Rackmounts Coming

1356

Comments

  • Reply 41 of 114
    macintoshmacintosh Posts: 22member
    I will be in AI chat at 10:45 PM tonight. Here we go.
  • Reply 42 of 114
    matsumatsu Posts: 6,558member




    when you look at it DDR is really the best thing out there for system RAM. DDR266 (PC2100) is the best current price performance balance. It kinda sucks that after the LONG wait Apple would choose 266 and not 333 (which is out there, if not wide spread) or even 400 (which has scary big throughput, but isn't on anything, yet)



    RD-RAM for all it's 400Mhz system bus is not really that fast. It needs interleaving, and it's only 16bits wide. 533 is only marginally faster x2 (for interleaving) to 1066. Sounds impressive but still only 16bit and about the same throughput as PC 2100 (DDR266)



    Seems to me that DDR 333 (PC2700) is the fastest thing you can find in reasonable quantity and prices at the moment.



    DDR266 is horribly late (on Apple's part) but if it arrives it isn't at all behind what's currently available.
  • Reply 43 of 114
    nostradamusnostradamus Posts: 397member
    [quote]Originally posted by fellow722:

    <strong>Guess who aslso predicted Rackmounts? Last night, in the chat room. Here we go...</strong><hr></blockquote>



    Wow, big accomplishment. I see you read ThinkSecret since they've had an article predicting rackmounts at WWDC since May 1.



    <a href="http://www.thinksecret.com/features/rackmount2.html"; target="_blank">http://www.thinksecret.com/features/rackmount2.html</a>;



    [ 05-06-2002: Message edited by: Nostradamus ]</p>
  • Reply 44 of 114
    bodhibodhi Posts: 1,424member
    JD -



    Customers for Rackmounts and customers for Power Macs are completely and entirely seperate. My friend I speak of at the beginning of this thread has sold Power Macs direct from Apple with no graphics card, no modem, no ram, and no hard drive. These things are not important to this market.
  • Reply 45 of 114
    zozo Posts: 3,117member
    what I'm curious about is if there will be requirements to buy them



    Will you have to buy a minimum number of them? Can any Joe go into an Apple Store and buy one?



    Can you just take a rack-mount server and put it on your desk if you wanted? Use it as a normal desktop?
  • Reply 46 of 114
    spotbugspotbug Posts: 361member
    What about the possibility that Apple is announcing all new PMs on the 14th?



    Jobs mentioned the rackmounts at the keynote because of the audience, but the official announcment (with specs and pics) has to wait until the unveiling of all the new PMs.



    What do we think? It's just a thought that flew through my head. Now that I've typed it out, it sounds really unlikely. But that "Add Reply" button is so tempting! Argh!!! Wicked, wicked "Add Reply" button!!
  • Reply 47 of 114
    othelloothello Posts: 1,054member
    [quote]Originally posted by ZO:

    <strong>Can you just take a rack-mount server and put it on your desk if you wanted? Use it as a normal desktop?</strong><hr></blockquote>



    i was thinking about this last night.



    if i had a usb/firewire hub on my desktop, somehow connected to the rack server, plus the monitor that would do me. i use the cd drive every now but could use my firewire burner for that.



    then my machine could sit out of the way in a rack (perhaps with the rest of the machines in the studio...)



    hang on thats a crap idea! <img src="graemlins/oyvey.gif" border="0" alt="[No]" />
  • Reply 48 of 114
    razzfazzrazzfazz Posts: 728member
    [quote]Originally posted by cdhostage:

    <strong>If Apple wants to make a serious server, it needs to give it seriouis power.



    I don't think two 1.2 GHz G4s can cut it.

    </strong><hr></blockquote>



    Actually, servers are usually much more liekly limited by available I/O bandwidth than by available raw processing power.



    Also, concerning the whole renderfarm theory: Usually, machines in a cluster are not referred to as "servers" (unless the whole cluster's purpose is to be some sort of a load-balanced server), and as SJ obviously specifically said "dedicated server hardware", I think this pretty much indicates that clusters or renderfarms are NOT the primary field of use those rackmounts are intended for.



