I just finished building two render nodes for my little basement rendering
farm and stumbled across something of an oddity.

Machine "A" has an Abit IC7-G motherboard, 3.2 GHz Pentium 4 CPU
(hyperthreading enabled), and 1 GB of DDR400 dual channel Corsair
TWINX10243200C2PT dual-channel memory. So does machine "B". Other than the
O/S, the only software installed on machine "A" is POV-Ray for Windows v3.5.
The same is true for machine "B". In fact, both machines are *identical* in
every respect, right down to the manufacturer of the hard disk drives and
the trackball mice... identical in every way save one: Machine "A" is
running Windows XP Pro (SP1), but I didn't have an extra copy of XP lying
around when I built machine "B", so I installed my old copy of Windows 2000
Pro, upgraded to SP4.

I ran a benchmark that I've been using to test the rendering speed on my new
workstations, and over an average of 10 runs machine "A" (the one with
Windows XP) averaged 4 minutes 17 seconds. Now I always thought Windows XP
was a bit of a pig size-wise, so I thought for sure that the Win2K machine
"B" would--at the very least--edge out the performance on its XP-saddled
sibling. Wrong! The Win2K workstation was, in fact, 25% slower than the XP
machine over 10 tests... an average rendering time for exactly the same
scene with all the same settings of 5 minutes 26 seconds.

After checking the BIOS versions and other esoteric settings to ensure that
both machines really *were* identical except for their O/Ses, I went to
CompUSA and bought a second copy of Windows XP Pro which I installed on the
Win2K machine, and then I reran the benchmarks again. The rendering times
were now almost exactly the same as for the other Windows XP Pro
workstation. I'm left to conclude that, for POV-Ray and for that benchmark
at least, Windows XP Pro is significantly faster than Windows 2000 Pro.

How can this be? I thought the core processing speed that both O/Ses are
capable of was pretty much the same, and that the speed-ups in XP were
mostly in things like starting and stopping apps and the system itself. Was
this just some sort of weird fluke, or is the culprit something more subtle
like the hyperthreading processor--does it not like Windows 2000 or
something? If anyone has run into something like this before, I'd sure like
to hear about it.

Thanks...

Post a reply to this message

Darren New wrote:
> Scott Gammans wrote:
> 
>> How can this be?
> 
> 
> My understanding is that Windows XP understands what hyperthreading is, 
> while 2000 merely treats hyperthreading as two independent CPUs. This 
> could be part of the difference. XP might be avoiding scheduling 
> background tasks that would interfere with PovRay, while 2000 thinks 
> it's OK because they're on a different CPU.
> 

And that explanation would make sense, except that during both 
benchmarks I set Render Priority in POV-Ray to "Highest", so even if 
Windows 2000 was ignorant of hyperthreading, wouldn't it still avoid 
scheduling background tasks because POV-Ray was running at highest priority?

Post a reply to this message

Scott Gammans wrote:

>>
> 
> And that explanation would make sense, except that during both 
> benchmarks I set Render Priority in POV-Ray to "Highest", so even if 
> Windows 2000 was ignorant of hyperthreading, wouldn't it still avoid 
> scheduling background tasks because POV-Ray was running at highest 
> priority?
I think that win 2k does know what hyperthreading is and consider your 
box have only one processor

XP understand what Hyperthreading is and consider your box have two 
processors. So pov execute on one proc and the rest of the process are 
bounded to the second (virtual)

Post a reply to this message

I may remember this wrong, or I might remember something from another
Windows version or another OS altogether, but I have the faint memory
of reading something about Windows 2000 having some truely braindead
task scheduling principle that a process cannot get more than 75% of
the CPU, no matter what.
  That is, if you have only one CPU-intensive task running, Windows 2k
will keep the CPU idle 25% of the time for some braindead reason.

  If this what I remember is right, then in practice this means that you
can never render with POV-Ray at full speed in Windows 2000.
  I suppose the only way you could get 100% of CPU is to run two instances
of POV-Ray rendering eg. halves of the image. (Unless Windows 2000 has
some other weird limitation with regard to this.)

  Hyperthreading doesn't have anything to do with this.

-- 
plane{-x+y,-1pigment{bozo color_map{[0rgb x][1rgb x+y]}turbulence 1}}
sphere{0,2pigment{rgbt 1}interior{media{emission 1density{spherical
density_map{[0rgb 0][.5rgb<1,.5>][1rgb 1]}turbulence.9}}}scale
<1,1,3>hollow}text{ttf"timrom""Warp".1,0translate<-1,-.1,2>}//  - Warp -

Post a reply to this message

Darren New <dne### [at] sanrrcom> wrote:
> This is empirically untrue, unless the task manager and performance 
> managers lie. I've had many a Win2K box showing 100% CPU utilization. 
> The "System Idle Task" percentage goes to 0% when you have a CPU-bound 
> job, even at lowest priority.

  Are you sure there was only one unthreaded cpu-intensive application
running when you saw the 100% cpu utilization?

-- 
plane{-x+y,-1pigment{bozo color_map{[0rgb x][1rgb x+y]}turbulence 1}}
sphere{0,2pigment{rgbt 1}interior{media{emission 1density{spherical
density_map{[0rgb 0][.5rgb<1,.5>][1rgb 1]}turbulence.9}}}scale
<1,1,3>hollow}text{ttf"timrom""Warp".1,0translate<-1,-.1,2>}//  - Warp -

Post a reply to this message

Well this is a fascinating discussion--I had always thought that Windows 
  XP was only an incremental improvement over Win2K, but that belief 
seems to have been shattered not only by Warp and Darren's observations 
but my own two eyes as well.

When benchmarks consistently run 20-25% faster on the same workstation 
and the **only** change that was made was to upgrade the O/S, the only 
logical conclusion would seem to be that **something** in the O/S was 
improved between Win2K and XP.

Post a reply to this message

Warp wrote:
>   I may remember this wrong, or I might remember something from another
> Windows version or another OS altogether, but I have the faint memory
> of reading something about Windows 2000 having some truely braindead
> task scheduling principle that a process cannot get more than 75% of
> the CPU, no matter what.

I think you are wrong. Megapov often get between 98 to 100 % of the cpu 
on my laptop. And POVRAY run with the same way.

BTW I never heared about a microsoft operating system which limit by 
defaut the amount of processor time a process can use. . I think that 
this could be tuned, but I donn't know how.

Post a reply to this message

I think it has to do with the way winxp schedules on hyperthreaded
processors. On Win2k, it treats it as an ordinary MP machine, but with XP,
there are additional halt instructions in the idle loop of the OS, freeing
some cpu resources.

Post a reply to this message

"Darren New" <dne### [at] sanrrcom> wrote in message news:404115ee@news.povray.org...
> My understanding is that Windows XP understands what hyperthreading is,
> while 2000 merely treats hyperthreading as two independent CPUs. This

This is in fact correct. If running W2K on a CPU wuth HTT enabled I would
recommend using task manager to lock POV to a single (virtual) processor.

-- Chris

Post a reply to this message