Less could be more for Nehalem Mac Pros

Discussion in 'Mac Pro' started by $BPM$, Mar 12, 2009.

  1. $BPM$ macrumors newbie

    Joined:
    Mar 12, 2009
    Location:
    Wales UK
    #1
  2. Grimace macrumors 68040

    Grimace

    Joined:
    Feb 17, 2003
    Location:
    with Hamburglar.
    #2
    Yeah, weird - this goes against what most everyone has said so far (sets of three will run at 25GB/sec and single/dual-channel for additional chips that run a bit slower.) Real world tests matter, but we'll have to see this corroborated by more examples.

    But, if it's true, and 12GB of tri-channel RAM of the 2.26 starts kicking 16GB of RAM on a 3.2 for some applications...that will make things interesting...
     
  3. drnen macrumors regular

    Joined:
    Jul 1, 2007
    #3
    wow very interesting.. so really the choice is either 6 or 12. I was going to go with 8 just for future upgradeability (so id have 2GB sticks) but i guess i should just start with 6 and then down the line when i'm ready for more i can sell the old ones back to OWC..
    Kinda sucks though.. i really wanted to find a spot between 6 and 12 where i could still use 2GB sticks :/
     
  4. $BPM$ thread starter macrumors newbie

    Joined:
    Mar 12, 2009
    Location:
    Wales UK
    #4
    Yeah, exactly my situation, plumped for the standard 6 and will now probably get 6x2 in a few months. Makes you wonder how fast it would be with 6x4, and all this on the 2.26. The 2.93 for the people who've got the cash is going to be a monster!

    Going to be interesting to see if this also applies to other real world apps.
     
  5. Boneoh macrumors 6502

    Boneoh

    Joined:
    Feb 27, 2009
    Location:
    So. Cal.
    #5
    Your mileage may vary

    The way that I interpret these results is "Your mileage may vary".

    Benchmarks are only a guide, not some kind of wonderful absolute law handed down from on high.

    Use the benchmarks as a relative comparison between machines, or when making changes to your machine.

    I understand that using the 4th slot will force memory access to slow down somewhat. I will run some tests later today, once I get the ram.

    Even if the memory access is somewhat slower, ram access is an order of magnitude faster the disk. So it will be interesting to see how this all relates to real life applications. :confused:
     
  6. Daim macrumors member

    Joined:
    Jun 16, 2008
    #6
    How did they test that? I thought the 2GB models have not been shipped yet..

    edit: hmkay I got it :D
     
  7. Grimace macrumors 68040

    Grimace

    Joined:
    Feb 17, 2003
    Location:
    with Hamburglar.
    #7
    2GB chips are readily available from OWC, Crucial, NewEgg, etc.
     
  8. $BPM$ thread starter macrumors newbie

    Joined:
    Mar 12, 2009
    Location:
    Wales UK
    #8
    Like I said, will be interesting to see if this applies to other real world apps. Look forward to seeing your results.
     
  9. jjahshik32 macrumors 603

    Joined:
    Sep 4, 2006
    #9
    Wow looking at those benchmarks, I now believe that the 2.26GHz nehalem mac pro is WELL worth the buy and seems to be faster than even the 3.2GHz previous gen.

    But its weird so this time around you shouldnt fill up all the last 2 slots if you choose to go with 2GB DIMMs to be faster than actually filling the last 2 spots with 2GB of RAM?? So its faster overall to just leave it at 12GB. Because if you choose to add 4GB DIMMS in the last 2 slots it still slows down as to the 12GB configuration... kind of strange.
     
  10. drnen macrumors regular

    Joined:
    Jul 1, 2007
    #10
    Id still like to see the 2.26 with 8 gigs just to see what it looks like. With that config couldnt they put 2 sticks in each quad of DIMMs? Rather than filling up one side? There by not using up a 4th slot? I really have no clue if one of you know how it works that would be helpful. Id still rather get 2gig sticks that i could KEEP without having to remove entirely when i decide to upgrade..
     
  11. fabriciom macrumors 6502

    Joined:
    Feb 17, 2008
    Location:
    Madrid, España
    #11
    So resuming, to get better performance with most applications on the 2,26. Get the 8GB(4*2GB) and then get another 4GB(2*2GB) aftermarket? Also, would it be better just to get the 6gb that come with the system and exchange for 6*2GB of the same type of memory?

