AMD power usage trumps Intel -- but does anyone care?

AMD power usage trumps Intel -- but does anyone care?

Summary: I absolutely agree with David Berlind regarding the importance of low-power servers. [How a chill-pill for your server room improves your bottom line] What I don't understand, however, (and haven't for quite a long time) is why the power consumption issue is getting attention now that Intel is talking about its next-gen low-power offerings.

TOPICS: Processors

I absolutely agree with David Berlind regarding the importance of low-power servers. [How a chill-pill for your server room improves your bottom line] What I don't understand, however, (and haven't for quite a long time) is why the power consumption issue is getting attention now that Intel is talking about its next-gen low-power offerings.

Check the actual CPU power consumption figures at Lost Circuits (measured directly off the CPU) or system-level consumption numbers from Tech Report and we see the same trend.

Intel Xeon, Pentium 4, and Pentium D cores consume far more power at idle (if C&Q is enabled on AMD) and at load, regardless, than their 90nm AMD counterparts. In fact, if you look back at older reviews, Opteron has been generally lauded as a much cooler, quieter CPU than its Nocona/Irwindale counterpart.

Opteron 252s have been available for over six months; low power and ultra-low-power versions of Opteron have been available even longer, and the advent of AMD's dual core demonstrates beyond a shadow of a doubt whose got the lower-power chip—the Opteron 152 system draws 201W at full load, as compared to the 840 at 292W. Michael Schuette's direct-CPU measurements at Lost Circuits show Toledo 4800+ drawing 80W at full load, versus 136W on 840.

Don't get me wrong; I realize 80W still doesn't compare to the 32W limit on some of Intel's upcoming server products--but the point is, AMD processors have been pounding Intel chips in terms of power usage for the last 12 months...and no one seems to have given a damn. My point isn't just that you haven't mentioned it here before, but that no one seems to have made mention of it, period.

Coincidence? Luck? Conspiracy theory? I don't know. But having worked with Nocona and high-end Prescott, and seen how dramatically better Opteron and Athlon 64 are in terms of power draw, I do think the press and industry (in general) should have been more cognizant of AMD's benefits in this area.

Joel Hruska is an IT writer in Louisville, Kentucky.

Topic: Processors

Kick off your day with ZDNet's daily email newsletter. It's the freshest tech news and opinion, served hot. Get it.


Log in or register to join the discussion
  • I care, and the Pentium M is Intel's secret weapon


    Nice work and thanks for that link which measures CPU power consumption. I'll have to see if they include Pentium M numbers.

    I do have a comment. You're absolutely right that Intel's Pentium 4/D/XEON architecture as a whole is one of the worst power hogs out there. They use more in idle than AMD chips use at burst. However, Intel holds the lead with the Pentium M chip which will "officially" start appearing on the desktop and server market soon. AMD has the Turion, but their specification for power consumption is a little higher.

    It is important to understand the significance of the Pentium M because a recent benchmark test on Tomshardware showed that an unleashed Pentium M operating at 2.56 GHz using conventional air cooling with no voltage modifications beat the fasted Intel and AMD desktop processors. The Pentium M architecture is the future for Intel and Pentium 4/D/XEON is a dead end because you just can?t keep adding these power hungry processors to multi-core chips.
    • Servers

      All well and good if you don't need server performance, but the slight power edge of the PM vanishes pretty quickly when you add in the memory controller (not required by the AMD chips).

      What really makes for a brick wall, though, is the lack of 64-bit capability in the PM. Probably OK for most corporate desktops, since even an old 600 MHz PIII is plenty of performance there, but lethal for a server.
      Yagotta B. Kidding
      • 64-bit for most things is moot

        There are very few things that need 64 bit today. Unless you're talking about a database server that needs to address more than 4 GBs of memory, there is really no benefit in 64 bit today.

        As for the memory controller, whow me the benchmarks that back up your claim?
        • Memory controller power

          [i]There are very few things that need 64 bit today.[/i]

          Ask Intel for their presentation to JEDEC on the subject of server performance and its dependence on the amount of system memory. Intel is serious enough about the need for massive DRAM that they pushed (hard!) for the Fully-Buffered DIMM spec now going out.

          [i]As for the memory controller, whow me the benchmarks that back up your claim?[/i]

          George, you have my e-mail via ZD. If you want to know how I know what memory controllers use in the way of power, I'll give you credentials.

          Otherwise, just accept that I work with DRAM controllers and know what they use. A DDR controller will burn several watts, although the exact amount varies based on too many factors to enumerate in a TalkBack. An on-CPU controller will burn less than an external one for several reasons including that the CPU version can go straight to cache with minimal buffering compared to a north-bridge type.

