Earlier this year, I had a chance to try to break my own record set last year, and to break the current record set in spring 2010 for “world’s largest photo” during a trip to the UK. While some people plan every aspect of such an event, I’m a bit more flexible – I had the necessary equipment needed to create the image, but it was doubtful that my own computer would be able to handle the creation of an image 5 times larger than my Prague Gigapixel (which itself was almost impossible to create using that computer). In fact, the necessary computing power to create this image properly is about 10x what I have – I simply couldn’t do it by myself. Anyway, I knew that this would be solved with time, and I was in no hurry to finish the image – it would suffice to do the photography, and the rest would eventually fall into place, I hoped.
So I was thrilled to speak to Marcus Hartmann of Fujitsu, who is also an enthusiast of photography, especially panoramic photography. I learned that a cooperation would be possible between Fujitsu and my company, 360cities.net — in short, Marcus would be able to provide me with a world-record-sized computer so that I could create this world-record-sized panoramic image!
After a couple of weeks, I made the trip from Prague, Czech Republic to Augsburg, Germany, to the headquarters of Fujitsu. Marcus generously procured from me the very best and fastest version of Fujitsu’s Celsius workstation – a stupendous 192 gigabytes of RAM, and 2 processors with 6 cores each. To put that in perspective, most geeks in your neighborhood, in 2010, will be bragging about having 16 gigabytes of RAM, and 4 or 8 cores. This computer has what is currently the maximum amount of RAM that can be put into a normal Windows PC, and will probably stay that way for another two or three years (it uses 12 dimms that are 16GB each, and you can’t buy dimms that are larger than 16GB, yet)
Why is such a computer necessary, in order to create a photograph? Aren’t computers like this required for modeling weather systems, or simulating mechanical engineering models, and stuff like that? Well, yes – but the sheer size of this photograph puts this task in the same league as the tasks you might normally associate with requiring the absolute fastest machine available.
To give you an idea of the volume of data that this Celsius workstation has crunched through over the past few weeks: The total number of photos shot was around ten thousand. These RAW files took about 400 gigabytes of data. These images were all converted from RAW files into tiffs; then, groups of about 2000 images were dropped into Autopano Giga – without any knowledge of how the panorama was shot, this program analyzed the images, and fit them together – with a few more days of human effort, these images were joined together nearly perfectly. Then, a single, seamless panoramic image was created out of these few thousand images.
Because the final image is a 360 degree image, but it was shot from each of the four corners of a building, I rendered four different panoramic images, and then I had to align and blend each one. This required opening between two and four images at the same time, each of which was larger than the “world record” image I created last year. The final, completed panoramic image does not even fit into a single file – the largest type of image file in existence that can be opened by an image editing program has a size limit of 300,000 pixels in each direction – this image is 400,000 pixels long – the technical challenges of creating a photo that is larger than can be fit onto one file, is considerable!
All told, the hardware involved in stitching and rendering this London Gigapixel image has allowed an incredible boost in productivity compared with any other computer I would have used, period. For example, rendering a 200,000 x 100,000 pixel panorama with this 12-core 192GB ram Celsius workstation took less than 4 hours. On a “normal fast” 8-core, 16GB ram machine, it would have taken a day or two. There were a number of tasks that took hours instead of days.
Just as importantly, there are the smaller tasks that have to be done in a project like this – and this is where the large amount of RAM made an enormous difference in my productivity (and sanity!). For example, the tasts of manipulating and fine-tuning an image that requires many gigabytes of RAM just to open in Photoshop means that even with a computer that is normally considered very fast (say, 16 or 32 gigabytes of RAM) it wil quickly become unusable. Creating a layer mask with a slower machine might require 15 minutes of waiting between steps. With this machine, it was possible to work normally, as if it was a small photo being manipulated.
Add up all of these steps required for creating and editing this image, and I have two conclusions about my productivity: first, I was able to do multiple iterations of certain tasks, where I would be stuck with the first try if I was using a normal computer (or, extend the project by some weeks); second, I accomplished in a few weeks what would have otherwise required months. Everyone’s time is worth something, and the ROI on a very fast workstation for a task like this is simply undeniable.
So, thanks to Marcus Hartmann and the team at Fujitsu for making this world-record image possible to finish. Without you, I would still be working at this very moment….
Now, go and see it for yourself! http://www.360cities.net/london-photo-en.html