Chip Clock HT Cache Bus Speed --------------------------------------------------------- 5080 3.7 GHz YES 2MB 1066 MHz 5160 3.0 GHz NO 4MB 1333 MHz Does the .7 GHz and HT worth more then 4MB cache and higher bus speed? The application is VoIP so there is not a lot of IO so I would not think Bus Speed would matter. I am finding mixed information on HT, some say it is great, others say it actually slows things down, could this be why the new chips done have HT?><>Nathan Stratton CTO, VoilaIP Communications nathan at robotics.net nathan at voilaip.com http://www.robotics.net http://www.voilaip.com _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
On Sunday 03 September 2006 8:55 pm, Nathan Allen Stratton wrote:> Chip Clock HT Cache Bus Speed > --------------------------------------------------------- > 5080 3.7 GHz YES 2MB 1066 MHz > 5160 3.0 GHz NO 4MB 1333 MHz > > Does the .7 GHz and HT worth more then 4MB cache and higher bus speed? The > application is VoIP so there is not a lot of IO so I would not think Bus > Speed would matter. I am finding mixed information on HT, some say it is > great, others say it actually slows things down, could this be why the new > chips done have HT?HT helps to avoid context switches when used with an HT-aware scheduler. it can be very significant for I/O in Xen because everything a DomU does has to pass through Dom0 for I/O. dedicating an HT thread exclusively for Dom0 is almost as good as giving it a full real core, but usually much cheaper. AFAIK, all Xen schedulers, all modern Linux, BSD and Solaris kernels are fully HT-aware, and make good use of it. I also think (but not sure) that some servicepacks for win2k and winXP gave them most, if not all, of the advantages of a modern scheduler. the only remaining cases where HT impacts negatively are heavy single threaded tasks. there, you''d like to dedicate the whole processor to a process during its timeslice, without other tasks taking up resources (cache, rename registers, FSB, etc) with server-like workloads, even more with Xen, HT is usually a good thing. i have no clue why newer chips doesn''t have it, neither why AMD never bothered with something like that. maybe there''s some other consideration i don''t know about. -- Javier _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Nathan Allen Stratton wrote:> i have no clue why newer chips doesn''t have it, neither why AMD never > bothered > with something like that. maybe there''s some other consideration i don''t > know about.The latter thing is quite simple: AMD threw the more transistors of their new production process in AMD64, the 64bit extension; Intel made HT instead. cu cp _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
On Mon, 4 Sep 2006, Christoph Purrucker wrote:> Nathan Allen Stratton wrote: > > > i have no clue why newer chips doesn''t have it, neither why AMD never > > bothered > > with something like that. maybe there''s some other consideration i don''t > > know about. > > The latter thing is quite simple: AMD threw the more transistors of their > new production process in AMD64, the 64bit extension; Intel made HT > instead.Yet, there is some reason why Intel does not have it in the 51xx. -Nathan _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
On Sun, 3 Sep 2006, Javier Guerra wrote:> On Sunday 03 September 2006 8:55 pm, Nathan Allen Stratton wrote: > > Chip Clock HT Cache Bus Speed > > --------------------------------------------------------- > > 5080 3.7 GHz YES 2MB 1066 MHz > > 5160 3.0 GHz NO 4MB 1333 MHz > > > > Does the .7 GHz and HT worth more then 4MB cache and higher bus speed? The > > application is VoIP so there is not a lot of IO so I would not think Bus > > Speed would matter. I am finding mixed information on HT, some say it is > > great, others say it actually slows things down, could this be why the new > > chips done have HT? > > HT helps to avoid context switches when used with an HT-aware scheduler. it > can be very significant for I/O in Xen because everything a DomU does has to > pass through Dom0 for I/O. dedicating an HT thread exclusively for Dom0 is > almost as good as giving it a full real core, but usually much cheaper. > > AFAIK, all Xen schedulers, all modern Linux, BSD and Solaris kernels are fully > HT-aware, and make good use of it. I also think (but not sure) that some > servicepacks for win2k and winXP gave them most, if not all, of the > advantages of a modern scheduler. > > the only remaining cases where HT impacts negatively are heavy single threaded > tasks. there, you''d like to dedicate the whole processor to a process during > its timeslice, without other tasks taking up resources (cache, rename > registers, FSB, etc)All our applications run on Linux and all are threaded.> with server-like workloads, even more with Xen, HT is usually a good thing.So, would it make sense to go with the older 5080 and take the extra speed and HT vs the newer 5160? The only unknown I have is on the CACHE, 2 MB vs 4 MB.><>Nathan Stratton CTO, VoilaIP Communications nathan at robotics.net nathan at voilaip.com http://www.robotics.net http://www.voilaip.com _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
