Posted by Kishore Singhal on May 11, 2010
High accuracy analog simulations face the classic problem of scalability. Doubling the size of a design roughly quadruples the size of the matrix and dramatically increases a simulation’s run time to the point where a design team simply cannot wait for it to finish. Conversely, the team cannot proceed without knowing that the design is functioning. Conundrum!
Happily, improvements in processor technology have provided new machines with multiple processing cores on the same chip. Although HSPICE has had the ability to use these cores for several years now that hasn’t stopped us from making improvements in our multi-core processing capabilities.
Synopsys recently released a new capability in HSPICE that improves the scalability of simulations on machines with four or more cores. This new high-performance multi-processing capability is seeing great scalability on multiple core machines:
|Number of Cores||Average Speedup|
These are average speedups on top of the over 5x speed improvements on single-core HSPICE that our team has already achieved over the past few years. Many of our simulation tests on four cores are reaching even higher performance while maintaining the golden accuracy you demand from HSPICE.
I’d like to go into more detail about how this great new technology works but given that it is patent-pending, I will save that discussion for another blog entry.