Get the latest tech news
SIMD < SIMT < SMT: Parallelism in Nvidia GPUs (2011)
ogrammable NVIDIA GPUs are very inspiring to hardware geeks, proving that processors with an original, incompatible programming model can become widely used. NVIDIA call their parallel programming model SIMT - "Single Instruction, Multiple Threads".
"Scalar spelling", where you write the code of a single thread using standard arithmetic operators, is arguably a better interface than SIMD loops with ugly assembly-like opcodes. [Disclaimer: NVIDIA doesn't disclose implementation details, and I'm grossly oversimplifying, ignoring things like data forwarding, multiple access ports, and synthesizable vs custom design]. However, I believe the term SIMT was coined by NVIDIA, and perhaps it's a shame that it (apparently) didn't catch, because the architecture "deserves a name" – not necessarily true of every "new paradigm" announced by marketing.
Or read this on Hacker News
