Please create an account to participate in the Slashdot moderation system

 


Forgot your password?
Close
typodupeerror
×

Supermicro Announces Quad-Opteron 1U Motherboard 158

Posted by CowboyNeal from the tiny-powerhouses dept.
hpcanswers writes "Supermicro, a producer of systems for the high-performance computing market, has announced a 1U-sized quad Opteron motherboard for the OEM market. The product, which is on display at CeBIT this week, supports both HyperTransport and PCI Express. It also consumes 1000 watts of power. Supermicro's announcement is all the more interesting because the company has historically only supported Intel processors."
This discussion has been archived. No new comments can be posted.

Supermicro Announces Quad-Opteron 1U Motherboard More

Supermicro Announces Quad-Opteron 1U Motherboard

Comments Filter:

  • Historically huh? (Score:4, Informative)

    by Anonymous Coward on Saturday March 11, 2006 @12:47PM (#14898493)
    Funny, I've got 20 1U and 2U supermicro opteron servers. Are you sure you researched this statement?
  • Isn't that equal to 1 Kilowatt? First we have ONE TRILLION BYTES, and now ONE THOUSAND WATTS! One million shames on you, Slashdot.

  • Space heater (Score:4, Informative)

    by iamplupp ( 728943 ) on Saturday March 11, 2006 @12:50PM (#14898503) Homepage
    "It also consumes 1000 watts of power" 1000W seems a little high... Four dual core opterons doesent need half of that! Even less for single core. The article suggest using 55W opterons.

    • Re:Space heater (Score:4, Informative)

      by CrayHill ( 703411 ) on Saturday March 11, 2006 @01:03PM (#14898551)
      As per TFA, they use a 1000-watt power supply. It does not consume a 1000 watts of power.
      • DOH.
        And if it doesnt consume 1KW, why does it need a 1KW psu? 20% or so headroom are sensible, of course, but 4 opterons dont even burn 300W in worst case situations...

        • Re:Space heater (Score:5, Informative)

          by Homology ( 639438 ) on Saturday March 11, 2006 @01:17PM (#14898602)
          Few power supplies much better than 80% efficiency, so with a 1KW PSU you can expect 800 W that is usable. The Opteron 870 is rated as max 95W, so four of them gives approximately 400W. A few SCSI hard disks that may use as much as 30W at startup, much memory, lots of cooling, and whatever the motherboard itself consumes, and we have easily another 100W. While a 1KW PSU seems much, it does not seem excessive.

          • A few SCSI hard disks that may use as much as 30W at startup

            True enough. To avoid the power draw of a handful of drives starting at once, most (all?) mid-to-high performance RAID systems offer staggered spinup on drives.
          • Startup load isn't an issue with SCSI hard drives. Any half-decent SCSI controller can be programmed to spin drives up one at a time. From what I understand, this is the default behavior for most RAID controllers.

            Parallel ATA does not have this capability (drives spin up the instant the power is turned on). I don't know how SATA handles this....

            (For the uninformed, hard drives require a significantly higer amount of power to initially turn on than they need to operate once the platters are spinning. A h
            • The original MacPortables (including backlit) could not power up without their battery for this very reason. The 1.5A power supply wasn't enough to initially spin the hard disk. If you used a later 2.0A or 2.5A powerbooko supply, they could start without the battery.

              hawk, who still has his
          • the 1000W are on the output side, not the input one...

          • Re:Space heater (Score:4, Informative)

            by evilviper ( 135110 ) on Saturday March 11, 2006 @10:43PM (#14900724) Journal
            BAD MODS! NO COOKIE!

            Few power supplies much better than 80% efficiency, so with a 1KW PSU you can expect 800 W that is usable.

            No, NO, NO, NO, NO! It doesn't matter if you have a 10% effecient power supply, or a 100% effecient power supply. A 1000WATT power supply will OUTPUT 1000WATTS. The difference in effeciency is how much INPUT power it will need to do that, and how much waste heat it will produce in the process.

            Power supplies are not, and have never been, rated by their INPUT. That would be just stupid, as it would seriously penalize those companies that make more energy-effecient units.

        • Re:Space heater (Score:2, Interesting)

          by Anonymous Coward
          Probably because they just happen to be manufactured that way now. Remarkable leaps forward in switched mode semiconductor PSUs lately. Take a look at this baby Mini ITX PSU [mini-itx.com]
          Just because this device says it will deliver 1KW at peak I don't expect that was a design parameter. Most interesting is that it fits into a 1U profile, I expect this is a tiny little beast.

