Is it possible for a render farm to be fast *AND* cheap?

Posted on January 31, 2013 in Blog News

If we at Pixel Plow were looking for an online render farm (which we wouldn’t be, because we’ve already got one!), we would be very interested in two things – price and speed of service. If we can’t deliver those things to you, we’ve got a problem.

We ran a test job (a tiled still) a while back with one of our major competitors, and ran the same job on the Pixel Plow render farm. Our timer was started when the job file was submitted (to rule out upload time as a factor) and stopped when the results were delivered (our case), or the file was made available for download (their case). Times shown are HH:mm:ss and all prices were converted to USD at time of job run.

Check out these results:

Competition @lowest priority
Completion in – 6:33:36 (ready for download)
Cost – $260.33

Pixel Plow @low priority
Delivered in – 8:20:28
Cost- $58.14

Pixel Plow @high priority
Delivered in – 3:33:22
Cost- $191.92

Turns out Pixel Plow’s high priority job cost less and was ready to go faster. In fact, once our competitor’s was finished, we still had to go download it – Pixel Plow sends jobs back to a storage folder automatically. And look at the price on Pixel Plow’s low priority job…

How can Pixel Plow’s prices possibly be that low? In this case, our competitor charges for renting the entire farm for the block of time that your job is running, whereas Pixel Plow only charges for the actual CPU time used to compute your render.

To elaborate…if your render, whether still or animation, would leave one or more nodes of our competitor’s render farm idle at some point during the job, they charge you for use of the idle CPU time even though it’s of no benefit to you. In this scenario, if farm X had 196 nodes, and you sent them a 200 frame animation where each frame took about an hour to render, farm X would render the first 196 frames on their 196 nodes during hour 1 and the last 4 frames on 4 of their nodes during hour 2. The problem is that they charge for the second hour of 192 nodes sitting there doing nothing in addition to the 4 nodes that are doing something useful.

Want to know more, or have questions about the test? We’d love to hear from you.

Facebook Icon Twitter Icon Visit Our Vimeo Page