Dual Processor v. Hyperthreading
 

Since there is often discussion of the relative merits of dual processors versus single, and the new hyperthreading on the PC side, I thought I would chime in with some recent tests (albeit not the most scientific, I tried to eliminate any variables from the tests but there are a couple, but they're instructive).

I have a dual 2.0GHz Xeon machine running Win2000Pro with 1.5Gig RAM and just purchased a 3.0GHz Hyperthreaded XPPro with 510Meg Ram for my wife. I set up Vegas4 and AfterEffects5.5pro on the new machine with the same plug-ins, etc. I then rendered 3 different projects (1 in Vegas, 2 in AE) on each machine.

Interestingly, the dual processor machine was faster in all 3 tests. It was markedly faster in AE, and somewhat (~10-15%) faster in Vegas. I watched the performance monitor and can only surmise that: (1) the extra memory made a big difference overall; and (2) AE is not hyperthread aware perhaps, whereas Vegas is.

I'll try a little more scientific approach if folks are interested and post times, etc.

interesting Stuart. thanks for the feedback. its good to know. Post up more times!

I'm not surprised by this result. I can't recall the exact number, but hyperthreading is like having 1.5 processors, maybe a tad less.

The dual processor system yields about 1.8-1.9 processors (there is some overhead lost to the operating environment).

Following this logic, the Xeons run 1.8 processors in an aggregate 4.0GHz. The P4 runs 1.5 virtual processors in a total of 3.0GHz.

The Xeons should do the work faster.

I agree with Will. AE renders are CPU intensive and AE is SMP-aware, so your box was faster on the basis of CPU power, not RAM.

Hyperthreading may actually slow a system down in some cases, or so I understand.

What would be really interesting IMO would be a comparison of rendering times with HT enabled and disabled.

I haven't explored the ability to disable it. I'll look it up (or do you know how?) and try it out.

HT can be disabled in the BIOS. So it should be possible to do a real-world test on the impact of HT without even opening the box.

If your bios + Windows fully support Hyperthreading all applications
(unless written to detect HT) will see it as a dual CPU system
instead of single. Or in other words, they will use it the same
way as a dual CPU system. This ofcourse does not mean you
get the same performance increase.

Can you check whether Windows 2000 identifies your processor
as a Hyperthreading processor (it should say so somewhere
in your system properties), because I don't think Windows 2000
supports them out of the box. A later Service Pack (3?) might
have added support for them. This is important because a
OS that doesn't know about Hyperthreading will not gain any-
thing by having it!! Also your BIOS must have it enabled as
pointed out above.

In fact, the few tests that I've seen on the web suggest a marginal improvement at best. But they've fallen short of being exact comparisons. Which is why I asked Stuart to run tests on his wife's system with HT on and off. Without changing any other parameters, it should give an idea of what HT really can do...beyond the hypeW2K does not support HT. But his wife's system, which has the HT processor, runs XP, which does support HT. It's unlikely that Stuart's dual Xeons in his W2K workstation are HT, so the fact that W2K doesn't support this technology is moot.

I must have mixed up the OS-es... Sorry. I know for sure I once
saw a Windows 2000 Server machine identifying a processor
as a HT processor. Perhaps there was a special extension
installed or something.