHz), plus lower cost (since they can use cheaper 2Gbit memory chips).

      Later, when 8Gbit memory chips appear (and 4Gbit memory chips fall in price), NLST could start using 4GBit memory chips for even better performance.

      • 1 Reply to vicl2010v
      • Another reason Samsung is able to do 32GB using 4Gbit memory chips maybe because they are "dual-die" memory chips.

        Supposedly it would not be possible to do this conventionally (without NLST IP) with single-die memory chips.

        Single-die means you have one memory chip inside the packaging.
        Dual-die means you have two memory chips packaged inside one package (so it looks like one bigger chip) - but you may save some space.


        The article on NLST below suggests NLST is able to achieve:
        - 8GB using 1Gbit memory chips
        - 16GB using 2Gbit memory chips
        - (and probably 32GB using 4Gbit memory chips)

        using conventional single-die memory chips (thanks to NLST's Planar-X tech):

        http://www.theregister.co.uk/2009/11/11/netlist_hypercloud_memory/
        Netlist goes virtual and dense with server memory
        So much for that Cisco UCS memory advantage
        By Timothy Prickett Morgan
        Posted in Servers, 11th November 2009 18:01 GMT
        ...
        The company also developed a memory packaging technology called Planar-X, which allows for two PCBs loaded with memory chips to be packaged together relatively inexpensively to share a single memory slot. This technique is cheaper and more reliable, according to Duran, than some of the dual-die packaging techniques memory module makers use to make dense memory cards out of low density and cheaper memory chips.
        ...
        Using the Planar-X double-board designs, Netlist can take 1Gb memory chips and make an 8GB memory module that costs only 20 to 30 per cent more than a standard 4GB module using 1Gb chips; using 2Gb chips, it can make a 16GB module, something no one else can do yet.

    • From the first two posts in this thread about Needham presentation:
      http://messages.finance.yahoo.com/Stocks_%28A_to_Z%29/Stocks_N/threadview?m=te&bn=51443&tid=12087&mid=12087&tof=5&frt=2
      Re: NLST Needham presentation Jan 14, 2010 - part1

      It is clear that NLST HyperCloud 16GB memory modules are made using 2Gbit memory chips. Others have to use 4Gbits memory chips (which are more than 2x more expensive) to make 16GB.

      So how is it that Samsung is saying:
      - 16GB modules being made using 2Gbit memory chips
      - and 32GB memory modules using 4Gbit memory chips

      If they were earlier making the 16GB memory modules thanks to MetaRAM, or Inphi and are using the same process to create 32GB (i.e. use 4Gbit for those), are they still in relationship with Inphi ?

      Did they sell any 16GB modules using 2Gbit memory chips ?

      Did they use MetaRAM buffer chips or Inphi buffer chips ?

      MetaRAM says it sold $37,000 worth but "none were used in commercial sales" in NLST/MetaRAM court dockets.

      Inphi says "In addition, no iMB product has been sold." in NLST/Inphi court docket 19.

      So what is happening here - or is GOOG somehow implicated if it had subcontracted buffer chip manufacture to third parties (as court documents suggest):

      http://messages.finance.yahoo.com/Stocks_%28A_to_Z%29/Stocks_N/threadview?m=te&bn=51443&tid=14733&mid=14813&tof=1&frt=2#14813
      Re: update on the various court cases 2 .. GOOG infringing part1/2


      So NLST has so far isolated their legal challenges to the big players (GOOG) who were acting as instigators, or the smaller companies who held IP (MetaRAM), or the component makers (Inphi) but hold no IP.

      But would Samsung be continuing that violation of NLST IP even now after MetaRAM demise ?

      • 1 Reply to vicl2010v
      • part 2/2 ..


        http://www.computerworld.com/s/article/9174330/Samsung_ships_32GB_DRAM_server_memory_module
        Samsung ships 32GB DRAM server memory module
        New 40nm modules boosts dual-CPU storage capacity to 394GB, doubling previous capacity
        By Lucas Mearian
        March 29, 2010 01:11 PM ET
        ...
        The module doubles the 16GB capacity of Samsung's previous high-end module, which used 2Gbit chips based on 50 nanometer (nm) lithography technology.

        The new module is based on 4Gbit DDR3 chips using Samsung's latest 40nm lithography technology. Samsung had started using the 40nm process to produce memory products last July.


        Last year's Samsung announcement:

        http://techreport.com/discussions.x/16600
        Samsung ships 16GB DDR3 memory modules
        by Cyril Kowaliski — 6:00 AM on March 19, 2009
        ...
        Samsung claims these are the "world's first and smallest high-density memory modules" to use 2Gb, 50nm DRAM chips, which are packed in dual-die configurations to allow for 16GB registered DIMMs. The modules themselves are rated for operation at 1066Mbps (or 1066MHz), and Samsung says it's the first to offer 16GB modules that can operate at 1.35V. It's not entirely clear whether the modules can actually hit their top speed at 1.35V, though.




        Here are some PRs from the past from memory module makers using MetaRAM (which has since conceded to NLST):

        http://www.digitimes.com/news/a20080820PR200.html
        Hynix demonstrates DDR3 R-DIMM using MetaRAM technology at IDF
        Press release, August 20; Esther Lam, DIGITIMES [Wednesday 20 August 2008]

        Hynix using MetaRAM "chipset" - MetaRAM memory module has Hynix logo on it (page 10):
        http://www.ansoft.com/ie/Track2/DDR3%20Memory%20Module%20Design.pdf

        STEC - was using MetaRAM
        http://www.smartmodular.com/binary/files/8GBDRAMMetaram.pdf

        SMOD was also using MetaRAM "chipset":
        http://www.epn-online.com/page/new56803/smart-launches-8gb-dual-rank-ddr2-rdimms.html
        SMART launches 8GB dual-rank DDR2 RDIMMs
        04/03/2008
        ...
        The new module combines SMART's new DDR2 packaging technologies with the MetaRAM chipset architecture.


        http://www.theregister.co.uk/2008/02/25/weber_metaram/
        MetaRAM double stuffs servers with memory
        256GB box for $500k $50k
        By Ashlee Vance in Mountain View • Get more from this author
        Posted in Servers, 25th February 2008 20:25 GMT

        Best of all, the likes of Hynix and Smart Modular - two early MetaRAM customers - can double-stuff their hardware without shifting over to pricier DRAM chips.
        ...
        Smart Modular has agreed to offer an 8GB DIMM for $1,500 that includes the MetaRAM chipset. That's waaay below the $5,000 that the device would usually cost. As mentioned, Hynix will sell product as well.
        ...
        TSMC makes the chips for MetaRAM, which charges about $200 for 8GB DIMM parts and $450 for 16GB DIMM parts. Both Intel, which has invested in MetaRAM, and AMD are backing the technology, and server makers such as Rackable Systems, Appro and Verari are expected to begin offering the double-stuffed memory soon.

    • bump!! 3

    • bump!! 6

    • bump!! 4

    • View More Messages
 
NLST
1.70-0.03(-1.73%)Aug 26 4:00 PMEDT