    -Thanks
     
  12. shokunin macrumors regular

    Joined:
    Jun 7, 2005
    #12
    I never fully understood how you could have triple channel and then a single channel that runs slower. I guess even if that is the case, you'll have to wait for whatever processes is running in the slower ram to finish before moving on to the next operation.

    Usually RAM configurations default to the lowest denominator. I would of figured that it would just default to dual-channel since it's 4 slots of ram, while would theoretically be 33% slower. Although many hardware review sites have seen minimal impact of triple-channel versus dual-channel.

    Ideally, Apple should have had 6 DIMM slots per CPU. Maybe in the next revision we will actually get a case redesign to accomodate a wee bit more space for more RAM.
     
  13. shokunin macrumors regular

    Joined:
    Jun 7, 2005
    #13
    You would need to equally split the RAM between processors as each processor will only see the RAM attached to it. SO for 8 Gig, sure, put 2x2GB next to each processor. However, it'll only run in dual-channel mode.
     
  14. drnen macrumors regular

    Joined:
    Jul 1, 2007
    #14
    I get that, i just wanna see what kinda hit that relates to in performance. If it doesnt end up being a big deal then maybe i just take a couple months dealing with dual-channel and then when i decide to upgrade ram later on to 3-channel i dont need to ditch any of the sticks currently in the system. I wish barefeats had included the 8gig spec in that bench
     
  15. Boneoh macrumors 6502

    Boneoh

    Joined:
    Feb 27, 2009
    Location:
    So. Cal.
    #15
    Memory benchmarks

    Hi, y'all!

    Just got my memory from OWC.

    In a nutshell, my 2.26 octa results:

    12 gb (6 * 2gb) max speed was 10,421 MB/sec.
    16 gb (8 * 2gb) max speed was 9,024 MB/sec.

    Seems that the timings for 12 gb were not that much different than 16 gb.

    I am looking for my original benchmark (6 * 1gb), can't seem to find it now. If my memory is right, the 6 gb max speed was in the 8,000 range.

    Here are the details:

    ====================
    12 gb memory test
    ====================

    dlt run-stress-test -d 5M -v OS

    diglloydTools 1.1.0 64-bit Copyright 2006-2008 diglloyd, Inc. All Rights Reserved.
    Thursday, March 12, 2009 12:47:50 PM PT
    Physical memory (MB): 12288

    ********************************** Stress Test *********************************
    System memory: 12GB
    Test duration: 5m
    CPU percent: 100% (per cpu core)
    Num CPU threads: 16
    Memory per thread (requested): 729.6MB

    ==> Determining amount of non-pageable memory
    (this phase can take several minutes before stress-testing begins)


    Checking for virtual memory paging for 11.4GB
    No paging detected for 729.6MB (good)
    No paging detected for 1.42GB (good)
    No paging detected for 2.14GB (good)
    No paging detected for 2.85GB (good)
    No paging detected for 3.56GB (good)
    No paging detected for 4.27GB (good)
    No paging detected for 4.99GB (good)
    No paging detected for 5.70GB (good)
    No paging detected for 6.41GB (good)
    No paging detected for 7.12GB (good)
    No paging detected for 7.84GB (good)
    No paging detected for 8.55GB (good)
    No paging detected for 9.26GB (good)
    No paging detected for 9.97GB (good)
    No paging detected for 10.69GB (good)
    *** PAGING DETECTED after 11.4GB ***
    Allocated 729.6MB per thread (11.4GB) for 16 threads, but paging detected.
    Paging detected, reducing memory usage and retrying with 98% of previous value.

    Checking for virtual memory paging for 11.2GB
    No paging detected for 715.0MB (good)
    No paging detected for 1.40GB (good)
    No paging detected for 2.09GB (good)
    No paging detected for 2.79GB (good)
    No paging detected for 3.49GB (good)
    No paging detected for 4.19GB (good)
    No paging detected for 4.89GB (good)
    No paging detected for 5.59GB (good)
    No paging detected for 6.28GB (good)
    No paging detected for 6.98GB (good)
    No paging detected for 7.68GB (good)
    No paging detected for 8.38GB (good)
    No paging detected for 9.08GB (good)
    No paging detected for 9.77GB (good)
    No paging detected for 10.47GB (good)
    *** PAGING DETECTED after 11.2GB ***
    Allocated 715.0MB per thread (11.2GB) for 16 threads, but paging detected.
    Paging detected, reducing memory usage and retrying with 98% of previous value.

    Checking for virtual memory paging for 10.95GB
    No paging detected for 700.7MB (good)
    No paging detected for 1.37GB (good)
    No paging detected for 2.05GB (good)
    No paging detected for 2.74GB (good)
    No paging detected for 3.42GB (good)
    No paging detected for 4.11GB (good)
    No paging detected for 4.79GB (good)
    No paging detected for 5.47GB (good)
    No paging detected for 6.16GB (good)
    No paging detected for 6.84GB (good)
    No paging detected for 7.53GB (good)
    No paging detected for 8.21GB (good)
    No paging detected for 8.90GB (good)
    No paging detected for 9.58GB (good)
    No paging detected for 10.26GB (good)
    No paging detected for 10.95GB (good)
    Allocated 700.7MB per thread (10.95GB) for 16 threads, reverifying paging behavior...
    Done determining pageable memory.
    NOTE: Avoid starting applications while tests are running; system response will be very sluggish.

    Memory usage per thread: 700.7MB
    Volumes to test: "OS"
    Read-only: false

    Creating 16 CPU thread(s) at 100% utilization, each consuming 700.7MB for a total memory use of 10.95GB
    1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

    Starting...
    1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 started.
    Waiting for all threads to start...
    All threads were started.
    Thread 0 started OK.
    Thread 1 started OK.
    Thread 2 started OK.
    Thread 3 started OK.
    Thread 4 started OK.
    Thread 5 started OK.
    Thread 6 started OK.
    Thread 7 started OK.
    Thread 8 started OK.
    Thread 9 started OK.
    Thread 10 started OK.
    Thread 11 started OK.
    Thread 12 started OK.
    Thread 13 started OK.
    Thread 14 started OK.
    Thread 15 started OK.
    16 cpu threads started successfully using a total of 10.95GB of memory, to be accessed continually.


    =====> Use ctrl-C to stop <=====
    Note: speeds denote memory copying speed (memmove); multiply by 2 for total bandwidth.

    Read-write mode for volume "OS" using temp file __DiskWhacker__217.6
    10s: 56754.4MB @ 9529 MB/sec (best)
    20s: 58155.8MB @ 10355 MB/sec (best)
    30s: 64461.8MB @ 10260 MB/sec
    40s: 63060.5MB @ 10225 MB/sec
    50s: 63060.5MB @ 10226 MB/sec
    60s: 61659.1MB @ 10316 MB/sec
    Maximum bandwidth so far: 10355 MB/sec
    70s: 65863.2MB @ 10184 MB/sec
    80s: 62359.8MB @ 10174 MB/sec
    90s: 63060.5MB @ 10410 MB/sec (best)
    100s: 65863.2MB @ 10218 MB/sec
    110s: 61659.1MB @ 10421 MB/sec (best)
    120s: 64461.8MB @ 10324 MB/sec
    Maximum bandwidth so far: 10421 MB/sec
    130s: 65162.5MB @ 10274 MB/sec
    140s: 60958.4MB @ 10359 MB/sec
    150s: 64461.8MB @ 10308 MB/sec
    160s: 63761.1MB @ 10267 MB/sec
    170s: 62359.8MB @ 10251 MB/sec
    180s: 60958.4MB @ 10317 MB/sec
    Maximum bandwidth so far: 10421 MB/sec
    190s: 65162.5MB @ 10294 MB/sec
    200s: 63761.1MB @ 10231 MB/sec
    210s: 63060.5MB @ 10214 MB/sec
    220s: 63060.5MB @ 10254 MB/sec
    230s: 63060.5MB @ 10168 MB/sec
    240s: 62359.8MB @ 10298 MB/sec
    Maximum bandwidth so far: 10421 MB/sec
    250s: 65162.5MB @ 10366 MB/sec
    260s: 63060.5MB @ 10241 MB/sec
    270s: 62359.8MB @ 10221 MB/sec
    280s: 62359.8MB @ 10233 MB/sec
    290s: 62359.8MB @ 10236 MB/sec
    300s: 65162.5MB @ 10227 MB/sec
    Maximum bandwidth so far: 10421 MB/sec
    Memory bandwidth: 10421 MB/sec
    Waiting for tasks to exit...16...16...15...15...14...13...12...11...11...9...7...6...4...2...done.
    Waiting for DiskWhackerTask for volume "OS" to exit...done.
    Test duration: 300.0 seconds...

