5 ms·

I'm curious if anyone knows why chips like these are being used today. If you already have software that parallelizes to 16 cores, why not just use a few more

• patrickgzill 17y ago
  Part of it is software licensing that charges per-socket or per-core. For Sun they do have certain applications in mind that are very parallel, such as web serving and message routing, where memory latency can serve to slow things down; the idea of hardware threads is to keep the CPU busy in spite of waiting for a fetch from main RAM.
• oiuyhgnjm 17y ago
  It's a lot easier to share a database across 128 threads on a single machine with uniform memory than shard it across 64 dual core machines with gigabit ethernet.