On 2/11/2016 7:36 AM, Sven Littkowski wrote:
> Hmmmm, this sounds to me that here is another good option for a useful
> improve of POV-Ray's capabilities. I should go on the nerves of CLipka
> with it (he'll love to hear from me again)...
>
Or you could get yourself a computer with a modern FPU.  Celeron was 
designed to be "cheap" and they did that by having a weak FPU and no 
cache. (The theory was Word(tm) doesn't need an FPU).
You are trying to do the most complicated things with the least powerful 
machine out there.

Post a reply to this message

Yes. All my powerful machines were stolen. And I have not yet recovered
enough, to get similar machines back...

Post a reply to this message

hi,

On 11/02/2016 14:00, dick balaska wrote:
> Or you could get yourself a computer with a modern FPU.  Celeron was
> designed to be "cheap" and they did that by having a weak FPU and no
> cache.

my elderly Celeron-based system needs replacing soon, which CPU would
you (and other readers) recommend?

-- 
regards, jr.

Post a reply to this message

On 2/12/2016 6:38 AM, jr wrote:
> hi,
>
> On 11/02/2016 14:00, dick balaska wrote:
>> Or you could get yourself a computer with a modern FPU.  Celeron was
>> designed to be "cheap" and they did that by having a weak FPU and no
>> cache.
>
> my elderly Celeron-based system needs replacing soon, which CPU would
> you (and other readers) recommend?
>
i5's are "cheap".  $500 for a decent system.  That will give you 4 cores.
I just priced an i7 box for just under $1000. The i7's are all "double 
threaded" which means 2 instructions at a time can be processed.  So in 
a 4 core CPU you get a virtual 8 cores.
My (4) core i7 box averages [1] twice as fast as my (4) core i5(s).
( http://www.buckosoft.com/bsac/ or
   http://www.buckosoft.com/bsac/meta/ )

If you do a lot of POV-Ray and can afford it, get a 6 core i7 or even 
the 8 core i7.

Intel's new number system sucks and you can't know your players without 
a scorecard.  I use this one.
https://en.wikipedia.org/wiki/Intel_Core

The only thing being an i7 means is it is double threaded.  There is 
still a *huge* variance between different models of i7. You could get 
between 2 and 8 cores. So make sure you check the model number of the 
chip!  BestBuy advertises boxes with "latest generation Intel(R)(C) i7 
technology!"  This is not enough information to make an informed decision.

You can google "i7-4960x vs i7-4930k" and cpuboss.com will give you 
graphs comparing the two chips.

If you can afford to be top dog, then you want the i7-5960X.

dik

[1] I have scenes where the i7 is only 20% faster than the i5. I have 
other scenes where the i7 is 10x faster(!) than the i5. (I don't know if 
that is a Windows (i7) vs. Linux (i5) issue.)

Post a reply to this message

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA256

Le 12/02/2016 16:46, dick balaska a écrit :
> 
> [1] I have scenes where the i7 is only 20% faster than the i5. I
> have other scenes where the i7 is 10x faster(!) than the i5. (I
> don't know if that is a Windows (i7) vs. Linux (i5) issue.)

My measures with i7 and povray (Linux): the HT-core is worth only 20
to 25% percent of a real core. At least for benchmark scene.

The compiler also play a significant role, and the intensive memory
access of a scene might too (if you are lucking enough for your parsed
scene to fit in the L3 cache of the CPU, big win): remember, despite
the number of cores, there is only one bus to reach the RAM.

http://wiki.povray.org/content/User:Le_Forgeron/vault/Compilation


-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2

iJwEAQEIAAYFAla+C+kACgkQhKAm8mTpkW2BJQP/Rd3WMMvYuQqxV+gxYzZCZCir
ZHhLeTq1ajk5Yg9M2tdgb2O9EXkW/RVta3ZSEk0DjC+zv0AFtZLX7k7vo7Bp9HQq
p/sokek+MVhJ1lKxbTc4m/8AyphKWntURkaRoWhcLfJ+vkFuIlaWErx0id6Qv/qx
xOpyqMRepGReCYKD2+8=
=54Qf
-----END PGP SIGNATURE-----

Post a reply to this message

On 2/12/2016 11:44 AM, Le_Forgeron wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA256
>
> Le 12/02/2016 16:46, dick balaska a écrit :
>>
>> [1] I have scenes where the i7 is only 20% faster than the i5. I
>> have other scenes where the i7 is 10x faster(!) than the i5. (I
>> don't know if that is a Windows (i7) vs. Linux (i5) issue.)
>
> My measures with i7 and povray (Linux): the HT-core is worth only 20
> to 25% percent of a real core. At least for benchmark scene.
>
> The compiler also play a significant role, and the intensive memory
> access of a scene might too (if you are lucking enough for your parsed
> scene to fit in the L3 cache of the CPU, big win): remember, despite
> the number of cores, there is only one bus to reach the RAM.
>
> http://wiki.povray.org/content/User:Le_Forgeron/vault/Compilation

Interesting.  20 years ago, the Intel compiler was the gold standard for 
Pov-Ray compiles.  I would have thought that gcc would have picked up 
whatever they were doing by now.

dik

Post a reply to this message

Here is an actual CPU-Test ...

http://www.pcgameshardware.de/CPU-Hardware-154106/Tests/CPU-Test-Benchmark-AMD-Intel-675663/

You can also look into the Luxmark-Results to get a good candidate ...
http://www.luxmark.info/top_results/LuxBall%20HDR/OpenCL/CPU

Post a reply to this message

On 2/13/2016 2:33 AM, Theogott wrote:
> Here is an actual CPU-Test ...
>
>
http://www.pcgameshardware.de/CPU-Hardware-154106/Tests/CPU-Test-Benchmark-AMD-Intel-675663/

I don't know...
An 8 core 5960K is only 21% faster than my 4 core 4770K?  That doesn't 
seem right.

Post a reply to this message

On 2/13/2016 8:53 AM, dick balaska wrote:
> On 2/13/2016 2:33 AM, Theogott wrote:
>> Here is an actual CPU-Test ...
>>
>>
http://www.pcgameshardware.de/CPU-Hardware-154106/Tests/CPU-Test-Benchmark-AMD-Intel-675663/
>>
>
> I don't know...
> An 8 core 5960K is only 21% faster than my 4 core 4770K?  That doesn't
> seem right.
>
>
I find that in my scenes changing the Render_Block_Size to about 8*, 
allowing blocks to be released faster, generally speeds up the render. I 
imagine most testing use the default settings.

*
YMMV


-- 

Regards
     Stephen

Post a reply to this message

Am 13.02.2016 um 09:53 schrieb dick balaska:
> On 2/13/2016 2:33 AM, Theogott wrote:
>> Here is an actual CPU-Test ...
>>
>>
http://www.pcgameshardware.de/CPU-Hardware-154106/Tests/CPU-Test-Benchmark-AMD-Intel-675663/
>>
> 
> I don't know...
> An 8 core 5960K is only 21% faster than my 4 core 4770K?  That doesn't
> seem right.

Number of cores isn't the only thing affecting performance. There's also
CPU clock, main memory bandwidth, cache layout and size, instruction
set, and so forth.

Most notably however, the benchmark score reflects a blend of real-life
scenarios; Most pieces of software may not even be prepared to
distribute their workload across 16 threads at all, and others may scale
poorly with the number of available cores due to a non-linear increase
in inter-thread communication overhead.

POV-Ray is a pretty exceptional piece of software, in that it requires
virtually no inter-thread communication (radiosity pretrace being a
notable exception).

Post a reply to this message