[dmeyer]

So lets say, for sake of argument, that OpenRL gains support from many of the major renderers, Maxwell included, and that the $1500 caustic card would deliver a 5x speedup compared to today's top end single processor system. This is essentially what they are promising. Who knows how successful it will be, it may very well join the ranks of the discarded ART VPS rackmount boxen in our server room, but lets just posit for a moment.

Let's also say that the 'top end' computers could be built for $1500 each.

So to make a small personal render farm that would equal the speedup of the card, you're looking at:

~$1500 per computer x 5
~$140 for Windows x 5
= $8200 for hardware & windows

Plus $ for maxwell node license? For other renderers one uses? Networking? Also roughly 30x the power consumption? Physical space?

Not to mention that the farm won't help you at all for interactive preview, which is the real advantage here. (Except for perhaps V-Ray RT)

I'm not going to order one because it currently only has support from Mental Ray, and I'd rather eat crushed glass than use that, but if modern softwares develop plugins for it I'd get one.

[JDHill]

Problem is, I don't see where they specify what the 5X (3-5X the site says) speedup actually means. Intel, on the other hand, has been quoting Phi performance with respect to specific machine configurations, running specific classes of simulation. Here, they reference HP and Dell machines (with 10 specific models comprising the list of qualified host machines), but don't specify whether they're talking about a machine equipped with a single 2603 or dual 2687Ws. The one card is 30W and the other 65W, which might seem to be suspiciously efficient for such an increase in performance. But it's not that I'm actually questioning the claim -- I have no basis for doing so -- I just wonder what they mean by it. Maybe it's more accurately specified somewhere else that I haven't seen.

[dmeyer]

Problem is, I don't see where they specify what the 5X (3-5X the site says) speedup actually means. Intel, on the other hand, has been quoting Phi performance with respect to specific machine configurations, running specific classes of simulation. Here, they reference HP and Dell machines (with 10 specific models comprising the list of qualified host machines), but don't specify whether they're talking about a machine equipped with a single 2603 or dual 2687Ws. The one card is 30W and the other 65W, which might seem to be suspiciously efficient for such an increase in performance. But it's not that I'm actually questioning the claim -- I have no basis for doing so -- I just wonder what they mean by it. Maybe it's more accurately specified somewhere else that I haven't seen.

Agree and of course I'm sure the speedup would depend hugely on the optimization of individual plugins. But even a 2-3x speedup for interactivity would be worth it for us. Artists waiting for preview renders are a lot more expensive than hardware.

[mtripoli]

Additionally (I'm not sure how familiar you are so forgive me if this is elementary), Dropbox doesn't actually upload your entire file if you save a change to it; it only uploads the new parts of that file. Pretty intelligent really. My point is, if using cloud-hosted software, the advantage in computation speed may more than make up for a sluggish internet speed.

I knew this, partially. I say partially in that I knew it would save edits made to doc's and such. However, just the other day I uploaded three raw videos I shot on my phone for a project I'm doing. I hadn't edited anything and figured I could make flying changes. I dug into the Dropbox/file change business. Dropbox uses "rsync" under the hood. It keeps track of 4M blocks at a time (it doesn't care about a "whole" file). If things change within a 4M block then it only changes that block. This is done locally on your machine. You can already see where this is going. If you edit a portion of a video (for example), though much of the 4M blocks haven't changed, they've "moved" relative to where they were. This then makes much of the file "look" different necessitating much more, if not all, of the file to be re-uploaded. This doesn't matter much to me; I rarely upload video, mostly working doc's, schematics, etc. For these the transfer times are almost instantaneous and I can work with others on projects very smoothly. If I were working with video I'd find some other method to manage this (perhaps chop up the video into manageable pieces and work on those). Good info though, thanks for mentioning this, I probably wouldn't have looked into it otherwise.