    Bye,

    RazzFazz
  • Reply 49 of 114
    crusadercrusader Posts: 1,129member
    There probably will be a set of requirements a person/organization will have to meet to buy a rackmount, but probably only verification a person has enough $$$ to pay for one.
  • Reply 50 of 114
    razzfazzrazzfazz Posts: 728member
    [quote]Originally posted by Matsu:

    <strong>



    when you look at it DDR is really the best thing out there for system RAM.

    </strong><hr></blockquote>



    Tthere's a little more to memory than just raw theoretical (!) peak bandwidth.





    [quote]<strong>It kinda sucks that after the LONG wait Apple would choose 266 and not 333 (which is out there, if not wide spread) or even 400 (which has scary big throughput, but isn't on anything, yet)</strong><hr></blockquote>



    Neither PC2700 nor the 400MHz variant have been formally standardized yet, as far as I know.





    [quote]<strong>RD-RAM for all it's 400Mhz system bus is not really that fast. It needs interleaving, and it's only 16bits wide. 533 is only marginally faster x2 (for interleaving) to 1066. Sounds impressive but still only 16bit and about the same throughput as PC 2100 (DDR266)

    </strong><hr></blockquote>



    The point is that, being only 16 bits wide, it is much easier to interleave multiple rambus channels than it is to interleave multiple DDR channels. Also, not that dual channel PC1066 has about 4.2GB/s theoretical peak bandwidth - not quite "about the same as PC2100".



    Bye,

    RazzFazz
  • Reply 51 of 114
    daveleedavelee Posts: 245member
    I thought that PC2700 had been approved by JEDEC (or whatever the body is).



    You can certainly buy modules from Micron (Crucial).
  • Reply 52 of 114
    g-newsg-news Posts: 1,107member
    You can also buy DDR.3200 modules now, from Samsung, in 128 and 256MB sizes, 512 coming a little later this year.



    The main problem is that, while DDR333 is already fairly supported, there's no chipset I know of that supports DDR 400 atm.



    G-news
  • Reply 53 of 114
    matsumatsu Posts: 6,558member
    I could be wrong (it has happened before) so someone please explain this to me, I must be missing something 'cause the numbers don't add up.



    take DDR variants



    DDR266 x 64bits = 17024



    RD-RAM 533 x 16bits x 2 (interleaved channels) = 17056



    That looks 'about the same' to me.



    But I could still be off. Exactly how many RDRAM channels are commonly interleaved? I was under the impression that 2 was by far the most common number.



    AFAIK, both DDR266 and RD-RAM 533 use a 133Mhz 'bus'. DDR gets it's double data rate by having two signals (one rising and one falling) on each cycle. RD-RAM creates 4 signals per clock by having the same, one rising one falling, arrangement as DDR but with two waves 90 degrees out of phase -- about the same thing as DDR II (think of it as QDR, quad data-rate) will do.



    Someone who undertands this better than me please explain. How does RD-RAM get a peak data-rate of 4200MBps ???
  • Reply 54 of 114
    g-newsg-news Posts: 1,107member
    BTw if you go over to <a href="http://www.tomshardware.com"; target="_blank">www.tomshardware.com</a> you'll find a huge article comparing all memory types to each other.

    it's clear that while RDRAM is the fastest, a good CL2 DDR266 can beat a CL3 333 or even 400 module.

    So if Apple "only" delivers DDR266 but go for CL2, it's fine with me.

    Of course DDR-333 with CL 2.5 or evne 2 would be even better.