          Problem is this: There's a growing drive here in Europe towards "green" computing. A well designed micro-cluster of nano form factor boards can be

        • Re:Space heater (Score:3, Interesting)

          by mihalis ( 28146 )
          f it doesnt consume 1KW, why does it need a 1KW psu

          My guess is the designers want to ensure excellent peak current capability. Maybe when the board switches on the transient power draw troubles lesser power supplies that nominally cover the steady state power demand. Certainly if what I've read about hi-fi amps is at all representative you are better off with plenty of headroom when trying to drive speakers with a spiky signal (that is, music), and so it seems to me a server would enjoy better reliability

    • Re:Space heater (Score:3, Interesting)

      by Courageous ( 228506 )
      For server solutions like these, typically the power supply is REDUNDANT. I.e., it's two of 'em, with a failover capability. This is very common, particularly in rackmounts.

      C//

    • You're a moron (Score:3, Informative)

      by Dr. Spork ( 142693 )
      If you read the article, you'd find that each of the Opterons in it will use 55 watts. Since you're stupid, I won't take any chances and do the calculation for you: 55 X 4 = 220 Watts of power (plus a bit extra for running the chipset and drives). That's not bad for what is essentially 8 cpu's! The recommended (read the spec sheet) 1000 watt power supply is there to insure perfectly stable voltages. It would be overkill for anything except applications where maximum uptime is essential (like web serving).

      F

      • For the sake of comparison, I run a 53 Watt PSU on this system, which draws about 12 Watts from the wall. Yaay * 8 for cool-running VIAs!
        • Ah but that's the trick, what takes me 2 minutes on an AMD64 takes you 17 on a VIA box. And I know as I run a 1.2Ghz C3.

          VIAs are good for only a few things

          1. Text processing [e.g. email, web, usenet]
          2. Low power NAT/dhcp/dns server for a house [which is what I use mine for]

          I have Gentoo Linux installed on mine and my AMDX2 while taking 9 times the power [or TDP] can usually finish emerges in a very small fraction of the time the C3 takes.

          This is like the Crusoe vs. Rest debate. Their CPUs would run with
          • You should setup a cross-compiler on the AMD box, and build binary packages for your C3. Then you can just scp them over, and not spend an extemely long time building software on the C3.

            It takes a little bit of time to get setup, but it quickly pays for itself. I do the same thing here, I have a SH3 box that I build packages for on this dual Xeon. If I tried to build them on that box, it would probably take weeks to build the system.

            emerge crossdev, and read the docs :)
            • or just use distcc and put -m32 in your CFLAGS. :-)

              Point is I don't really have a lot installed so doing a weekly [re: monthly] update isn't a significant problem.

              Mostly my point was even though the C3 lacks ALU horsepower it's more than enough to act as a highly flexible NAT/router/etc box. Certainly more flexible than say a $100 linksys router and only costs slightly more anyways.

              Tom

    • Re:Space heater (Score:5, Informative)

      by ivan256 ( 17499 ) * on Saturday March 11, 2006 @02:01PM (#14898763)
      Even dual processor 1U Xeon boxes typically have dual-redundant 1000W power supplies.

      You never know what an enterprise user is going to stick in the expansion slots.
      • If you mean with "dual redundant 1000W" 2 redundant 500W PSUs, then i agree.
        If you mean 2 1000W psus, then you are posting bullshit.

        • I mean two 1000 watt supplies, though 750's are common too. I have yet to see a name brand 1U dual Xeon server with 500's. By name brand, I mean Intel whitebox, IBM, Dell, etc... The Supermicro 1U dual Xeons use two 750's inserted from the back. The Intel whitebox servers (rebranded by tons of places) use dual 1000's, one in the front and the other in the rear, but they may offer 750's as an option.

          Just because there's a 1000 watt supply in there doesn't mean the machine is pulling 1000 watts. Without any a
    • 1000W seems a little high

      Hitting the 1000W mark allows them to use cool
      new terms like '1KW'. It sounds more hip than 1000W.
      Just like chip technologies, where 130nm sounds cooler
      than .13 micron.

      And what is with this term 'bubbler'? In my day we
      called it a bong. First time I heard this I expressed dismay,
      but the young people just laughed.

      Kids today! If it wasn't holding up my onions,
      I would use my belt to smack some sense into them.

