just had a little quantum leap experience replacing my
old P4 2.6 Ghz with an i7-970 3.2 Ghz (6 physical cores).

Results of the POV-Ray benchmark using 64-bit beta 38:

One CPU:   893 pps
All CPUs: 6591 pps ;) ;) ;)

Post a reply to this message

Christian Froeschlin <chr### [at] chrfrde> wrote:
> just had a little quantum leap experience replacing my
> old P4 2.6 Ghz with an i7-970 3.2 Ghz (6 physical cores).
>
> Results of the POV-Ray benchmark using 64-bit beta 38:
>
> One CPU:   893 pps
> All CPUs: 6591 pps ;) ;) ;)

congrats.  I see MCPov will suddenly be much more useful...

Post a reply to this message

Le 27/08/2010 23:59, Christian Froeschlin nous fit lire :
> just had a little quantum leap experience replacing my
> old P4 2.6 Ghz with an i7-970 3.2 Ghz (6 physical cores).
> 
> Results of the POV-Ray benchmark using 64-bit beta 38:
> 
> One CPU:   893 pps
> All CPUs: 6591 pps ;) ;) ;)
> 
We are missing the OS & memory (size & type) if we want to compare.

And how you get that number (because all I got is time: render, photon,
bounding & parser)

Are we making a new benchmark site/database ?

rot47, because it's not about bragging.

>6[ H:E9 :f hg_)[ 36E2bg @? =:?FI ec3:EDi =6DD E92? ba D64@?5D E@
C6?56C[ D@ E92EVD 23@FE g`ha AAD [ 2?5 :EVD @?=J b]bbvwK[ D@ J@FC ?F>36C
D66>D 2 3:E =@HX

#6?56C %:>6i
  !9@E@? %:>6i      _ 9@FCD  _ >:?FE6D  _ D64@?5D W_]a_b D64@?5DX
              FD:?8 `g E9C625WDX H:E9 _]abb r!&\D64@?5D E@E2=
  #25:@D:EJ %:>6i   }@ C25:@D:EJ
  %C246 %:>6i       _ 9@FCD  _ >:?FE6D b` D64@?5D Wb`]bga D64@?5DX
              FD:?8 `d E9C625WDX H:E9 bee]bdb r!&\D64@?5D E@E2=

Post a reply to this message

> We are missing the OS & memory (size & type) if we want to compare.

Sorry, Windows 7 x64 with 6 GB, 3 x 2GB Mushkin DDR3 1333 Mhz,
although BIOS settings seem to indicate its running at 1066 Mhz.
I'm a bit hazy on modern technology details; the last system I
assembled myself was a P66 with 8MB back in 1994 ... I initially
considered ordering Corsair DDR3 1600 Mhz but the vendor told me
the 1600 Mhz wouldn't benefit my configuration unless I overclocked
(it's one of these sites where you configure your system online).

Also my initial enthusiasm was dampened by some blue screens and I
actually found some problem using memtest86, seems two of the three
DDR's had problems and it's running on 2GB for the time being but
seems to be stable now (I hope ...).

> And how you get that number (because all I got is time: render, photon,
> bounding & parser)

In the Windows version there is menu entry "Run benchmark" which
will render the benchmark scene and yield the pps value (and only the
pps value) in a message box. In fact many people testing CPU speeds
seem to use POV-Ray 3.7 just for its benchmark feature and list it
next to other benchmarks such as Cinebench, e.g.

http://www.tabsnet.com/index.php?option=com_benchmark&task=list&bid=5&sysid=0

I didn't try to render benchmark.pov manually because the
menu item takes care of using correct render options.

Post a reply to this message

Le_Forgeron wrote:

> >6[ H:E9 :f hg_)[ 36E2bg @? =:?FI ec3:EDi =6DD E92? ba D64@?5D E@
> C6?56C[ D@ E92EVD 23@FE g`ha AAD [ 2?5 :EVD @?=J b]bbvwK[ D@ J@FC ?F>36C
> D66>D 2 3:E =@HX

my trace time was reported as about 38 seconds using
12 threads. But it seems PPS was given using "elapsed time"
of 40 seconds (it was also in message output). Maybe this
accounts for some of the difference. Not quite sure where
the difference comes from, though.

