Overclocking in Business PC's (drafting)

Discussion in 'Business & Enterprise Computing' started by NSanity, Apr 30, 2012.

  1. Zendude

    Zendude Member

    Joined:
    Mar 31, 2005
    Messages:
    23
    ^ This.

    As IT manager for a small architectural firm, I can wholeheartly agree with this solution.

    Overclocking used to be a hit and miss affair, but with autoOC now on the K series cpu's, it's a breeze.

    If I were in your shoes NSanity I would buy the fastest K processor on the 1155 socket, get a nice big cpu air cooler and autoclock. Throw in as much good quality ram (Corsair or whatever), a gen 3 SSD and away you go. It's as good as you can do without going uber-expensive on dedicated Workstation gear.

    The good thing with these systems is when the next gen componentry comes along, you gut the box, new internals, away you go again. The old internals can be used to upgrade Admin and back end machines if needed.

    Just have a few spares or even a complete box to do rapid swap outs when someone's machine goes south.

    I wouldn't push too hard to get bleeding edge performance from your cadstations. I found that users got really annoyed when their machines would not work at all. Complaints about the machine not fast enough were rare - users are funny creatures : most adapt to whatever they are given.

    If the user complains about their machine I would say "Sure, if you get accounting to approve $2000 I will happily upgrade your box". It seemed to quieten them down a bit.....

    TLDR Version : buy 1155 K series, auto OC on big air coolers and have spare parts ready. Upgrade on the next gen.
     
  2. metamorphosis

    metamorphosis Member

    Joined:
    Feb 25, 2002
    Messages:
    2,111
    For what it's worth, NSanity, here's my 2c.
    The fewer cores you have, generally-speaking, the more OC you can get out of the individual cores before reaching the same temperature and/or stability problems.
    And the new i5/i7's single-core performance isn't significantly advanced over the on C2D's.
    So.
    If you wanted to save yourself an arm and a leg,
    and also not waste money by potentially blowing up a new CPU (despite what other's have said about warranty, explaining it to your computer store may be more difficult),
    I would recommend going with a bunch of high-end second-hand C2D computers - yes, you heard me.
    The E8500's routinely go up to 4ghz (easy) from their stock 3.16ghz, on stock cooling, and I can confirm have excellent single-core performance. Getting them stable is not terribly hard. If I wasn't running a largely silent solution, I could get mine up to 4.3 pretty easy.

    On the other hand, if money is no object it's not hard to get an i5 up to 4.5 with non-stock cooling.

    @other guys please calm down, it doesn't help the thread.
     
  3. underskore

    underskore Member

    Joined:
    Nov 5, 2002
    Messages:
    4,314
    Location:
    3198
    if the drafties aren't careful their modeling techniques could be pushing the model & assembly complexity up quite a bit. making they sure all know what they're doing is a must.


    (im stuck using IV on 32bit XP at my current job, fore some reason they are happy to fork out for IV subscriptipons and fancy hardware but not for an up to date 64bit windows :sick: 12gb i cant touch in my machine...)
     
  4. Domokun

    Domokun Member

    Joined:
    Nov 28, 2005
    Messages:
    2,691
    I apologise if this has already been mentioned and I've skimmed over it (I blame lack of sleep), but why not take advantage of the "turbo" feature on the Intel processors and simply increase the "turbo" multiplier for single-core operation. For example, I could set my 2700K up as follows:

    1-Core: 4.7GHz (47 x 100MHz, Overclocked)
    2-Core: 3.7GHz (37 x 100MHz, Stock)
    3-Core: 3.6GHz (36 x 100MHz, Stock)
    4-Core: 3.5GHz (35 x 100MHz, Stock)

    Since the processor is technically only overclocked during single-core operation, it should stay within reasonable power and thermal limits. That way, you'd have the best of both worlds. Fast single-core operation for "Inventor" and stable quad-core operation for multi-threaded applications.
     
  5. KonMan

    KonMan Member

    Joined:
    Jun 28, 2001
    Messages:
    589
    Location:
    Melbourne
  6. Zenskas

    Zenskas Member

    Joined:
    Jan 27, 2010
    Messages:
    1,884
    Location:
    Tas
    If only Intel made a K series i3 CPU. Turn off HT, OC the crap outta it and they would cost half as much as a 2600K yet perform just as well for single core use, plus have that extra core for the other 20% workload. Due to being only two cores they would also run a lot cooler and be more stable at high speeds.

    But I guess Intel and others are now thinking dual core is budget stuff only and that nobody would want to OC them anymore :thumbdn:
     
  7. OP
    OP
    NSanity

    NSanity Member

    Joined:
    Mar 11, 2002
    Messages:
    18,361
    Location:
    Canberra
    So I'm back.

    Testing we've done;

    * Intel 520 240GB vs OCZ Revodrive3 X2 MAXIOPS - no benefit whatsoever, despite being 5-20x faster than the intel, depending on queue depth.
    * i7-2600k/z68/32GB vs i7-3930K/X79/64GB - benefits with being able to run multiple products within Factory Design Suite Ultimate seamlessly (well as seamlessly as we can), a little faster with some tasks - but only worth it for *some* drafties.

    I sat down and watched a load of "everything" on with a 10500 part project, procmon and resource monitor.

    Inventor is just downright broken. It will just "go out for lunch" on a thread, and literally do nothing for a while (upto 5 minutes) during a load, then come back.

