Hacker News Reader

Comment by dekhn | Hacker News Reader

The Tera was not a success. I remember people trying to get me to use it for graph problems, but it didn't really provide anything I couldn't do with a regular computer at the time.

View on HN

Yeah, lots of people didn't try. What it provided was a way to get over 90% utilization of the processors (every processor issuing useful instructions most every clock tick) without worrying about cache misses. So relatively simple code, useful tools, and high performance. But if you didn't care about any of that, then it didn't provide much.

I'm familiar with how the architecture was designed, and if I had chosen, could have ported code. However, like I said, it didn't really perform any faster than a PC of the time on any code (even ones optimized for the platform). Writing massively multithreaded code, maintaining it, tuning it, and then keeping an entire separate platform for something that isn't measurably faster for a wide range of problems is a huge cost/burden to developers.

I certainly cared about performance but like most people we found that general purpose computers that ran all our codes quickly were not a cost/burden to developers.

I found lots of things that ran much faster on the Tera than on a PC. But yes, if you are happy with the performance of a general-purpose machine, then there's little reason to take on a super. Machine-room requirements alone would finance a lot of off-the-shelf hardware.

Reply on news.ycombinator.com