Also I noticed you used 15 threads, did this give you
any benefit over using 12?

Post a reply to this message

Christian Froeschlin wrote:

> I didn't try to render benchmark.pov manually because the
> menu item takes care of using correct render options.

Actually I'm now quite confused with respect to the benchmark.pov,
benchmark.ini and the built-in benchmark. In the changelog for 3.7
I only find mention of the new -benchmark for linux, but on the page
http://www.povray.org/download/benchmark.php I found a warning that
benchmark.pov is currently not suitable for benchmarking 3.7. In
the bug tracker FS#117 indicates the internal scene has changed.

Most notably benchmark.ini renders a 384x384 image while the
hardcoded benchmark uses 512x512. But despite the smaller size
it renders much slower (2:20 minutes for photons alone using
all CPUs).

Another minor peculiarity is that the built-in command
tells you it will not display anything, but then proceed
to open a render window.

Post a reply to this message

Le 28/08/2010 17:27, Christian Froeschlin nous fit lire :
> Le_Forgeron wrote:
> 
>> >6[ H:E9 :f hg_)[ 36E2bg @? =:?FI ec3:EDi =6DD E92? ba D64@?5D E@
>> C6?56C[ D@ E92EVD 23@FE g`ha AAD [ 2?5 :EVD @?=J b]bbvwK[ D@ J@FC ?F>36C
>> D66>D 2 3:E =@HX
> 
> my trace time was reported as about 38 seconds using
> 12 threads. But it seems PPS was given using "elapsed time"
> of 40 seconds (it was also in message output). Maybe this
> accounts for some of the difference. Not quite sure where
> the difference comes from, though.
> 
> Also I noticed you used 15 threads, did this give you
> any benefit over using 12?

It seems to help a bit. Very little for the benchmark.

Render Time:
  Photon Time:      0 hours  0 minutes  0 seconds (0.201 seconds)
              using 15 thread(s) with 0.229 CPU-seconds total
  Radiosity Time:   No radiosity
  Trace Time:       0 hours  0 minutes 31 seconds (31.103 seconds)
              using 12 thread(s) with 366.155 CPU-seconds total

Render Time:
  Photon Time:      0 hours  0 minutes  0 seconds (0.202 seconds)
              using 18 thread(s) with 0.230 CPU-seconds total
  Radiosity Time:   No radiosity
  Trace Time:       0 hours  0 minutes 30 seconds (30.756 seconds)
              using 15 thread(s) with 363.361 CPU-seconds total
Render Time:
  Photon Time:      0 hours  0 minutes  0 seconds (0.202 seconds)
              using 18 thread(s) with 0.230 CPU-seconds total
  Radiosity Time:   No radiosity
  Trace Time:       0 hours  0 minutes 31 seconds (31.320 seconds)
              using 15 thread(s) with 366.475 CPU-seconds total



Please notice that at such small time, it might also be purely
side-effects (it's unreliable). I might run it with various threads to
make a curve if I get some energy for it.

Post a reply to this message

Le 28/08/2010 18:13, Le_Forgeron nous fit lire :
 I might run it with various threads to
 make a curve if I get some energy for it.

Enjoy (but do not rely on it)
For sack of documentation, until 7 WT, it's one real core per thread.
(and until 12 WT, there is at least one "pseudo core" idle)
(done on 980X / Asrock X58 Extreme3, Linux Ubuntu  amd64, memory in
triple channel mode, PC3-10666 CL 9, might be running at a different
speed: memtest report a bandwidth of 16400 MB/s)

 WT			Trace time
 1			251,26
 2			123,438
 3			82,225
 4			61,38
 5			49,314
 6			41,176
 7			39,025
 8			37,058
 9			35,574
 10			33,814
 11			32,145
 12			31,204
 13			31,086
 14			30,963
 15			30,93
 16			30,965
 17			31,063
 18			30,782
 19			31,148
 20			31,021
 21			31,003
 22			31,179
 23			31,064
 24			31,162
 25			31,312
 26			31,123
 27			31,348
 28			31,295
 29			31,24
 30			31,741
 31			31,396
 32			31,615
 33			31,585
 34			31,627
 35			31,262
 36			31,122

Post a reply to this message