    Command "run-stress-test" executed in 390.66 seconds.

    ====================
    16 gb memory test
    ====================

    dlt run-stress-test -d 5M -v OS

    diglloydTools 1.1.0 64-bit Copyright 2006-2008 diglloyd, Inc. All Rights Reserved.
    Thursday, March 12, 2009 1:02:40 PM PT
    Physical memory (MB): 16384

    ********************************** Stress Test *********************************
    System memory: 16GB
    Test duration: 5m
    CPU percent: 100% (per cpu core)
    Num CPU threads: 16
    Memory per thread (requested): 972.8MB

    ==> Determining amount of non-pageable memory
    (this phase can take several minutes before stress-testing begins)


    Checking for virtual memory paging for 15.2GB
    No paging detected for 972.8MB (good)
    No paging detected for 1.90GB (good)
    No paging detected for 2.85GB (good)
    No paging detected for 3.80GB (good)
    No paging detected for 4.75GB (good)
    No paging detected for 5.70GB (good)
    No paging detected for 6.65GB (good)
    No paging detected for 7.60GB (good)
    No paging detected for 8.55GB (good)
    No paging detected for 9.50GB (good)
    No paging detected for 10.45GB (good)
    No paging detected for 11.4GB (good)
    No paging detected for 12.3GB (good)
    No paging detected for 13.3GB (good)
    No paging detected for 14.2GB (good)
    *** PAGING DETECTED after 15.2GB ***
    Allocated 972.8MB per thread (15.2GB) for 16 threads, but paging detected.
    Paging detected, reducing memory usage and retrying with 98% of previous value.

    Checking for virtual memory paging for 14.9GB
    No paging detected for 953.3MB (good)
    No paging detected for 1.86GB (good)
    No paging detected for 2.79GB (good)
    No paging detected for 3.72GB (good)
    No paging detected for 4.65GB (good)
    No paging detected for 5.59GB (good)
    No paging detected for 6.52GB (good)
    No paging detected for 7.45GB (good)
    No paging detected for 8.38GB (good)
    No paging detected for 9.31GB (good)
    No paging detected for 10.24GB (good)
    No paging detected for 11.2GB (good)
    No paging detected for 12.1GB (good)
    No paging detected for 13.0GB (good)
    No paging detected for 14.0GB (good)
    No paging detected for 14.9GB (good)
    Allocated 953.3MB per thread (14.9GB) for 16 threads, reverifying paging behavior...
    Done determining pageable memory.
    NOTE: Avoid starting applications while tests are running; system response will be very sluggish.

    Memory usage per thread: 953.3MB
    Volumes to test: "OS"
    Read-only: false

    Creating 16 CPU thread(s) at 100% utilization, each consuming 953.3MB for a total memory use of 14.9GB
    1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

    Starting...
    1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 started.
    Waiting for all threads to start...
    All threads were started.
    Thread 0 started OK.
    Thread 1 started OK.
    Thread 2 started OK.
    Thread 3 started OK.
    Thread 4 started OK.
    Thread 5 started OK.
    Thread 6 started OK.
    Thread 7 started OK.
    Thread 8 started OK.
    Thread 9 started OK.
    Thread 10 started OK.
    Thread 11 started OK.
    Thread 12 started OK.
    Thread 13 started OK.
    Thread 14 started OK.
    Thread 15 started OK.
    16 cpu threads started successfully using a total of 14.9GB of memory, to be accessed continually.


    =====> Use ctrl-C to stop <=====
    Note: speeds denote memory copying speed (memmove); multiply by 2 for total bandwidth.

