Hi all, just a quick question. I’m test rendering a few frames from a shot to get an idea of rendering times. I’m on a dual processor workstation and although I’ve got the render command set to ‘Localhost, Localhost’, my processor usage never goes higher than 50%. There are two mantra processes running, each using 25%. Why isn’t Houdini using and the available processor power?

Many thanks,

JT

Are you using Apprentice? If you are, you're limited to only one CPU.

No. Full version. As I said, when I render using ‘localhost,localhost’, there are two mantra processes running, except each one only uses 25%. I don’t experience this problem with Linux and was wondering if it could be due to XP’s hyper-threading?

Regards,

JT

Yes, very possible. The good news is that if it is related to hyperthreading, it's likely that you *are* getting full burn of your cpus, but boy, that's distracting! My understanding of the process is that if the app in question isn't able to make use of HT, it will always indicate half if the available cpu being used. This isn't bad, it should still be using it all.

I have this feeling that in 10 years people are going to look back on Intel's implementation of hyperthreading technology and laugh their asses off.

Cheers,

J.C.

We just got a new single opteron with hyperthreading and running two seperate mantras, not multi-threaded just two different renders uses 100% cpu. Have you tried localhost,localhost,localhost,localhost. Sounds crazy but maybe you can actually run 4 mantras at once on 2 processors….. :?

Hiya, yep, tried the localhost,localhost,localhost,localhost, option. Works on paper, all four mantras fire up and then promptly sit there doing nothing using 0% cpu usage!

I tried turning hyperthreading off, started a render using localhost,localhost, and the cpu usage went to 100% on both processors! However, it took longer to do one frame than with hyperthreading enabled using only 25% on a single mantra! After which it refused to start the next frame. Very strange.

Thanks for the input guys, at least i'm not totally stupid. Thought i might have needed to tick a “please render quickly” box or something.

regards,

JT

We just got a new single opteron with hyperthreading and running two seperate mantras, not multi-threaded just two different renders uses 100% cpu.

I don't believe that Opterons, or any other AMD processor, support hyperthreading. It seems to be an entirely Intel thing. AMD does have something called HyperTransport, which basically refers to AMD's memory bus and its protocol.

HyperThreading is really just a better instruction scheduling algorithm, which utilizes the P4's execution units more effectively by trying to fit 2 instruction threads into the P4s extremely long instruction pipeline. And it certainly sounds better from a PR standpoint than “we had to improve our scheduling algorithm because previously, many of the execution units were dormant”. :?

The Althon and Opteron have shorter instruction pipelines (Athlon has 10, Opteron 12), so they wouldn't benefit nearly as much from such a strategy. The P4s, by comparison have very deep pipelines (original P4s have 20 stages, the new prescott P4s have 31). Hyperthreading helps feed those very hungry pipes.

This is also the main reason why an Opteron at 2.0Ghz can take on a 3.2Ghz P4 - the Opteron can execute an instruction in 12 clocks, whereas the P4 takes 31 clocks to execute an instruction (it's a little more complicated than that, actually, but that's the basic reason why AMD is ‘lagging’ in the Mhz race - they don't need to run as fast).

Anyways, the idea that a single hyperthreaded CPU gives you 2 logical CPUs , or 100% more performance, is just PR hype. It's probably closer to a 10%-15% improvement, which is why your “CPUs” are only running at around 25%.

sorry I got totally confused there, we have two new render boxes one is a dual opteron and one is a P4 with hyperthreading, in my mail I managed to mix the two specs into one impossible one….. When I get a chance I try racing them and see which one wins. :wink: