Forum Replies Created

Page 5 of 13
  • I doubt we’ll hear anything until the 980 Ti and Titan II are out. Wouldn’t really
    make sense for Adobe to do an update only to have to do another when two more Maxwell
    cards come out. Most likely NVIDIA’s waiting until AMD releases its 3K series.

    Ian.

    ——–
    SGI Guru

  • My main thought on this is that, with modern sw techniques now available, it really
    ought to be possible to code the app so that it can learn by itself what the optimal
    settings are as one’s work progresses, in all sorts of ways.

    For largely non-technical users (ie. those who cannot be expected to understand how
    much RAM a particular scene might require to render), the most appropriate question
    is likely to be, what are the generally overall best settings, and what are the
    plus/minus consequences of deviating from these?

    Really though, the app itself ought to be able to learn & adapt; the parameters that
    controlled this process could be adjustable, but expecting every user to hard fix
    these settings isn’t realistic now, the whole process is too complicated and technical
    IMO (eg. re your comment about fragmented memory).

    Ian.

    ——–
    SGI Guru

  • New 700 cards are easily available from numerous sources in Europe,
    that’s not a problem. Price-wise, they’re cheaper than the crazy
    pricing of this past summer, but still a bit too costly in most
    locations. Used ones don’t yet go for what one might call bargain
    prices, ie. not usefully less than new, though I suppose hunting
    for a 6GB 780 would be a good win if one can find them. Only the
    780 Ti would be quicker than two 580s though.

    I mentioned the 900s just so the caveat is there, eg. in case the guy
    suddenly decides to blow some savings. 😀

    Ian.

    ——–
    SGI Guru

  • If it has to be an all-new card, probably a GTX 770, or if you can find a deal,
    a GTX 780. If you can get used cards instead, then look for two used GTX 580 3GB.

    Don’t buy GTX 970/980 yet as they’re currently not properly supported.

    Ian.

    ——–
    SGI Guru

  • Ian Mapleson

    December 30, 2014 at 11:16 am in reply to: CC 2014 RAM Previews take too long to Start processing

    Looking at the settings, I have a followup question: I’d read that it’s normally
    best to leave at least a couple of cores/threads reserved for other applications,
    but Alex has that set to zero. What effect might this have in practice, ie. with
    reserved CPUs = 0? Also, I see at the bottom it states the actual no. of CPUs
    which will be used is thre – how is this number derived? I’d have thought with an
    i7 970 it would still work ok with 4 or even 5 cores, leaving 1 for other tasks.

    Or are these both akin to the RAM Preview checkbox, ie. the best settings varies
    between scenes, etc.?

    Ian.

    ——–
    SGI Guru

  • Using a 780 means just adding some text to the ray trace text file. That’s rather
    different to any more involved changes which may or may not allow a card to function
    properly for all possible tasks which may try to use it.

    Btw, there are 6GB 780s, why not use those?

    Ian.

    ——–
    SGI Guru

  • yonni,

    One cannot judge the CUDA performance of a GPU based on the no. of cores or the
    memory bus width. Read my previous posts in this thread discussing the GTX 580;
    it has far fewer cores and yet is very strong for CUDA, because the shaders run
    at a much higher speed. NVIDIA halved the shader clock after the 500 series (in order
    to make power delivery easier to design & control), negating what judgements one might
    normally make based on simple things like the no. of cores.

    This is why the 580 is faster than all the 600 series cards for AE, and it beats the
    lesser 700 series cards aswell (only the 780 and 780 Ti are faster), eg. two good 580s
    are faster than a Titan Black for AE. See my results here, the Arion page and other test
    results such as the Blender Cycles benchmark.

    Indeed, for some tasks involving CUDA, the 580 beats all the 700 series cards (check the
    reviews on tomshardware).

    Not all tasks use CUDA in the same way though, eg. for AE, a 580 is faster than a 680 by
    quite a margin, but for Arion it’s the other way round, especially for multiple GPUs
    (Arion scales better).

    Beware judgments based on simple things like no. of cores, bus width, etc. They can be
    very misleading, certainly in this instance.

    Ian.

    ——–
    SGI Guru

  • Except I think most people would prefer proper support rather than an ‘iffy hack’,
    so until then, if someone asks, I’ll still answer in the negative.

    Ian.

    ——–
    SGI Guru

  • yonni writes:
    > Not really an option. this is for a home rig and I want the card to be suitable for
    > games. also I live in Israel and used K6000\K5000 even if used will be tto expensive.

    In that case go for a 780, 780 Ti, or multiple 580s. Or of course multiple 780/Ti cards
    if the budget permits.

    > Could you suggest a good format for test purpose?

    To be honest I can’t comment on which would be the best, but usually I thought video was
    either uncompressed or something like JPEG, but the best thing to do is simply experiment
    with the various available formats, see how it affects performance.

    > I have an ssd for the OS and a standard 7200RPM HD for storage. the footage is on that
    > disk. do you think it’s not fast enough for 1080p real time Playback?

    Certainly not if it’s uncompressed data, and I’ve a feeling png isn’t a compressed format.
    I would never use a single drive for any kind of HD stream. For mechanical drives, 2 or 4
    disks in RAID0 or RAID10 is best, though a single SSD is definitely fast enough. More
    costly of course, but the performance is leaps & bounds beyond a rust spinner. Mind you,
    512GB capacity SSDs are lot more affordable now, eg. the MX100.

    Also, for working with AE, you should absolutely have a separate SSD acting as a cache
    drive for AE.

    > sorry for being noob but exactly you mean by i\o setup?

    I mean the storage system.

    Ian.

    ——–
    SGI Guru

  • yonni writes:
    > I really wanted those 4 gigs of ram, they do help me a lot in Mudbox.

    One option is to hunt for a used K5000, I managed to get a couple, also some used
    6000s for an excellent price.

    > 780ti is a bit out of my budget, how bout gtx 780?

    Also ok, though not as quick as two 580s. Not sure where 780 pricing is these days,
    I guess it depends on whether you want max speed or efficiency.

    Having said that, did one vendor make a 6GB 780? Can’t recall offhand…

    > get real time Playback even without any effects on the footage (1080P MOV png).

    Is png really the best image format? Try using other formats, that might help.

    > -When I set the project to use Cuda acceleration I do get the Yellow stripe on
    > the timeline but after 2 seconds the playback get’s very choppy and many frames drops.

    What type of storage medium are you using? Best to use an SSD, or at the very least a good
    RAID0 or RAID10 of quick drives.

    > something is just not working right there..

    Sounds like the bottleneck is elsewhere, perhaps the I/O setup.

    Ian.

    ——–
    SGI Guru

Page 5 of 13

We use anonymous cookies to give you the best experience we can.
Our Privacy policy | GDPR Policy