> -----Original Message----- > From: xen-users-bounces@lists.xensource.com > [mailto:xen-users-bounces@lists.xensource.com] On Behalf Of > Nathan Allen Stratton > Sent: 04 September 2006 14:29 > To: Christoph Purrucker > Cc: xen-users@lists.xensource.com; Javier Guerra > Subject: Re: [Xen-users] Xeon 5160 vs 5080 > > On Mon, 4 Sep 2006, Christoph Purrucker wrote: > > > Nathan Allen Stratton wrote: > > > > > i have no clue why newer chips doesn''t have it, neither > why AMD never > > > bothered > > > with something like that. maybe there''s some other > consideration i don''t > > > know about. > > > > The latter thing is quite simple: AMD threw the more > transistors of their > > new production process in AMD64, the 64bit extension; Intel made HT > > instead. > > Yet, there is some reason why Intel does not have it in the 51xx.Put quite simply: Hyperthreading makes more sense on a very long pipeline (because on any branch-mispredict or some other pipeline stall, there is a second chance to have something useful to do, ready to run). As I''m sure plenty of people are aware, Hyperthreading can give better performance in some cases, but there are plenty of cases where it actually performs worse. Since both the AMD processors and the newer Intel processors don''t have very long pipelines, they also haven''t got Hyperthreading. Of course, everything in the processor design is a balance between the performance gain and cost in transistors. -- Mats> > -Nathan > > _______________________________________________ > Xen-users mailing list > Xen-users@lists.xensource.com > http://lists.xensource.com/xen-users > > >_______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Hi> Chip Clock HT Cache Bus Speed > --------------------------------------------------------- > 5080 3.7 GHz YES 2MB 1066 MHz > 5160 3.0 GHz NO 4MB 1333 MHz > > Does the .7 GHz and HT worth more then 4MB cache and higher > bus speed? The application is VoIP so there is not a lot of > IO so I would not think Bus Speed would matter. I am finding > mixed information on HT, some say it is great, others say it > actually slows things down, could this be why the new chips > done have HT?Hyperthreading is only bad in a few cases. One of the I heard a lot about was running *old* versions of interbase with hypertheaded CPU, where the "one-process" design of interbase in conjunction with the required locks lead to the problem of having the process "jumping" between the virtual cores, taking about all the CPU load for nothing. The solution in that time was to disable Hyptertheading OR to set affinity to one cpu only. (Current builds don''t have that problem as they use NPTL.) However by bet is that the 5160 will outperform the 5080 for it''s 4 MB cache alone! (Yes running virtual hosts does benefit from more "virtual cores", but those also need to share the cache...) Regards, Steffen _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Nathan Allen Stratton wrote:> Chip Clock HT Cache Bus Speed > --------------------------------------------------------- > 5080 3.7 GHz YES 2MB 1066 MHz > 5160 3.0 GHz NO 4MB 1333 MHz > > Does the .7 GHz and HT worth more then 4MB cache and higher bus speed? The > application is VoIP so there is not a lot of IO so I would not think Bus > Speed would matter. I am finding mixed information on HT, some say it is > great, others say it actually slows things down, could this be why the new > chips done have HT?I recently ordered a Dell Precision 490 workstation with a single Xeon 5160 (Woodcrest) CPU instead of the Xeon 5080 (Dempsey) CPU. My reasons was a cooler CPU (85W vs 130W Peak TDP) and a faster, more effective CPU (4MB shared vs 2MB per core L2 Cache, 1333 MHz vs 1066 MHz FSB speed). Dell''s 490 price with Xeon 5160 was lower than for an equally configured 390 single socket workstation with the slower Core 2 Duo Extreme 2.93 GHz CPU. The following benchmark by Phoronix with FC5 Linux proves that even the 2,66 GHz Xeon 5150 in general is faster than the 3,77 GHz Xeon 5080 with exception for little lower RAMspeed http://www.phoronix.com/scan.php?page=article&item=521&num=1 More about the Xeon Woodcrest/Bensley platform vs Opteron at Tom''s Hardware http://www.tomshardware.com/2006/06/26/xeon_woodcrest_preys_on_opteron/index.html The new Mac Pro is also based on Xeon 51XX Woodcrest: http://www.anandtech.com/mac/showdoc.aspx?i=2816 Rgds Terje _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users