Results 1 to 34 of 34
  1. #1
    Join Date
    Jul 2008
    Location
    New Zealand
    Posts
    1,208

    Choosing the right CPU..

    Hi all,

    So we're planning to replace our older servers with newer builds, and i've had a hard time choosing a good CPU. Mainly, i'm a bit confused with the benchmarks i've seen compared to the prices.

    The 3 that i've been considering;
    Intel Xeon X3470 $325
    Intel Xeon W3565 $295
    Intel Xeon E5620 #390

    I basically want a pretty high-end CPU, without paying the extra $$ for the super-new-high-end ones.

    Would anyone be able to tell me whether the E5620 would perform much better then the X3470? (So far, all i know is that it uses ~30% less power..). In some benchmarks i saw, E5620 got 4k, while the X3470 got 6k (not sure how reliable it was, but just got me questioning the CPU's)

    thanks

  2. #2
    Join Date
    Jun 2008
    Location
    San Diego, CA
    Posts
    510
    The X/E5xxx series is dual-socket capable, and that is part of the premium you are paying. X/W3xxx CPUs are not dual-socket capable.

    If you are only going for a single-socket server, then you can go with one of the other CPUs you listed.

    Also note that the CPUs are different sockets. X3470 is a LGA1156 Socket while the other two are LGA1366. You get more memory bandwidth with the 1366 socket because it is triple-channel memory (vs. dual).

    Performance wise, the two single-socket CPUs are within a few % of each other. The E5620 is a good 40% slower than the other two, but since you can install up to 2 of them in one server, if you need more power, you get the most performance from the dual socket solution (and the most cost).

    So in general:
    2 x E5620 > W3565 ~= X3470 > E5620.

    Hope that helps.

  3. #3
    Join Date
    Jul 2008
    Location
    New Zealand
    Posts
    1,208
    Ah wow. Thank you so much! Learned alot more about CPU's by reading your post

  4. #4
    As was said, the x3470 is lga 1156 and the others lga1366. as to performance, sure, the 1366 has more memory bandwidth, but the entire corei7-architechture has plenty of memory bandwidth in general. Tomshardware did a test a while back showing a 0% (yes, 0%) performance difference across all of their tests between running an i7 at 1600mhz vs 800mhz. While the artifical memory bandwidth did show a proportional difference, the real world processing power difference was under 1% in all other tests, even artificial tests. So as far as that goes, I wouldn't expect the slower bus speed to impact the real world performance of the 1156 under any typical scenario.

    In my experience, the x34xx cpus and the e55xx cpus are the same performance clock-for-clock, so the x3470 will outperform the e5520 by a wide margin. Not totally sure about the e5620, but I would assume you'd see the same thing, with the x3470 being 500mhz higher, you'd see a good 20% performance improvement. Comparing the x3460 to the E5520, I saw a performance difference, in terms of requests / second I could handle on a cpu-bound workload, that directly correlated to the difference in cpu mhz.

    The flip side is you can get 2x E5620's into a system, so that could be of value to you. Even so, a dual E5620 system is going to require a larger motherboard, which requires a larger more expensive case, the motherboard will cost about twice as much money, and you'll need 2 cpus as well at twice the money, adn also end up using more power, in order to get total processing power of about 1.5x what you'd see on the x3470. for my money I'd buy two x3470 systems instead. There are also some 6 core E56xx processors I believe, which, at around $600 apiece, provide a rather substantial performance benefit at a relatively reasonable price all things considered. If I just wanted maximum processing power per dollar, I would say a system using the cheapest available 6 core xeon is going to be on par with a system using the x3470, while allowing you to pack more processing power into fewer servers, which saves you hassles in terms of system administration.
    Phoenix Dedicated Servers -- IOFLOOD.com
    Email: sales [at] ioflood.com
    Skype: iofloodsales
    Backup Storage VPS -- 1TBVPS.com

  5. #5
    Join Date
    Jun 2008
    Location
    San Diego, CA
    Posts
    510

  6. #6
    Quote Originally Posted by brentpresley View Post
    Best bang for the buck Server CPU out there right now is the AMD Opteron 6128.

