Oompa's Profile

Member since April 10, 2008

  • Name

    Oompa Loompa

  • Location:

    United States of America

Favorite Files

Recent Posts

  1. Comment - AMD quad-core Opteron servers claim performance records

    (Apr 11, 2008 - 9:46 PM)

    That's partially my point - the systems being compared in the article are not really comparable. The x7350 has been available for close to a year now - why compare it with the latest and greatest Opteron 8356 which you cannot even order yet? And the SPEC numbers reported seem to be hand picked - take a look for yourself (http://www.spec.org/). It looks like the article was taken mostly from an AMD press release. I'm not against AMD, I'm against this type of misleading article. For those of us who work on very large scale systems (256 procs and up), it's a disservice to see this type of article. AMD chips have always done well on SPECfp - but that's not the only benchmark and it only tells you a small piece of the performance equation. For example, a 2-way Intel Xeon E5472 posts 114 base and 139 peak on CINT2006 RATES while the 2-way AMD Opteron 8356 only scores 90 base and 104 peak and the 4-way AMD Opteron 8356 posts 161 base and 185 peak. The 2-way to 2-way doesn't look good for AMD on anything but floating point performance. Sure, the 4-way AMD system has more throughput than the 2-way Intel system for a single board system - but it consumes nearly twice the wattage and costs about 140% more. For cluster and grid systems like those I work on, I could not cost-justify an Opteron 8356 system - period - not even in the dual-socket configurations (where it only delivers similar or lower performance on average compared to the cheaper Xeon system).

    I wish the situation were different - but AMD really needs to do more than just push quad socket systems versus dual socket systems. For a workstation (and if cost is no object), by all means - get a 4-way Opteron system. But for large scale server systems - there's really no incentive to go with AMD right now. If they can deliver better frequencies (say 3.1 GHz), that would make a dual socket blade look much more appealing and would definitely make the market more competitive.

  2. Comment - AMD quad-core Opteron servers claim performance records

    (Apr 10, 2008 - 2:01 PM)

    This is an apples to oranges comparison. Go look at the SPEC numbers for yourself and you will see that even the HP ProLiant BL460c (2.33 GHz Intel Xeon E5410 in a 2-way configuration, 8 cores) beats the HP ProLiant BL685c G5 (2.3 GHz AMD Opteron 8356 4-way configuration, 16 cores) in many SPEC result and baseline scores (of course not rates due to the 2-way vs. 4-way configuration differences).

    To try to make this a bit more apples to apples in comparison, let's look at two different HP systems.

    The HP ProLiant BL685c G5 server blade costs
    $13,697.00 for the 8354 AMD Opteron 4-way configuration - the 8356 AMD Opteron is not available for order at this time and would add approximately $600 to the system cost ($150 price difference per processor times 4 processors). As far as power consumption is concerned, the cumulative ACP for the processors alone in this configuration is 300W.

    The identically configured HP ProLiant BL460c server blade costs $5,067.00 for the Intel Xeon E5450 2-way configuration. As far as power consumption is concerned, the cumulative ACP for the processors alone in this configuration is 160W.

    Looking at what I can actually order today - there is no comparison. I can buy 2 of the Intel systems for considerably less than one of the AMD systems. Since these are blade systems - the Intel offerings become even more attractive. I can buy a 16 blade Intel based system much cheaper than an 8 blade AMD system. The Intel system will outperform the AMD system. The only plus for the AMD system will be in total power consumption - which is negated if you choose to purchase a comparable 4-way Intel system from a different vendor (HP just doesn't offer that configuration with the newer Intel processors).

    So, there you have it. But, don't take my word for it - go and investigate the real performance numbers at http://www.spec.org/ and configure/price compare systems for yourself http://www.hp.com/ or look at other vendor offerings for a better comparison.

    Whatever you do - do NOT make assumptions and purchase decisions based on this article alone as it presents a biased opinion drawn from only a few SPEC numbers taken from incomparable systems. The Intel Xeon X7350 had SPEC numbers published over 9 month ago.