    • You do realize more than just the processors require power.
      You have
      • Fans(many in servers)
      • Hard Drives
      • RAID Controllers
      • Fiber Card(possibly)
      • NICs
      • CD/DVD ROM
      • USB Ports may be required to power something plugged in

  • It all makes sense (Score:5, Funny)

    by Ancil ( 622971 ) on Saturday March 11, 2006 @12:52PM (#14898510)
    the company has historically only supported Intel processors.
    That explains it. They probably have a whole bin of 1,000-watt power supplies sitting around they need to get rid of.

    • The back story on this....

      "Tom."

      "Yah, George?"

      "I just got our quarterty inventory report--we've got TEN extra pallets of 1,000-watt power supplies! I'm steamin' mad about this--I want them priced to MOVE!!"

      "Well... uh... I don't think there's much dema.."

      "I don't give a rats tail about what you think. Do whatever it takes to move those power supplies!!"

      In news today, Supermicro announces a quad Operton motherboard for 1U chassis...."

  • Intel's dominance at play here (Score:5, Informative)

    by hirschma ( 187820 ) on Saturday March 11, 2006 @12:54PM (#14898518)
    Supermicro has offered AMD solutions [supermicro.com] for a quite while now - just not under their "main" brand name. If you don't know that their Aplus products exist, you won't find them. Although I'm sure no one would go on record, I'd wager that Intel has pressured a heavily Intel-dependent vendor to not promote AMD's product.

    In fact, go to SuperMicro's home page [supermicro.com], and you'll notice no mention or links to their AMD based products.

    This isn't the first time that this has happened. When AMD first shipped the Athlon, very few board makers dared to ship Athlon solutions [tomshardware.com] for fear of Intel shorting them on chipsets. I recall, but cannot substantiate, that Asus and Abit first shipped Athlon boards under a "shadow brand", much as Supermicro is doing here.

    I, for one, cannot wait to buy some of the Supermicro^h^h^h^h^h^h^h^h^h^h, um, Aplus gear.

  • FINALLY. (Score:5, Funny)

    by blair1q ( 305137 ) on Saturday March 11, 2006 @12:58PM (#14898532) Journal
    Finally.

    A pizza box that will actually cook your pizza.

  • I liked SuperMicro... (Score:4, Interesting)

    by (H)elix1 ( 231155 ) <slashdot.helix@nOSPaM.gmail.com> on Saturday March 11, 2006 @01:26PM (#14898641) Homepage Journal
    I've still got some SBU and DBU mainboards still chugging away. The problem was I made the switch to AMD and SuperMicro waited years to make the transition. To bad because I liked their kit and in the PII/III era, and Asus, Abit, and SuperMicro were solid recommendations to friends and family. They sort of fell off the recommendation list because of the preference of the AMD CPU's. Now that they are working back into the workstation/server market, I'll definitely take a look the next time I do a system update.

    Yes, they had a white box label that did AMD stuff. Whatever. I'm glad to see the 'pro' brand get with the program.

  • how about 16 cores? (Score:4, Informative)

    by Ankou ( 261125 ) on Saturday March 11, 2006 @01:39PM (#14898686)
    Maby you cant fit it in 1u, but Tyan makes the K8QW that supports 8 Opterons with the M4881 add on processor board. Meaning you got 16 cores of pure powa. Go ahead, compile the internet. [tyan.com] I'd be interested in knowing if there was anything higher than that.

  • Is this Google's new brain? (Score:3, Interesting)

    by Dr. Spork ( 142693 ) on Saturday March 11, 2006 @01:43PM (#14898697)
    A Supermicro source rushed to claim some of the glory tied to rumors that Google has become a large Opteron shop.

    "There's some truth to the rumors" about Supermicro supplying gear to Google, the source said. "It was happy days around here."

    This is pretty slick hardware, and given Google's recent and complete switch to AMD, this seems like a good match. Opterons are just awesome, even I'm seriously thinking of buying a 165 (slobber...).

    On an unrelated note, did you notice the chipset is made by NVidia? Wow, they've come a long way! I'm impressed to see this kind of iron from a company that used to live off pimply gamers. With PCI-X supported, I can't help but wonder about what framerates one could get in a properly multithreaded game.

    • PCI-X and PCI-Express are not the same thing ;-)
      • Aah, oops, my bad. Thank you for the correction. I meant PCI-Express, though I noticed that the motherboard only supports 8X while most graphics cards want 16. And strangely, the intergrated video is ATi even though the chipset is NVidia. This thing is really some sort of chimera!
        • Most graphics cards want 16, unless they are in SLI mode, then they need 8x each. Interestingly this motherboard [supermicro.com] has two 8x PCI-Express slots (as well as 2 PCI-X slots), but still has the ATI onboard video. If only the two 8x slots were capable of SLI, this thing could be the ultimate motherboard for very rich gamers. With a multithreaded game and 8 cores, and 32 GB of RAM, the CPU and memory sure won't be the bottleneck in the system.