    8 cores at 2GHz for ~$280.
    At a passmark score of 5000 (cpubenchmark.net), I'm not terribly impressed. About on par with an x3440, which is certainly a good cpu, but at around the same price and same benchmark, I wouldn't really be interested unless you could tell me the benchmark was somehow biased in favor of the intel cpus.
    Phoenix Dedicated Servers -- IOFLOOD.com
    Email: sales [at] ioflood.com
    Skype: iofloodsales
    Backup Storage VPS -- 1TBVPS.com

  7. #7
    Join Date
    Jun 2008
    Location
    San Diego, CA
    Posts
    510
    I never said it would beat the snot out of the other chips. Just best bang for the buck, especially considering you can load up to 4 or more of these into a server with the right board. And that is when things REALLY shine. That X3440 is single-socket only.

    Oh, and the current boards that run the AMD 8 and 12 core chips are fully upgradeable to BullDozer when AMD releases it in 3-6 months. That is 16-24 cores in one socket on the horizon. Can you say VPS wet dream?

  8. #8
    How much does it cost to get a 4 or more socket system going? Any links to relevant motherboards there?

    On single socket, getting the same performance for the same price certainly makes it a viable option, but nothing to get too excited about.
    Phoenix Dedicated Servers -- IOFLOOD.com
    Email: sales [at] ioflood.com
    Skype: iofloodsales
    Backup Storage VPS -- 1TBVPS.com

  9. #9
    Join Date
    Jun 2008
    Location
    San Diego, CA
    Posts
    510

  10. #10
    Quote Originally Posted by brentpresley View Post
    http://www.provantage.com/tyan-s8812wgm3nr~7TYA90A9.htm

    For people running on per-server, not per-socket licenses, this is an extremely attractive option.
    thanks for the link
    Phoenix Dedicated Servers -- IOFLOOD.com
    Email: sales [at] ioflood.com
    Skype: iofloodsales
    Backup Storage VPS -- 1TBVPS.com

  11. #11
    Join Date
    Jul 2008
    Location
    New Zealand
    Posts
    1,208
    Would anyone know why the Intel Xeon X3460 benchmarks better then the X3470 ? Also the X3460, costs ~$5 more then the X3470..

    EDIT: was going to get the Intel Xeon W3565, but NE prices are $600 lol.

  12. #12
    do you know if the 6128's power rating is comparable to intel's? I believe they measure it slightly differently, but either way, the AMD shows 115w vs a much lower amount for a comparable performance intel. Any other gotchas to look for?
    Phoenix Dedicated Servers -- IOFLOOD.com
    Email: sales [at] ioflood.com
    Skype: iofloodsales
    Backup Storage VPS -- 1TBVPS.com

  13. #13
    Quote Originally Posted by bhavicp View Post
    Would anyone know why the Intel Xeon X3460 benchmarks better then the X3470 ? Also the X3460, costs ~$5 more then the X3470..
    probably just because the margin for error in the test is greater than the performance difference between the two cpus
    Phoenix Dedicated Servers -- IOFLOOD.com
    Email: sales [at] ioflood.com
    Skype: iofloodsales
    Backup Storage VPS -- 1TBVPS.com

  14. #14
    Join Date
    Jun 2008
    Location
    San Diego, CA
    Posts
    510
    Quote Originally Posted by funkywizard View Post
    do you know if the 6128's power rating is comparable to intel's? I believe they measure it slightly differently, but either way, the AMD shows 115w vs a much lower amount for a comparable performance intel. Any other gotchas to look for?
    AMD uses "ACP" for their power measurements. It's effectively "nominal" power draw.

    I can tell you this: 8 of these CPUs, with 96GB RAM, and 12 x SATA 7200 RPM 3.5" HDDs pull 3.5 Amps on a 208V circuit at full load.

  15. #15
    Join Date
    Jun 2008
    Location
    San Diego, CA
    Posts
    510

  16. #16
    Join Date
    Jul 2008
    Location
    New Zealand
    Posts
    1,208
    Quote Originally Posted by brentpresley View Post
    The Opteron 6128 is HE, so i guess a tad lower performance as well?

  17. #17
    Join Date
    Jun 2008
    Location
    San Diego, CA
    Posts
    510

  18. #18
    Quote Originally Posted by brentpresley View Post
    AMD uses "ACP" for their power measurements. It's effectively "nominal" power draw.

