Soon It Will Be Cheaper to Compute Than to Cache


At the NewTeeVee Live conference in San Francisco yesterday, I interviewed Sam Blackman, co-founder and CEO of Elemental Technologies, about trends on computing and storage costs. Elemental makes software that takes advantage of graphics processors to provide super fast transcoding, which is the process by which video content is formatted for different devices. The idea of storing one copy of a piece of content and formatting it on the fly makes intuitive sense when compared to storing four or five copies of that content and then delivering it only when needed.

But for now the cost of storing multiple copies is actually cheaper than doing the transcoding in real time. That difference in cost has been a barrier keeping some service providers from buying products that offer capabilities such as those of Elemental’s new server, which launched yesterday. However, Blackman believes that will change within the next three years as prices for processing fall faster than the price of storage. A similar trend is enabling people to store more of their data online, rather than on their hard drives. The cost of bandwidth is so low that folks don’t mind either streaming content or looking it up online rather than downloading it and keeping it on their computers. Check out Blackman’s comments in the video below.




The value is in the ability to customize streaming content/advertising on the fly – otherwise it would still be cheaper overall to use transcoding technology and to store several versions of the content ahead of time. For example, for scalability, one would have to have these GPUs at each of the node of a CDN where several disk versions of the pre-transcoded content would otherwise suffice.

The economics to watch is not just compute vs. disk costs but whether the increased monetization from customization would exceed the increased cost of computing.

On the comments about thousands of different devices – they don’t use thousands of content formats.

p.s On the other trend of storing all the data online, it is true that for majority of consumer media activity that would suffice. No one can possibly store all available data locally anyway. Nonetheless cheaper bandwidth does not mean lower speed of light – so for performance reasons as well as for disconnected operations – several applications will continue to benefit from local copies of data.


I agree to the comment above that storage will still be much cheaper, if you think about moving to SSD technology which consumes even less power.
I think it has to be calculated according to the application. In real time video processing , it might make sense, but when there is a lot of identical copies that can be prepared ahead time it is very hard to compete.

Sam Blackman

Hi folks,

Really good set of points. On the power consumption issue, it’s going to be interesting. As ophirk points out, if there is nothing done to customize the video for the end user, than it will take a long time for compute time to be cheaper / consume less power than storage. However, as I note at the end of the video, if customizing the video based on the end user (product placement, etc) to increase monetization is required, then the real-time processing that has to happen obviates the need to store hundreds of renditions. I talked about this a bit at the end of the video above.


Where’s the power usage in this comparison? With a modern GPU pulling down nearly 200 watts at load, data centers would be hard pressed to scale out many GPU’s and cool them, whereas storage uses relatively no power (less than GPU’s idling). Storage is also more flexible in workloads, because it can be used for anything, but GPU’s are only utilized for specialized applications.

Comments are closed.