        • I noticed that the motherboard only supports 8X while most graphics cards want 16. And strangely, the intergrated video is ATi even though the chipset is NVidia.

          This is a server motherboard, not a workstation board. On a server board, PCI Express x8 slots (4GB/sec) are meant for things like dual port 4X Infiniband cards, dual channel Ultra320 SCSI cards, and 2GB/sec Fibre Channel cards. On a server board, PCIe x8 slots are not meant for graphics cards because most servers only need very basic integrated

  • Much better use of space (Score:4, Funny)

    by mpcooke3 ( 306161 ) on Saturday March 11, 2006 @01:52PM (#14898733) Homepage
    This will allow for much better use of space in data center racks.

    For example I could take out 8x 1U Intel based Dell web servers and replace it with 1 8way supermicro Opteron machine.

    The Opteron server couple probably serve content faster than the 8 Intel based Dells plus with the additional 7u space I'll easily have enough space to put a saucepan on top to boil water for tea, thereby saving myself the 80p I normally spend in the data center vending machine.
    • No, swapping 8 single units for one octal unit is usually a bad move for webservers. Individual operating systems can typically withstand far fewer than 1000 simultaneous connections, and when one component fails on a big, expensive, octal server, the whole thing is down and you lose 8 servers. Also, the amount of disk you can strap into an 8-way system is limited by the controllers, not the CPU speed, so for storage of actual web content, it can be very expensive or very inefficient unless you buy a great
      • I was actually being sarcastic but since you seem to have taken me seriously...

        It all depends on your web serving requirements, if you are big then the 8ways are better. You probably wouldn't actually remove older machines, but starting to use 8 way machines in a 1U slot does actually make sense if you are doing mass web serving. You want to cram in as much as possible since you are then maintaining less machines and have more cost effective use of space + more req/sec peak load handling per $$$. (of course

  • Off topic but... Why? (Score:5, Interesting)

    by nroose ( 738762 ) on Saturday March 11, 2006 @02:08PM (#14898796) Homepage
    Why does every server have it's own transformer/converter/rectifier/power supply (I think you know what I mean, that thing that converts from AC to DC)? It seems to me that it would make more sense for a room of servers to have a DC supply for all the computers.

    What's more, most server rooms have a power backup unit, which converts from AC to DC and back to AC again, just so that the computers can convert it back to DC. This is terribly inefficient just in terms of electricity, and it also creates a whole lot of heat, just so that we can air condition these rooms with huge air conditioners!

    It would just seem to make sense to me that the world of computing would come up with a standard for using DC, and then companies would build big power supplies that would offer redundancy, power backup, and current conditioning. It would save money, power and space.
    • For more information on this topic, see here [slashdot.org].
    • Because when AC and DC is involved, people get confused: It all goes wrong when a radio contestent is asked to spell AC/DC [ozbizz.com] and can't.

    • Re:Off topic but... Why? (Score:4, Informative)

      by Rick Genter ( 315800 ) <rick DOT genter AT gmail DOT com> on Saturday March 11, 2006 @04:24PM (#14899281) Homepage Journal
      It would just seem to make sense to me that the world of computing would come up with a standard for using DC, and then companies would build big power supplies that would offer redundancy, power backup, and current conditioning. It would save money, power and space.


      In fact, in the telco world, this is exactly how it works. The standard is to use -48vDC. Sun (among other manufacturers) makes servers that run directly off of DC [sun.com] (the Netra 120 on the referenced page).
    • AC servers cater to the lowest-common denominator, the guy that wants to plug it in on his desk.

      At 40kW though (really anything over 25kW in the US), a single/redundant -48V converter for the rack makes a lot of sense : you need a bunch of small 30A feeds like the Sun 20k's (12 per machine!).

      The trouble with 48V in high-power applications is that you need a lot of copper. A typical DC powerplant in the telecom world is good for a maximum of 400kW. That power plant is about 3,000 square feet (granted it i
    • Why does every server have it's own transformer/converter/rectifier/power supply (I think you know what I mean, that thing that converts from AC to DC)? It seems to me that it would make more sense for a room of servers to have a DC supply for all the computers.

