I have a new HP computer with an Intel I7 Quad-Core 2.66GHz processor running
Vista Home Premium 64-bit.

I successfully installed POV-Ray 3.6, then ran the benchmark twice, producing
times of 6:46 and 6:45.

I then successfully installed 3.7 beta, which supports the multiple processors,
and re-ran the benchmark twice, producing times of 7:52 and 7:49.

I noticed that the initial phase, with the incrementing photon count, took much
longer under 3.7, approximately 6:15, and the photon count bounced up and down
quite a bit: up to 2000, back to 1000, up to 4000, back to 2000, etc.  Once the
actual rendering began it went very quickly, with the Windows Task Manager
monitor showing all of the CPUs pegged.  However, the first phase took so much
longer that the total time was longer by a little over a minute under 3.7.

Is this to be expected?

Post a reply to this message

"wisemaj" <nomail@nomail> wrote:
> I have a new HP computer with an Intel I7 Quad-Core 2.66GHz processor running
> Vista Home Premium 64-bit.
>
> I successfully installed POV-Ray 3.6, then ran the benchmark twice, producing
> times of 6:46 and 6:45.
>
> I then successfully installed 3.7 beta, which supports the multiple processors,
> and re-ran the benchmark twice, producing times of 7:52 and 7:49.
>
> I noticed that the initial phase, with the incrementing photon count, took much
> longer under 3.7, approximately 6:15, and the photon count bounced up and down
> quite a bit: up to 2000, back to 1000, up to 4000, back to 2000, etc.  Once the
> actual rendering began it went very quickly, with the Windows Task Manager
> monitor showing all of the CPUs pegged.  However, the first phase took so much
> longer that the total time was longer by a little over a minute under 3.7.
>
> Is this to be expected?

**** I will withdraw this question!!  ****

Sorry, newbie error.  I wasnt' aware of the Run Benchmark option under the
Render menu, and was simply loading and running the Benchmark file from 3.6 in
both cases.  Once I discovered that, and used "Run Benchmark", I got the
following results:

One CPU:   671.38 PPS (662.46 PPS CPU Time) 6:30 Clock Time
All CPUs: 3370.80 PPS (440.70 PPS CPU Time) 1:17 Clock Time

Now that's a beautiful thing!   Thanks.

Post a reply to this message

wisemaj schrieb:

> Sorry, newbie error.  I wasnt' aware of the Run Benchmark option under the
> Render menu, and was simply loading and running the Benchmark file from 3.6 in
> both cases.  Once I discovered that, and used "Run Benchmark", I got the
> following results:
> 
> One CPU:   671.38 PPS (662.46 PPS CPU Time) 6:30 Clock Time
> All CPUs: 3370.80 PPS (440.70 PPS CPU Time) 1:17 Clock Time
> 
> Now that's a beautiful thing!   Thanks.

Be aware that those benchmarks cannot be compared.

On a 4-core machine, a (wall clock) speedup factor like 3.0 to 3.5 is 
realistic.

Post a reply to this message

clipka <ano### [at] anonymousorg> wrote:
> wisemaj schrieb:
>
> > Sorry, newbie error.  I wasnt' aware of the Run Benchmark option under the
> > Render menu, and was simply loading and running the Benchmark file from 3.6 in
> > both cases.  Once I discovered that, and used "Run Benchmark", I got the
> > following results:
> >
> > One CPU:   671.38 PPS (662.46 PPS CPU Time) 6:30 Clock Time
> > All CPUs: 3370.80 PPS (440.70 PPS CPU Time) 1:17 Clock Time
> >
> > Now that's a beautiful thing!   Thanks.
>
> Be aware that those benchmarks cannot be compared.
>
> On a 4-core machine, a (wall clock) speedup factor like 3.0 to 3.5 is
> realistic.

clipka,

Thanks for the input.  Not sure exactly how they do this, but Intel claims a
capacity of 8 (simultaneous?) threads for the I7 processor even though it's
described as "Quad-Core".  Windows Task Monitor shows 8 CPU's and all of them
peg at 100% during the benchmark. Would that change your 3 to 3.5 speedup factor
estimate?

Post a reply to this message

wisemaj schrieb:

> Thanks for the input.  Not sure exactly how they do this, but Intel claims a
> capacity of 8 (simultaneous?) threads for the I7 processor even though it's
> described as "Quad-Core".  Windows Task Monitor shows 8 CPU's and all of them
> peg at 100% during the benchmark. Would that change your 3 to 3.5 speedup factor
> estimate?

What you see there is "hyperthreading" - the capability of each 
individual core to have a certain number of tasks assigned at once (in 
this case two), and essentially switch back and forth between these 
whenever one of them falls idle, e.g. when waiting for data to be 
fetched from main memory into the processor cache. From the OS point of 
view, this is easiest handled by treating each core as two individual ones.

I never examined the effect of this on rendering time before, but the 
figures from a scene I'm currently working on might give an impression 
(run on an i7 machine):

- Restricted to 4 threads:
667 seconds wall-clock time, for a CPU-to-wall-clock ratio of 3.88

- With full 8 threads:
443 seconds wall-clock time, for a CPU-to-wall-clock ratio of 6.99

So yes, there's a significant effect to be seen there.

Post a reply to this message