Hyper-Threading vs. not - performance impact

[Phynaz]

Originally posted by: Maverick2002
Then by your logic HT is utterly worthless, and so are most high end multi-core systems. I'm going by task manager utilization which stays at 100% for 3ds max/vray rendering on all threads.

No, I'm saying it's worth while because there aren't applications that can max out cpu resources with a single thread. Therefore system throughput increases if you can run a second thread to use some of those idle resources.

 

[mozartrules]

Originally posted by: PhynazHow are you reporting cpu usage, task manager?

Task manager in Windows (used only for development IDE), top in RH5 Linux (X55xx), topas in AIX (P595 with 64 4.2GHz Power6 processors). The comment above was from an 8 core Linux box. But I agree that this doesn't necessarily show that HT isn't fully utilized. The fact that application throughput goes down (quite a lot) as soon as I have more threads than real cores is the best indicator that I am doing well keeping the CPU pipeline full. A lot also depend on the compiler, this is where Intel's compiler really shines.

A nop loop does calculate a lot (assuming that the optimizer dosn't remove it, any C++ compiler worth using will do that). The CPU doesn't know that you aren't doing anything useful!

But I agree that to get get this kind of behavior you have to at a minimum study the assembly (and fix where needed) and program with some knowledge of the caching scheme (padding structs to fit cache lines, ...). Applications written in anything higher level that C++ need not apply.

 

[Phynaz]

Never mind