    I can tell you this: 8 of these CPUs, with 96GB RAM, and 12 x SATA 7200 RPM 3.5" HDDs pull 3.5 Amps on a 208V circuit at full load.
    That's the information I was looking for, thanks.

    I've got 24x 3.5" 7200 rpm sata, 72gb ram, and dual E5506 doing 2.5 amps 230v if memory serves, and a similar config with dual E5520 doing 5a @ 120v. I'd estimate I'd drop about 1a @ 120v or 0.5a @ 230v with 12 hard drives instead, but I'm not certain if this was under full load or not. If I recall, I saw about 100w extra power use with both cpus cranking at full blast, so for sake of argument, let's say 3a @230v with 24 drives and 72gb ram and dual cpu, or most likely 2.5a @230v with 12 drives, 72gb ram, and dual cpu.

    I'll gladly take an extra 1a @ 230v or 2a @ 120v to get similar performance, seeing as how the E5645 is rated for the same power use as the E5520, but two of the E5645 will perform similarly to 4 of the opterons, while saving $300+ on the motherboard, requiring a smaller case and lower power supply, and with 2x E5645's costing the same as 4 of the opteron cpus.

    In the UK, that extra 1a @ 230v is worth a good $80 / mo, so you'd save easily $1000, if not $1400, in the first year alone, and still get the same processing power.
    Phoenix Dedicated Servers -- IOFLOOD.com
    Email: sales [at] ioflood.com
    Skype: iofloodsales
    Backup Storage VPS -- 1TBVPS.com

  19. #19
    Join Date
    Jun 2008
    Location
    San Diego, CA
    Posts
    510
    We do not see the performance you are quoting.

    For us, 2 x L5520 CPUs in one of our grids does about as much work as 2 x Opteron 6128s. (8 real + 8 HT cores vs. 16 real cores). And I can load up more VPS's on the AMD box, which makes its revenue much higher per box.

  20. #20
    Quote Originally Posted by brentpresley View Post
    We do not see the performance you are quoting.

    For us, 2 x L5520 CPUs in one of our grids does about as much work as 2 x Opteron 6128s. (8 real + 8 HT cores vs. 16 real cores). And I can load up more VPS's on the AMD box, which makes its revenue much higher per box.
    I agree with you. the 6128 should perform similarly to the 5520. What I was saying was, the 6 core intel should perform about twice as fast as either the 5520 or the 6128, costs about $600 per cpu, and uses the same power as a 5520. The motherboard also would be in the $400-$500 range, vs close to $800 for the opteron board you linked me. And then of course, with the opteron, you have to find a case that's large enough and has enough power to support it. I'm not sure what that entails, but I would assume it would cost more than the cases I can use for the dual xeons.

    Even taking away the case considerations, you can use easily 200w less power and save at least $300 up front by going a dual xeon route compared to a quad opteron route, and get about the same performance. It won't necessarily be equal in all circumstances (like with tons of VMs where more slower real cores might be better than fewer faster real cores), but even in that circumstance, you have to decide if using an extra 200 watts of power, and spending a few hundred bucks more up front is justified. It might be, it might not.
    Phoenix Dedicated Servers -- IOFLOOD.com
    Email: sales [at] ioflood.com
    Skype: iofloodsales
    Backup Storage VPS -- 1TBVPS.com

  21. #21
    Join Date
    Jun 2008
    Location
    San Diego, CA
    Posts
    510
    The 6 core Intel is not 2X as fast as a 5520. It's 50% faster, but not 100% faster. I have several of these, and while they are nice, fast, and great machines, they just are not 2X as fast.

    Just that fact alone throws a monkey wrench in all your calculations.

  22. #22
    Join Date
    Jun 2008
    Location
    San Diego, CA
    Posts
    510
    And when you think of rackspace costs and the fact that I can cram more cores into the same Us using AMD, then the price/performance metric of the 6128 is even better.


    BTW, this is a much better set of benchmarks than passmark, and far more relevant to our discussion:
    http://www.anandtech.com/show/2978/a...6-core-xeon/14

    There Anand compared Intel 6 Core at 2.93GHz vs. AMD 12 cores at 2.2GHz.

    Essentially, 1 REAL + 1 HT core on Intel at ~3GHz equals 2 REAL AMD cores at 2GHz. Ballparking it.

  23. #23
    Join Date
    Feb 2002
    Location
    New York, NY
    Posts
    4,612
    The Opteron 6128 is a great chip. In a hosting environment, I'd much rather have the extra cores that they provide, even if each core is slower.
    Scott Burns, President
    BQ Internet Corporation
    Remote Rsync and FTP backup solutions
    *** http://www.bqbackup.com/ ***

  24. #24
    Join Date
    Feb 2011
    Posts
    580
    Our selection criteria is simple: we order the fastest Xeons available that fit into 60 TDP. A dual socket 1U server in such config fluctuates its power draw by no more than 1 amp depending on the load. A year ago this meant 8 cores per server, now it is 12 cores per server.

  25. #25
    Join Date
    Jul 2008
    Location
    New Zealand
    Posts
    1,208
    Would anyone know if the Corei5-2400 3.10 6M is comparable to the xeon X3470? The benchmark seems to be 6.5k for the i5, and ~6k for the Xeon. But the i5 is much cheaper then the Xeon, so i wasn't sure if the performance is actually the same or better. (Ie will the HT make a difference in the real world? - The app i am going to run doesn't actually support multi-threads, however, there will be around ~20 of these apps running at once. So i'm wondering if 8 HT cores will make a difference, or will the i5 still be faster.

  26. Quote Originally Posted by bhavicp View Post
    Would anyone know if the Corei5-2400 3.10 6M is comparable to the xeon X3470? The benchmark seems to be 6.5k for the i5, and ~6k for the Xeon. But the i5 is much cheaper then the Xeon, so i wasn't sure if the performance is actually the same or better. (Ie will the HT make a difference in the real world? - The app i am going to run doesn't actually support multi-threads, however, there will be around ~20 of these apps running at once. So i'm wondering if 8 HT cores will make a difference, or will the i5 still be faster.
    well, i5 2400 is sandy bridge based CPU which is a totally different animal than X3470 lynnfield.

    i5-2400 does have higher clock speed than X3470, therefore better raw performamce, but i5-2400 doesn't support HT nor ECC RAM, and it can't be installed on socket 1155 server board such as SM X9SCM/X9SCL.

    if you prefer "Xeon" class, Xeon E3 12xx sandy bridge is being sold already! Xeon E3 1230 (3.2Gz quad-core, 8x HT, 8M L3, ECC, 80w TDP) costs under $240, and it will beat X3470 (2.93G quad-core, 8x HT, 8M L3, 95w TDP) lynnfield very handily, and use less power too:
    http://www.provantage.com/intel-bx80...0~7ITEP3CY.htm

    HOWEVER, supermicro is yet to ship X9SCM/X9SCL server boards in mass quantities due to the infamous SATA-II bug found on H67/P67/C204/C202 south bridge. but it should be soon...

  27. #27
    Join Date
    Jul 2008
    Location
    New Zealand
    Posts
    1,208
    Quote Originally Posted by [email protected] View Post
    well, i5 2400 is sandy bridge based CPU which is a totally different animal than X3470 lynnfield.

    i5-2400 does have higher clock speed than X3470, therefore better raw performamce, but i5-2400 doesn't support HT nor ECC RAM, and it can't be installed on socket 1155 server board such as SM X9SCM/X9SCL.

    if you prefer "Xeon" class, Xeon E3 12xx sandy bridge is being sold already! Xeon E3 1230 (3.2Gz quad-core, 8x HT, 8M L3, ECC, 80w TDP) costs under $240, and it will beat X3470 (2.93G quad-core, 8x HT, 8M L3, 95w TDP) lynnfield very handily, and use less power too:
    http://www.provantage.com/intel-bx80...0~7ITEP3CY.htm

    HOWEVER, supermicro is yet to ship X9SCM/X9SCL server boards in mass quantities due to the infamous SATA-II bug found on H67/P67/C204/C202 south bridge. but it should be soon...
    Thanks for that, that's exactly what i was looking for. Only thing is, i'm told they won't ship until the official release date on 04/05/11

  28. #28
    Join Date
    Aug 2007
    Posts
    46
    I'm assuming these on Ebay probably have the SATA bug...

