With compositing power you can enhance low sample count cycle renders. The main idea is here scaling and filtering.

Here we have 2 blend files, first is our cycles rendering example and second is only for compositing. Of course you don’t have to use separate blend files. But here i made 3 renders from the example then i saved renders as PNG files and used them for the 3 inputs of the compositing (blue colored a, b and c).

Our compositing screen: 

When rendering our inputs, i set low samples like 25 and changed seed for each render. I know, 25 samples value is very low, but for my laptop 25 samples took 6.2 minutes. So, 75 samples image took about 24 minutes for me.

Number Red 1 is our first PNG file (25 samples and seed value 0). Number Red 2 is our cleaned image by means of 3 low sampled renders. Number 3 is 75 samples and seed value 0.

Let’s look closer a little more to 3 results (PNG, quality %100, resolution %100):

Red 1 is this (25 samples, 6.2 minutes rendering):

Red 2 is this (compositing view, no loss of time):

Red 3 is this (75 samples, 24 minutes):

As a result, i can’t say this compositing is the best way; you can also try other tips like here:

I am not so happy with this cleaning method, but i can only feel better that we can get rid of at least 30 percent of the fireflies faster. If you have a lot of time, you just need to increase the samples for better results. But if you have a slow rendering and don’t like waiting you can render 2 or 3 (or more) times with lower samples and different seeds, and use those images for cascading operations (blur, scale, filter) to reduce fireflies.

Other ways are upgrading your graphics card to a CUDA supported Nvidia card which will cost lots of money. If you have to renew your PC, this is going to be the best solution. Buy plenty of RAM (more than 4 GB), maybe SSD and a good CPU. I am not thinking of using bridged graphics card like Nvidia SLI. That gives at least %25 more power but is it worth to spend so much money?

I also tried to install OpenCL for ATI Mobility Radeon but there is no good news here. I could install OpenCL 1.0 but it has no use for hardware rendering in cycles. OpenCL 1.2 is also installed but I could only get CPU option. GPU rendering gave a blank screen.

If you are using a Core i7 (4 cores) or AMD FX multicore laptop, you better use blender’s internal renderer. And also you can’t upgrade to an Nvidia card because graphic cards are generally on board, or you cannot disassemble so easily like a RAM upgrade.

To read and download this document click here: http://www.scribd.com/doc/102620154/Clean-Fireflies