    CPU is dead, Memory is *slowly* leaking, Disk is doing nothing.

    We're going to try to get Autodesk involved with this project. I'm satisfied we cannot possibly throw any more hardware at it, at all.
     
    Last edited: May 24, 2012
  8. b00n

    b00n Member

    Joined:
    May 2, 2003
    Messages:
    235
    Location:
    Brisbane
    Have you spoken to your local autodesk reseller? as who we have our subscription with are very good in resolving this sort of things with our revit bim projects, as they have a good in house team.

    As for hardware in the architecture firm i admin we use revit/autocad/3d max and use HP z400/z420/z600 with minimum12gb ecc ram/120ssd/quadro 2000 or 4000 and z600 has quadro 6000 and dual xeons

    what sort of work are you using inventor for? as i do work also for a engineering firm using solidworks and we are able to import inventor files and use them in projects, and solidworks does support multi core processing.
     
  9. Luke212

    Luke212 Member

    Joined:
    Feb 26, 2003
    Messages:
    10,179
    Location:
    Sydney
    if the cpu is dead and memory increasing it may be decompressing textures or generating procedural textures or something like that. does it offload work to gpu? that would be a possible explanation. check to see if gpu is being used in that time. other than that maybe its poor programming and it is getting deadlocks causing timeouts. or there is incompatible hardware causing timing issues.
     
  10. millsy

    millsy Member

    Joined:
    Mar 31, 2007
    Messages:
    13,480
    Location:
    Brisbane
    That's ridiculous, some of this big software is rubbish at times, I remember Catia had a great habit of randomly crashing on you
     
  11. OP
    OP
    NSanity

    NSanity Member

    Joined:
    Mar 11, 2002
    Messages:
    18,361
    Location:
    Canberra
    Our Reseller is ImaginIT - in theory the largest reseller of Autodesk software in the world. They have no idea and don't actually listen to us when we discuss the scale of the project we're working on.

    HP Xeon based workstations have been mentioned by our reseller. But they cannot tell me why it will alleviate our problems. ECC ram is slower, we're not really moving *that* much in and out of ram (I don't think we're having memory errors that would be corrected/caught by ECC). The Xeon's are slower, PLUS Inventor for the parts we're complaining about is Single Threaded.

    On top of all this Inventor is Direct3D (and has been since 2006), not OpenGL as such there is Zero need for Quadro's.

    Finally, the client has a few HP Z series workstations - they are far and away slower than our drafting machines. We're talking magnitudes. Mostly they don't see the issues, because they only load up individual assemblies (i.e a part here and there) not entire sections of the factory.

    Factory design and Drafting.

    Moving applications isn't an option. The Client has about $400k of licensing with Autodesk, not to mention us.

    Inventor doesn't use CUDA at all. We're talking flat coloured surfaces, no textures. Some of the stuff its stalling on is the most basic of assemblies (Inventor stock hand rails and platforms), it makes no sense.

    The one thing it does stall on that makes sense, is about 1GB of Tekla modelled assemblies. Thats fine, I can take that - because its actually doing stuff when it does stall (as in the software - I can see in procmon that things are happening).

    Hardware wise I just can't see it. I mean i have a Dual E5 Xeon server sitting here i could run up for it - but the Z series workstations also suffer and they have ECC/Xeon's.

    I tend to find that just because you spend more money on an application, doesn't mean its well written.

    Note to everyone, the software doesn't crash - we don't lose work already completed. We're just sitting there twiddling our thumbs waiting for Inventor to come back from lunch or smoko or wherever the fuck it goes.
     
  12. millsy

    millsy Member

    Joined:
    Mar 31, 2007
    Messages:
    13,480
    Location:
    Brisbane
    Yeah, and you should get more quality for it. The licensing costs for Catia are astronomical to say the least...
     
  13. underskore

    underskore Member

    Joined:
    Nov 5, 2002
    Messages:
    4,314
    Location:
    3198
    10500 unique parts? surely they are not all using all that detail all the time... or do you mean 10500 instances?
     
  14. OP
    OP
    NSanity

    NSanity Member

    Joined:
    Mar 11, 2002
    Messages:
    18,361
    Location:
    Canberra
    Assemblies, parts are even more (X number of parts can make up an assembly).

    No we don't run at that detail all the time, but thats the measure when it comes to benchmarking. Note it took about 40 minutes to load on an SSD capable of 1600MB/sec and ~240k iops.

    We typically run at 2000-2500 assemblies. It's a big project, really big.
     
  15. m3k

    m3k Member

    Joined:
    Dec 29, 2007
    Messages:
    1,019
    simple case of autodesks inefficient coding. Im working with particle solver at uni... moved to blender because it uses all cores.

    seriously a case of an overclocked i5 is faster than a 3930k.. joke

    its a shame... free software is starting to beat them in terms of efficient loading of polygons and point cloud data... i was even going to buy a copy from my university but they just lost a sale :/

    i honestly see a day where blender will have enough plugins to replace max (especially fixing its UI) if it hasn't already.
     
    Last edited: Jun 7, 2012
  16. zero_velocity

    zero_velocity Member

    Joined:
    Sep 16, 2010
    Messages:
    2,867
    Location:
    QLD
    OP, why dont you OC 1 or 2 PC's and see how that goes? leave it for a month then make another decision... that way Im sure you will get the backing of your superiors when they see the improvements :thumbup:
     

Share This Page

Advertisement: