ANAHEIM, Calif. — Because of the large nature that EDA represents, EDA vendors are some of the pioneers of using compute-farms. This raises a question on the probability that these help make progress in the general parallel programming problem. A second question is whether startups, without legacy code to worry about, can optimize their software for these compute-farms.
At the Design Automation Conference this week in Anaheim, Calif., Dr. Luc Burgun, president and CEO of Emulation and Verification Engineering (EVE) SA, acknowledged that EDA tools are compute-intensive and that the EDA industry has put in place creative licensing models to take advantage of this.
Burgun, however, noted that a lot of algorithms are not so easy to distribute and parallelize. He explained to EE Times Europe: "If you take the example of distributed RTL/Gate simulation, researchers have been trying to accelerate the simulation for more than 3 decades with quite limited success. One issue is the huge amount of communication between the various components of the design. Another one is the difficulty to predict the switching activity of those different components, especially for low-power designs."
Similarly questioned about the compute-farm approach, Simon Davidmann, president and CEO of Imperas Ltd., a young company developing multiprocessing development tools (Thame, England), said a key use of compute farms and distributed computing is to provide resources for running regression tests.
Davidmann explained: "Currently, embedded software developers do not utilize compute farms as they historically wait for the hardware before verification begins. This will change with the software teams adopting simulation and we will see embedded software starting to use similar methodology for development as the hardware teams. This includes significant use of compute farms."
In an interview with EE Times Europe, Graham Bell, marketing counsel for EDA startup Extreme DA Corp. (Santa Clara, Calif.), replied in quite a straightforward way. He indeed said the compute-farm approach "is like the 2005 way of doing things. It is a little bit like yesterday."
Bell added: "You should not consider this to solve the parallel programming problem. If you want to use a workstation with eight cores, you will go fast. And, they are not that expensive."