    Read-write mode for volume "OS" using temp file __DiskWhacker__158.7
    10s: 33365.3MB @ 6676 MB/sec (best)
    20s: 51477.9MB @ 8196 MB/sec (best)
    30s: 47664.7MB @ 8897 MB/sec (best)
    40s: 48618.0MB @ 8537 MB/sec
    50s: 49571.3MB @ 8233 MB/sec
    60s: 50524.6MB @ 8667 MB/sec
    Maximum bandwidth so far: 8897 MB/sec
    70s: 48618.0MB @ 8864 MB/sec
    80s: 48618.0MB @ 8729 MB/sec
    90s: 45758.2MB @ 9010 MB/sec (best)
    100s: 47664.7MB @ 8925 MB/sec
    110s: 51477.9MB @ 8666 MB/sec
    120s: 49571.3MB @ 8842 MB/sec
    Maximum bandwidth so far: 9010 MB/sec
    130s: 50524.6MB @ 8414 MB/sec
    140s: 45758.2MB @ 8766 MB/sec
    150s: 49571.3MB @ 8762 MB/sec
    160s: 49571.3MB @ 8420 MB/sec
    170s: 47664.7MB @ 9024 MB/sec (best)
    180s: 51477.9MB @ 8774 MB/sec
    Maximum bandwidth so far: 9024 MB/sec
    190s: 47664.7MB @ 8833 MB/sec
    200s: 53384.5MB @ 8805 MB/sec
    210s: 43851.6MB @ 8883 MB/sec
    220s: 50524.6MB @ 8630 MB/sec
    230s: 49571.3MB @ 8445 MB/sec
    240s: 50524.6MB @ 8457 MB/sec
    Maximum bandwidth so far: 9024 MB/sec
    250s: 45758.2MB @ 8835 MB/sec
    260s: 49571.3MB @ 8457 MB/sec
    270s: 49571.3MB @ 8900 MB/sec
    280s: 47664.7MB @ 8327 MB/sec
    290s: 48618.0MB @ 8614 MB/sec
    300s: 51477.9MB @ 8448 MB/sec
    Maximum bandwidth so far: 9024 MB/sec
    Memory bandwidth: 9024 MB/sec
    Waiting for tasks to exit...16...16...16...16...16...16...15...15...15...14...14...13...13...12...11...10...9...9...6...5...4...1...done.
    Waiting for DiskWhackerTask for volume "OS" to exit...done.
    Test duration: 300.0 seconds...

    Command "run-stress-test" executed in 383.93 seconds.
     
  16. Umbongo macrumors 601

    Umbongo

    Joined:
    Sep 14, 2006
    Location:
    England
    #16
    Each processor's memory controller connects to 3 different channels. When all 3 are populated you get the full triple channel bandwidth. Each channel also supports 3 DIMMs but Apple chose to only support more than one DIMM on the first channel.

    Yes they should have.
     
  17. drnen macrumors regular

    Joined:
    Jul 1, 2007
    #17
    thanks for the info Boneoh- if you find that original info for the 6gig and you can run it against 8gig that would be much helpful too :)
     
  18. MADMAXX macrumors member

    Joined:
    Mar 9, 2009
    #18
    Will GFX cards have any impact on performance?
     
  19. VirtualRain macrumors 603

    VirtualRain

    Joined:
    Aug 1, 2008
    Location:
    Vancouver, BC
    #19
  20. nanofrog macrumors G4

    Joined:
    May 6, 2008
    #20
    From what I can tell, they didn't have enough room with the layout they chose combined with the recycled case (different internals only).
     
  21. Morriss macrumors member

    Joined:
    Jul 24, 2008
    #21
    It would be interesting to see the 2008 3.2GHz MP configured and tested with 6 x 2GB ram as well.
     
  22. $BPM$ thread starter macrumors newbie

    Joined:
    Mar 12, 2009
    Location:
    Wales UK
    #22
    Revised graphs now up on Barefeats with full RAM breakdown (inc 8gig)
     
  23. VirtualRain macrumors 603

    VirtualRain

    Joined:
    Aug 1, 2008
    Location:
    Vancouver, BC
    #23
    [​IMG]

    [​IMG]

    This app runs the same on 4 cores as 8 cores (at 8GB they are within 10%)... it seems it's memory limited not CPU limited. I'm mildly curious why the benchmark would suffer at 14GB but not 13GB.
     
  24. drnen macrumors regular

    Joined:
    Jul 1, 2007
    #24
    aah very good thats what i was looking for. So 8gigs is still a safe bet it looks like :)
     
  25. Fomaphone macrumors regular

    Joined:
    Jan 11, 2009
    #25
    how did you have after effects configured in that test, assuming you were running CS4?

    also, today is the day to decide... 12gb or 16gb ($75 more). what do you guys think?

    i'm inclined to go with the 12 and leave a few slots open for 4gb sticks when those become somewhat reasonably priced.
     

Share This Page