      If the voltage weren't so low and the currents so high, this might be practical, but with the large currents typically flowing on a supply of 5 Volts or less, it wouldn't take much resistance in the cables and connectors to drop the voltage excessive

    • What's more, most server rooms have a power backup unit, which converts from AC to DC and back to AC again, just so that the computers can convert it back to DC. This is terribly inefficient just in terms of electricity, and it also creates a whole lot of heat, just so that we can air condition these rooms with huge air conditioners!

      That's just completely, totally, untrue. Converting from AC to DC and back causes nominal losses.

      It's the conversion from 120v to 12v, then up to 120v again that is the real pr

    • I have a VIA mini-itx board that I run as a Web, mail, and MySQL server in my basement. I got the board used, with the 12 volt dc power supply. I use one of these [astrodyne.com] with a small emergency lighting battery for a simple, efficient UPS. The board has a 1 GHZ processor, and draws on average about 3 amps at 12 volts. It should run for about an hour on backup; luckilly I haven't had to test that lately :).

  • Standby spare (Score:3, Interesting)

    by Spazmania ( 174582 ) on Saturday March 11, 2006 @02:32PM (#14898873) Homepage
    Also included: a standby spare so that when that much heat in that small a package burns itself out in six months you don't have to wait for a replacement. Supermicro is generally pretty good but packing a kilowatt of consumption into a 1U package is about as smart as running your home PC in the oven set on bake.

    Maybe I'm just sore because I've spent the last few weeks identifying the bad ram in last year's opteron rackmounts from Penguin. 2 gig ECC dimms and I'm seeing a 40% failure rate from multiple manufacturers. They stacked the damn chips one on top of another. There's no where for the heat to go. Of course they're going to fail.

    • Re:Standby spare (Score:3, Interesting)

      by evilviper ( 135110 )

      Supermicro is generally pretty good but packing a kilowatt of consumption into a 1U package is about as smart as running your home PC in the oven set on bake.

      A kilowatt in a 1U package is generally called a "Deluxe George Foreman Grill".

      If you pack the front and rear panels with several high-speed ear-piercing fans, you could disperse that kind of heat. You'll need to wear hearing protection when you go into your server room, but it's certainly possible.

      Still, these are the kinds of absolutely ridiculous s

    • 2 gig ECC dimms and I'm seeing a 40% failure rate from multiple manufacturers. They stacked the damn chips one on top of another. There's no where for the heat to go. Of course they're going to fail.

      Cpnversely, one of the things that impressed me about the physical design of Sun's SunBlade-1500 was a shroud over the DIMM's to promote cooling of said modules - the other end of the shroud went to the chassis exhaust fan. Weird that we're having to worry about thermal management in memory - though the the in

  • "Consumes" 1000 watts "of power" (Score:3, Interesting)

    by Ancient_Hacker ( 751168 ) on Saturday March 11, 2006 @03:13PM (#14899032)
    Hate to be pedantic, but this pizza-box doesnt "consume" 1000 watts "of power". In many ways:

    • The sticker on the power supply may say "1000 watts", but that's an upper limit of what the power supply can do when maxed out. I suspect in typical setups the draw will be much lower. Opterons only draw about 80 watts, so a quad of them will only be about 300 watts. You're only going to max out with a lot of memory and many 10K RPM disks.
    • The unit doesnt "consume" the power, it gets mostly (99%) gets converted into heat.
    • Saying "1000 watts" is enough. No need to say "of power", the only meaning of watts is as a measure of power. I suppose we should be glad they didnt use one of the many wrong units, like watt-hours, or watts per hour.
    • That's a lot of watts in a 1U height! Do they suggest mounting it high on the rack so you can use it as a hair dryer?

  • 1000 Watt power supply, NOT 1000 Watts of power (Score:3, Informative)

    by speedplane ( 552872 ) on Saturday March 11, 2006 @04:15PM (#14899256) Homepage

    "It also consumes 1000 watts of power."
    WRONG!
    The board requires a 1000 Watt power supply, not neccesarily 1000 watts of power. The power supply is the upper limit of how much the board can consume. Most computers come with a 300 watt power supply even though they normally use only about 100 Watts.
    That being said, this board probably consumes quite a bit of power (but much less than 1000 watts) if it needs such a heavy duty power supply.
  • Let's say your electricity costs you $0.10 per kilowatt-hour. This machine will cost nearly $75/month to operate:

    1000watts / 1000 = 1kW-hour to operate for one hour

    ( 1 kw-hour * 24hours per day * 365 days / 12 months ) * $0.10 == $73.00 / month

Slashdot Top Deals

Love may laugh at locksmiths, but he has a profound respect for money bags. -- Sidney Paternoster, "The Folly of the Wise"

Close