How do you think the new GigE standards will influence the machine vision industry?
Respond or ask your question now!
Acceleware's solution was to use multiple GPUs with 128 core processors per GPU to accelerate the speed. "We really translated the algorithms and physics to something that ran on GPUs," Schneider says. "We did multiple GPUs—four, eight and so on—and then we gave it to our software partners to integrate our stuff. We had to overcome a lot of technical hurdles, but by the time it got to the end-user it was pretty easy. For them, it's a checkbox. It's really, really painless."
By using multicore with multiple GPUs, the simulations run at a minimum 10 times faster, with some products increasing as much as 35 percent. Doing the same simulations as before, that means 400 simulations at 15 minutes each, instead of 10 hours—a total of 100 hours instead of the previous 4,000.
The platform features a Sentaurus Device from Synopsys Inc., (Mountain View, Calif.), API layer and AXE Library form Acceleware and a GPU from NVIDIA (Santa Clara, Calif.). The Sentaurus Device is capable of simulating a range of semiconductor devices.
"Simulation times for a mobile phone that used to take 10 hours now only takes 15 minutes to compute," said an LG spokesperson. "This acceleration has opened up a whole new opportunity to reduce time to market and introduce new optimization algorithms into our processes."
In addition to adopting the multicore GPUs, the system had to deal with parallel problems.