    http://cgi.ebay.com/SUPERMICO-X9SCM-...item43a4d2bdd0

  29. Quote Originally Posted by affordaservers View Post
    I'm assuming these on Ebay probably have the SATA bug...

    http://cgi.ebay.com/SUPERMICO-X9SCM-...item43a4d2bdd0
    most likely they do have the buggy C204 south bridge.

    we got a X9SCM-F sample few days ago and supermicro wanted us to "swear" that we will never use the 4 SATA-II ports before they would release it.

    however, if you use hardware RAID card or just the two SATA_III ports on affected X9SCM(-F) board, you will be just fine. but, that ebay seller prices the board way too high! when SM releases X9SCM-F board in mass numbers, it should be under $200.
    Last edited by [email protected]; 03-01-2011 at 06:36 PM.

  30. #30
    Join Date
    Oct 2002
    Location
    Vancouver, B.C.
    Posts
    2,656
    Quote Originally Posted by bhavicp View Post
    Would anyone know why the Intel Xeon X3460 benchmarks better then the X3470 ? Also the X3460, costs ~$5 more then the X3470..

    EDIT: was going to get the Intel Xeon W3565, but NE prices are $600 lol.
    Because cpubenchmark.net is based on random user submissions with sample sizes too small to balance out other factors that affect performance.

    Try these benchmarks instead:
    http://www.spec.org/cpu2006/results/cpu2006.html
    ASTUTE HOSTING: Advanced, customized, and scalable solutions with AS54527 Premium Canadian Optimized Network (Level3, PEER1, Shaw, Tinet)
    MicroServers.io: Enterprise Dedicated Hardware with IPMI at VPS-like Prices using AS63213 Affordable Bandwidth (Cogent, HE, Tinet)
    Dedicated Hosting, Colo, Bandwidth, and Fiber out of Vancouver, Seattle, LA, Toronto, NYC, and Miami

  31. #31
    Join Date
    Aug 2007
    Posts
    46
    Quote Originally Posted by [email protected] View Post
    most likely they do have the buggy C204 south bridge.

    we got a X9SCM-F sample few days ago and supermicro wanted us to "swear" that we will never use the 4 SATA-II ports before they would release it.

    however, if you use hardware RAID card or just the two SATA_III ports on affected X9SCM(-F) board, you will be just fine. but, that ebay seller prices the board way too high! when SM releases X9SCM-F board in mass numbers, it should be under $200.
    Yeah, I was thinking that looked a bit high. Probably trying to take advantage of the impatient ones wanting one now!

  32. #32
    Quote Originally Posted by hhw View Post
    Because cpubenchmark.net is based on random user submissions with sample sizes too small to balance out other factors that affect performance.

    Try these benchmarks instead:
    http://www.spec.org/cpu2006/results/cpu2006.html
    +1

    spec.org

    That's what I've been preaching on this forum, too

  33. Quote Originally Posted by bhavicp View Post
    Thanks for that, that's exactly what i was looking for. Only thing is, i'm told they won't ship until the official release date on 04/05/11
    usually Intel likes to release their own server boards to go along with the "matching" Xeon CPUs. since the socket 1155 server boards are being delayed so that they can put on the bug-free south bridge, I guess that's why Intel didn't make a fuzz about the release of Xeon E3 sandy bridge line on Feb-20. for sure, Xeon E3's are on the market now! the pity is that we can only have a few samples of supermicro socket 1155 server boards to play with at this point in time.

  34. #34
    Join Date
    Jul 2008
    Location
    New Zealand
    Posts
    1,208
    Quote Originally Posted by hhw View Post
    Because cpubenchmark.net is based on random user submissions with sample sizes too small to balance out other factors that affect performance.

    Try these benchmarks instead:
    http://www.spec.org/cpu2006/results/cpu2006.html
    Ah awesome. Hopefully they get the new Sandy Bridge CPU's soon as well

Similar Threads

  1. Replies: 2
    Last Post: 11-19-2010, 03:39 PM
  2. Replies: 1
    Last Post: 10-29-2009, 11:16 PM
  3. Replies: 0
    Last Post: 08-11-2009, 11:42 AM
  4. Replies: 5
    Last Post: 06-26-2008, 06:03 PM
  5. Choosing a Host for a number of CPU Intensive sites
    By Domen Lombergar in forum Web Hosting
    Replies: 6
    Last Post: 02-12-2008, 03:40 PM

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •