 |
|
|
|
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Is there a POV-Ray patch or something for hyperthreading?
I just fired up POV-Ray 3.5 with P4 Xeon 3000 ( I'm on vacation: ), and
Skyvase on 640x480 AA 0.3 is a 9 seconds (SSE2 is detected, but the only one
of 'virtual' processors work). Result is bad - P4 Nothwood or AMD 1200 need
about 12 seconds for the same. With some other raytracers, Xeon 3000 can be
up to three times faster than Northwood 1800.
Any help, link, please? Or I need to launch two POVs?
I just tried with
POV-Ray and Ml POV 0.81, with different files, it works, both of procs are
on 100% (but seems that MlPOV don't recognize
SSE).
Anto
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Anto Matkovic wrote:
>
> Is there a POV-Ray patch or something for hyperthreading?
>
> I just fired up POV-Ray 3.5 with P4 Xeon 3000 ( I'm on vacation: ), and
> Skyvase on 640x480 AA 0.3 is a 9 seconds (SSE2 is detected, but the only one
> of 'virtual' processors work). Result is bad - P4 Nothwood or AMD 1200 need
> about 12 seconds for the same. With some other raytracers, Xeon 3000 can be
> up to three times faster than Northwood 1800.
>
> Any help, link, please? Or I need to launch two POVs?
> I just tried with
> POV-Ray and Ml POV 0.81, with different files, it works, both of procs are
> on 100% (but seems that MlPOV don't recognize
> SSE).
See -
http://tag.povray.org/povQandT/languageQandT.html#multiprocessor
See also -
http://www.it-berater.org/smpov.htm
--
Ken Tyler
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
I also just got a P4 with hyperthreading, and I found that running two
instances of POV-Ray actually runs less efficiently than running one
instance.
Running one instance could render a certain image in 35 seconds. Running two
instances at the same time, rendering the same image, took 1:20 seconds,
which is 10 seconds more than twice 35.
Someone wrote a patch that supports multithreading, and that may work better
with hyperthreading, but I haven't tried it.
- Slime
[ http://www.slimeland.com ]
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
"Ken" <tyl### [at] pacbell net> wrote in message
news:3F35319F.F6FA6A73@pacbell.net...
> See also -
> http://www.it-berater.org/smpov.htm
>
Thanks, I just figured how it works - works very well. Theo Gottwald did a
god job.
Anto
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
On Sat, 9 Aug 2003 18:39:29 +0200, "Anto Matkovic" <ant### [at] matkoviccom> wrote:
> Is there a POV-Ray patch or something for hyperthreading?
See: http://news.povray.org/povray.windows/26402/
ABX
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Thanks, Anto!
Since the new explanation "smpov How to" is advertised at povray.org I had
(alltogether) ... 1097 Downloads from SMPOV.
Even people tell me actually where the limits of the current implementation
are.
Mostly if they use tools like Poser, poseray, moray when there are "absolut
pathes in the include-files.
i dont have these tools currently so I could not test that till now.
So far I already have some ideas for a next version. Now I'll take a look if
someone has been
posting numbers if it makes sense to use SMPOV with hyperthreading.
--Theo
-----------------------------------------------------------------------
Distributed Network-Rendering or Local SMP-Rendering. With SMPOV and
POV-Ray 3.5. * Download free at: http://www.it-berater.org/smpov.htm
"Anto Matkovic" <ant### [at] matkoviccom> schrieb im Newsbeitrag
news:3f368408@news.povray.org...
>
> "Ken" <tyl### [at] pacbellnet> wrote in message
> news:3F35319F.F6FA6A73@pacbell.net...
> > See also -
> > http://www.it-berater.org/smpov.htm
> >
> Thanks, I just figured how it works - works very well. Theo Gottwald did a
> god job.
>
> Anto
>
>
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
"Theo Gottwald *" <The### [at] t-onlinede> wrote in message
news:3f38ec3c$1@news.povray.org...
> So far I already have some ideas for a next version. Now I'll take a look
if
> someone has been
> posting numbers if it makes sense to use SMPOV with hyperthreading.
>
Well, seems it haven't sense, even both of procs are on 100%.
Skyvase on 1024x768 AA 0.1:
skyvase.pov rendered in 1024x768 in 2 Tiles (H) from 08-14-2003 / 17:12:36
to 08-14-2003 / 17:13:14 (38).
Maybe I doing something wrong.
This is near the same as rendering with two POV's with two separate ini
files, every one with output to a different half-image file.
(17 and 22 seconds).
Single instance of POV need 29s for all.
However, good surprise for me is possibility to render radiosity (with some
my own files) in parts. I remebering many problems with
POV-Ray 3.0 in this case. Well, this is adavantage of POV 3.5...
Anto
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
On Thu, 14 Aug 2003 17:34:35 +0200, "Anto Matkovic" <ant### [at] matkoviccom> wrote:
> However, good surprise for me is possibility to render radiosity (with some
> my own files) in parts. I remebering many problems with
> POV-Ray 3.0 in this case. Well, this is adavantage of POV 3.5...
AFAIK problem with radiosity rendered in parts in 3.5 is the same like in POV
before 3.5. I would rather say that thanks to better processors we can use
better radiosity settings so artefacts are less visible. Can be this your
case?
ABX
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
> AFAIK problem with radiosity rendered in parts in 3.5 is the same like in
POV
> before 3.5. I would rather say that thanks to better processors we can use
> better radiosity settings so artefacts are less visible. Can be this your
> case?
>
This was with some entries for IRTC with POV 3.0 (1997, 98..). I stopped
render with
radiosity, closed
POV, and tried to continue render. New rendering part was a bit lighter or
darker than old part. No problem to fix this in bitmap
program, bur IRTC rule was: you can only convert image or ajust gamma or
brightness-contrast. So, Pentium 2 from these days had to work continiously
for days or
weeks.
Similar problem can be with some commercial raytracers, if they calculate
radiance map per bucket. You can see different radiosity from quad to quad
of the render view.
As I can see, POV-Ray 35 haven't this problem, because of writting '.rca'
files to HDD (I suppose this is a radiance map file, can't find anything
about in
docs for now), so it possible to rerun radiosity render.
How this work with simultaneuosly rendered parts, I don't know for now.
Anto
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Hi Anto,
Here are my numbers:
skyvase.pov rendered in 1024x768 in 1 Tiles (-) from 08-23-2003 / 11:53:28
to 08-23-2003 / 11:54:15 (47).
skyvase.pov rendered in 1024x768 in 2 Tiles (H) from 08-23-2003 / 11:54:22
to 08-23-2003 / 11:54:58 (36).
skyvase.pov rendered in 1024x768 in 2 Tiles (V) from 08-23-2003 / 11:55:04
to 08-23-2003 / 11:55:52 (48).
skyvase.pov rendered in 1024x768 in 4 Tiles (H) from 08-23-2003 / 11:56:10
to 08-23-2003 / 11:57:02 (52).
However - all rendered locally with my 2 CPU's. You see the Overhead when I
make 4 tiles makes it even slower then when I render this in 1 tile.
In one sentece: Scenes which need only very short rendering (30 Seconds) may
not be the best candidates for SMPOV and Hyperthreading. I have counted that
even before the second instance starts up, 4 Seconds are gone.
Your numbers:
38 Second - 29 Seconds = 9 Seconds diffrence. How long does it take for your
POV-Ray to load?
Remeber we load two instances of POV-Ray. Secondly they start up and load
the POV-File.
Did you include the "loading time" while measuring the 29 Seconds?
Maybe in that case POV-Ray was already loaded and shurely even the file.
Secondly Pictures which need more time to pre-process then actual render
time
maybe not the best candidates for "Tiling" in Hyperthreading Systems. (=
rendering them in pieces).
(General statement)
The point is that we start 2 instances of POV-Ray and EACH does the full
Preprocessing.
If you would have a time win of 20% for using the Hyperthreading, you have a
loose
for doing the Pre-Processing two times. I have no idea if "preprocessing"
can be "hyperthreaded" or not.
On my Dual-Athlon i've found that Preprocessing produces nearly no CPU-usage
(I used PartixGen - very intresting and fully distributable) - so I would
say give it a try.
With real SMP-Systems (my Athlon here) the situation is diffrent.
2. SMPOV and the Renderagent need also some calculation- and reaction time.
They are primarily made for pictures which need long time to render so
the "builtin reaction-times" are rather slow to save CPU-time.
Reaction time is, between SMPOV giving the Orders for Rendering out till
the RenderAgents react.
Secondly till they notice that POV-Ray is ready. Then thirdly SMPOV has to
get a message that all
the tiles are ready. And fourth it has to start PicPender.exe to "put al
lthe tiles together".
There are several reaction times that may add up to 5 Seconds. Add your
loading times and some
time for "Preprocessing"
3. Two instances of POV-Ray are started. (=Loading time for POV-Ray).
4. To find out where the times are, you can render it with SMPOV, make the
"Tiling" on 1
(so there will be only 1 instance started). Then make a second try (tling
on 2) - try "H" and "V"
there are diffrences. Then you have numbers in the log-file. These
numbers include some overhead,
but at least you are shure the settings are really the same for all runs.
Keep me informed if you make a second try.
--Theo
-----------------------------------------------------------------------
Distributed Network-Rendering or Local SMP-Rendering. With SMPOV and
POV-Ray 3.5. * Download free at: http://www.it-berater.org/smpov.htm
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
