Referencing this article.
When we talked to a few VC’s previously about APUs, we were asked to show that there would be demand. Kind of hard to do so in advance of the market, but we made rough estimates. Earlier this year, ClearSpeed took its reference design board and started selling it. Sure enough people bought it. Because it does a number of things quite well. At a lower power consumption.
The trick, as we told the VCs and others, is getting the applications on there. That is the hard part. And that is the part you need to make easy. With a standard board design and basic API. Write your code so that it is not tied to the board, but tied to the API, and have the API interface properly to the board. Board makers may need to write some low level routines, or even a few higher level libraries. As long as a compiler can generate what might seem to be a function call into this API, and pass data back and forth, we should be able to do something, and do it transparently to the end user.
The end user doesn’t care where the application runs. Only that it runs faster. Now the question I hope we get is not “prove that there will be a market” but how big is this going to get … I have a nice answer for that. Hopefully I will get a chance to give it.
Viewed 4553 times by 1249 viewers
[...] Scalability.org has a number of posts on APUs (of which FPGAs are a subset). A couple of weeks ago, there was a post about some discussions the scalability folks had with VCs. More recently there was commentary on APU programming and a discussion on the APU market and possible success factors. As I have said before, I like the concept of accelerator cards, and the potential processing power they bring to the table. What makes me more circumspect are the various barriers to adoption, and the need (or realization of one) for APUs in the mainstream scientific computing market. I just feel that the way they are marketed is somewhat limiting. Vendors need to find applications that show a clear benefit and also ways to make code development a lot easier and scalable (the latter is an area where scalability.org is as good a resource as any) [...]