Posts by Patrick Massey

1) Message boards : Number crunching : Haswell TSX (Message 75665) Posted 26 May 2013 by Patrick Massey Post: I'd get the K version because it's unlocked. Depends on what you're doing. I haven't read into Haswell too much yet, but with Ivy Bridge the K versions lack features that the non-K has (for example, I do not have PCI passthrough (VT-d) on my 3570K, but I'm not planning on running HyperV or Xen on here so it doesn't matter right now).
2) Message boards : Number crunching : Best operating system for crunching (Message 75664) Posted 26 May 2013 by Patrick Massey Post: The newer i3's are quad core processors rather then dual core with hyper-threading, which should yield much better results. I don't believe that's true - they're all 2 cores and 4 threads as far as I'm aware(?) Yes. They are dual cores with HT. Color me wrong. I could have swore I read otherwise, but checking it now shows you're right.
3) Message boards : Number crunching : Best operating system for crunching (Message 75626) Posted 19 May 2013 by Patrick Massey Post: I found that the latest i3s run REALLY well. I found that my first generation (arrandale) i3 gets more granted credit if I run 3 tasks rather than 4 - presumably because there's more L2 cache available per task, and the loss is only for a hyperthreaded core, leaving more resources for the remaining core. The newer i3's might be more efficient with their cores though... The newer i3's are quad core processors rather then dual core with hyper-threading, which should yield much better results. In general on multi-threaded/core systems, I've found that best results come from not running 100%, leaving one core/thread to handle OS tasks. For multi-threaded (HT and AMD-FX) the shared L2 does make a huge difference in performance for some projects (I know that it does for Primegrid).