I, in no way, consider myself a PC arch expert. Nevertheless, I must respond. I think the major difference has to do with the shared 4MB L2 that the Woodcrests use and the amount of L1 present. Each pair of Woodcrest cores shares the pool of L2 and they each have 32K of L1 to load instructions from. The Opteron uses dedicated 1MB of L2 cache and 64K of L1 per core. I've yet to have time to read the Woodcrest whitepaper on how the L2 pool is partitioned while being used. I do know that the 5150 uses a 1000Mhz bus and the 5160 uses a 1333Mhz bus. The Opteron series all use 1000Mhz bus as well.
If you break down the Mhz and bogomips on the 275 versus the 5160 and the 5150, you'll find out that they all have .5 bogomips per Mhz. In an ideal setting, they should perform identically. My guess would be that it's the shared L2 and the O/S that's causing the slow-down for you. You could have too much stuff that just isn't getting in and out fast enough. The O/S and/or the applications may not be utilizing the chips to their potential. There was a time, not too long ago, when the Linux SMP kernels didn't schedule threading and single processes well. In some cases, you were better off running a single-core kernel. If you want to truly compare your boxen, purchase/download a benchmarking tool like Winstone or Sisoft Sandra. ------------------------------------------- Glen Batchelor IT Director All-Spec Industries phone: (910) 332-0424 fax: (910) 763-5664 e-mail: [EMAIL PROTECTED] ------------------------------------------- www.allspec.com ------------------------------------------- > -----Original Message----- > From: [EMAIL PROTECTED] [mailto:owner-u2- > [EMAIL PROTECTED] On Behalf Of Gabriel Green > Sent: Thursday, June 14, 2007 12:34 AM > To: u2-users@listserver.u2ug.org > Subject: Re: [U2] New UV system -- AMD Opteron v. Xeon? PC architecture > experts? > > Considering all the discussion here, I'd like to ask a question and make a > statement for those a little more familiar with server architecture > > I have a year old Sun Fire x4200 server, two Dual Core Opteron 275s, which > performed much better with UV and in general than my brand spankin' new > Dell > PowerEdge 1950, with two Xeon Woodcrests (Core 2 Duo-based). Each machine > has 16G RAM and similar hard drive configurations. Both run Windows > Server > 2003 x64 Standard Edition R2 Service Pack 2 and UV 10.2. > > Get this-- > > The new Dell benchmarked lower and does something very strange. On the > Dell, the System process (not System Idle Process) often jumps between 3 > and > 20 percent, all the time. My suspicion is that AMD Opteron based > multiprocessor (meaning multiple physical processors) perform better > because > they can transfer data between physical processors and associated DIMM > slots > across the hyper-fast HyperTransport bus, and not the slower (in the > Dell's > case, I believe ~700 MHz) system bus on Intel-based machines. > > Because the Sun machine is faster we are moving our production UV system > back to it this weekend. > > Anyone want to confirm this for me? Or does anyone know how I can track > down what the "System" process is actually -doing-? > > > Just my experience, thanks-- > Gabe > ------- > u2-users mailing list > u2-users@listserver.u2ug.org > To unsubscribe please visit http://listserver.u2ug.org/ ------- u2-users mailing list u2-users@listserver.u2ug.org To unsubscribe please visit http://listserver.u2ug.org/