    G-news
  • Reply 55 of 114
    g-newsg-news Posts: 1,107member
    Quite simple:

    RDRAM with 1066MHz (that's on a 533MHz bus, which is essentially a 4x133 bus) is currently 16 bits wide (32bit coming), interleaved, the performance doubles theoretically, gives:



    1066MHz x16 bits = 17056 mbits/sec

    x2 for interleaving = 34112 mbits/sec

    /8 for bytes = 4264 MB/sec = 4.2GB sec



    And for DDR:

    200MHz x 64 bit = 12800 mbit/sec

    x2 for DDR400 = 25600mbit/sec

    /8 for bytes = 3200 MB /sec = 3.2GB /sec



    That will lead to the conclusion, that the maximum DDR you can get today maxes out at a theoretical 3.2GB /sec, while PC1200 RDRAM would max out at 9.6GB/sec (that's for teh 32bit flavor, 16 bit would be 4.8GB/sec)...but there are theoretial values and reality is much more positive for DDR-SDRAM than the numbers would suggest.



    G-News



    Ps: the mistake you made was assuming that RDRAM runs at 400, resp. 533MHz, however that is the bus that runs at those (ie 4x100 and 4x133MHz), the RDRAM itself runs at twice the "speed".



    [ 05-07-2002: Message edited by: G-News ]



    [ 05-07-2002: Message edited by: G-News ]</p>
  • Reply 55 of 114
    andersanders Posts: 6,523member
    [quote]Originally posted by There is no g5:

    <strong><a href="http://www.gvstore.com/gvs90apg4wor4.html"; target="_blank">http://www.gvstore.com/gvs90apg4wor4.html</a>;



    maybe Apple is just going to slap a silver Apple logo on these.....



    Jet</strong><hr></blockquote>



    I predict that thats exactly what they are gonna do. Look at the specs of that thing. If you add the ram and extra HD to the PM G4 in the Apple Store the price is the same, which means that they must get them from Apple sans case. So they coorporate with Apple already now and how much could it cost Apple to buy the firm and add "rack mounted" to their list of products?



    Think Limited.
  • Reply 57 of 114
    matsumatsu Posts: 6,558member
    Oh my poor little brain. I need a drink, is it too early for champagne and OJ ??? Does that make me gay? Anyway, I almost get it now.



    the RD-RAM itself runs at twice the speed of the bus. The bus itself is a quad-pumped affair. So 1066 is not 1066 because of interleaving, per se, it actaully runs at twice the speed internally, but transfers data over a 533Mhz bus?



    Wouldn't the bus limit how much data can go between the RAM chips and the CPU?



    OI, I'm confused.
  • Reply 58 of 114
    g-newsg-news Posts: 1,107member
    Here's a correct version of your scoreboard.

    Not very well done, but it was quick







    G_News
  • Reply 59 of 114
    g-newsg-news Posts: 1,107member
    I can't tell you how exactly RDRAM works, in fact only RAMBUS can, I guess, but that's how it is.

    That's also why RD-RAM modules need heatsinks...at [crossout]800MHz+[/crossout] things get hot.



    edit: ok at 400MHz+ then



    G-News



    [ 05-07-2002: Message edited by: G-News ]</p>
  • Reply 60 of 114
    razzfazzrazzfazz Posts: 728member
    [quote]Originally posted by Matsu:

    <strong>the RD-RAM itself runs at twice the speed of the bus. The bus itself is a quad-pumped affair. So 1066 is not 1066 because of interleaving, per se, it actaully runs at twice the speed internally, but transfers data over a 533Mhz bus?



    Wouldn't the bus limit how much data can go between the RAM chips and the CPU?

    </strong><hr></blockquote>



    Actually, it goes like this:



    The new P4s feature a 133MHz FSB, which is then quad-pumped to effective 533MHz. It is 64 bits wide.



    Those RDRAM modules are internally clocked at 533MHz, double-pumped to effective 1066MHz. Two 16-bit-wide channels are then interleaved.



    133x4x64=533x2x2x16



    Bye,

    RazzFazz



    EDIT:

    Tom's Hardware has a comparison of DDR and RD-RAM. Go <a href="http://www.tomshardware.com/mainboard/02q1/020318/i845ddr-01.html"; target="_blank">here</a> and scroll down for a table that sums it all up pretty nicely.



    [ 05-07-2002: Message edited by: RazzFazz ]</p>
Sign In or Register to comment.