          You can get a baseline on the power for both a DRAM controller and the FSB by simple arithmetic. It's not trivial thanks to the need to terminate the transmission lines involved.
          Yagotta B. Kidding
  • I see where your mistake is...

    It's not full load that matters, it's average and practical loads that includes the effects of power management.

    When you look at that, your data sheet level arguments are watered down to an amateur analysis.

    Second: You are also dismissing price elastacity in the processors. If you're product is not quite the best (and Intel's is not), there is this thing called "priced cheaper" ! Surprisingly, if option B costs less, then X watts are maximum load AIN'T that important NO MOE.

    So again, your analysis needs some work before it's fully baked and sent to the spellchecker.
  • Meta-story

    The really interesting part is how the trade press treated power as a big issue in the Athlon v. PIII comparison, a non-issue when the P4 turned into a building heater, and now it's a big story again with the PM.

    You'd almost think that they were working for Intel.
    Yagotta B. Kidding
  • AMD preemtpive strike on INTEL

    Power does not matter that much for a single user with a single PC, but when you build rack servers, you face a big problem. Putting two Xeons each using 151 watt max is very stressful for 1U or 2U servers, you need to bigger power supply, better cooling, and lot more eletricity cost. One primary reason INTEL can't have dual core Xeon is power consumption. Dual core Xeons uses at least 200 watts, and it is insane to put 2 such processors into 1U or 2U chassis.

    INTEL will launch a dual core processor called Sossaman32 next year based on the Pnetium-M. It's 32 bit, because INTEL hasn't figured out how to copy AMD64 to Pentium-M yet.

    But, AMD seems to be always a few moves ahead. It silently launched the dual core Opteron 865HE and Opteron 860HE, which is at 55 watts. You can have 8 such babies for 16 way computing and total power is just 440 watts.

    This is not surprising because Turion64 MT is rated at 25 watts. So there is no problem for AMD to do dual core opteron64 at 50 watts. And we can see AMD can do quad-core opteron around below 100 watts.

    Keep in mind single core Xeon runs at max power of 151 watts.
  • Power usage is important

    With today's rising cost of electricity because of rising fuel prices, you pay to use the server and pay to cool it with AC. Let say you have a 1000 processors in your server farm. AMD sucks 55 watts the Xeon sucks 151 watts that is 96 watt difference x 1000 or 96kw per hour. At $.10 per 1kw that is $9.60 per hr x 24 x 365 = $84,096 per year. Now you pay to cool it. At about 1/10 that or $8409.60 or $92,505.60 total just for the difference in CPUs only. There is saving on CPU price and mainboards and productivity over the Xeon. I was sold on the Athlon for quite some time.

    • AMD power does not trump Intel sales...

      Your example is a poor one... any 1000 server farm will have been offered a deal by Intel that would far outweigh the notional power savings you mention. That's why Intel is still mashing AMD's butt. Furthermore, as soon as Intel heard AMD were developing AMD64 (another AMD technical lead) they said "Heck, let the little guy pay for it, then when we run with it we'll STILL make more sales and with no dev costs!"

      Any and all ways, Intel wins, always.
      • It isn't notional and you miss the point

        Those numbers are not in stone buddy. Those numbers were for comparison purposes only. Besides, the savings would be greater since the processors are not changed yearly. If I can buy a AMD processor cheaper then Intel so can that server farm. AMD will have to come up with a better counter offer. AMD powered machines are slightly cheaper than Intel machines from the big computer houses. That is what competition is all about.
        • No you miss the point AMD fanboy

          Intel hunts down leads and cherry picks the highest revenue customers targetting them with sweeteners that will undercut the point at which AMD can sell a processor without taking a loss. That's why they're being sued. That's what being a monopoly is all about. Technical "superiority" aint worth squat - the only superiority I know about is on the bottom line.
          • Listen here goofball

            If Intel can undercut AMD's price and still make a profit, them AMD doesn't have a legal leg to stand on. That is what competition is all about. You can only dump for so long before the share holders will scream at you. As long as they are making money, they will never be accused of dumping. I bet the suit will go nowhere.
  • 4 way Opteron server, total 145 watts cool!

    Rackable Systems' new C2002 server, winner of the Best Server award, draws an astoundingly low peak of 145 watts in a configuration featuring two dual-core 1.8 GHz AMD Opteron HE processors (four total cores), 8 GB DDR400 RAM and one terabyte of SATA storage leveraging Hitachi 500 GB hard drives, with a mere 500 BTU/hour of peak thermal output. Mounted back-to-back in Rackable Systems' cabinets, the C2002 provides the greatest possible efficiency in power consumption and heat evacuation.
  • IDF: INTEL advsetised for AMD

    INTEL is telling people to measure performance-per-watt: of course AMD is the winner, Opteron has higher performance and lower power usage.

    INTEl is doing free advertising for AMD...