On 2017-10-15 15:39:59 +0000, William F Pokorny said:

> On 10/14/2017 01:25 PM, E-mailyvo s gmx.net wrote:
>> On 2017-10-13 16:19:13 +0000, pkoning said:
>> 
>> 
>> I see a completely different result, 3.8.0 is 5 to 6 times *slower* than 3.7.0.
>> 
>> Benchmark.pov scene with exactly the same settings:
>> 3.8.0 >> 583,35 seconds
>> 3.7.0 >> 114,76 seconds
>> 
>> Glasschess.pov scene:
>> 3.8.0 >> 143,23 seconds
>> 3.7.0 >>   23,85 seconds
>> 
>> A while back I noticed that 3.7.1 was a bit slower then 3.7.0 but not 
>> as extreme as the difference between 3.7.0 and 3.8.0
>> 
>> I verified the compiler options and I made no mistake there, optimized 
>> settings for release.
>> 
>> It would be good to know if the Windows or/and Linux version show the 
>> same difference between 3.7.0 and 3.8.0
>> 
>> BTW: your 3.7.0 version isn’t the latest, “r9” from august 2017 is the 
>> last version (compiler used is LLVM 8.1.0 (clang-802.0.42)):
>> http://megapov.inetart.net/povrayunofficial_mac/finalpov.html
>> 
>> -- 
>> Yvo
>> 
> 
> I'm running 3.7.0 and 3.8.0 versions from early July. I see the 
> following results for glasschess.pov which are in line with what we 
> expect - a slight slowdown 3.7.0 to 3.8.0 and one which is a little 
> worse when threads > cores; Excepting, where new intel/amd optimized 
> noise is used (new to *nix compiles) and helps substantially. Where the 
> new noise code helps we are usually faster - though with that code too 
> - we have seen a few scenes coming in slower, but not run down why 
> those particular ones slower with the new noise code.
> 
> I am running:
> 
> Ubuntu 16.04
> g++ 5.4.0 20160609
> i3-4130 CPU @ 3.40GHz
> 
> On what hardware are you guys running?
> 
> What happens if you use +wt<n> to run threads<=cores?
> 
> Note. These days timing is very sensitive to how the cache(s) on the 
> system are being used when you do timing renders. 5-6x slower seems 
> extreme for cache impacts. I've seen as much as about 2x slower on a 
> busy system on occasion, but never more - and you are seeing similar 
> across multiple renders...
> 
> glasschess.pov
> 
> 3.7.0
> -------------------
> 65.86user 0.03system 0:17.53elapsed  +wt4
> 65.97user 0.00system 0:17.53elapsed  +wt4
> 131.83
> 
> 42.68user 0.02system 0:22.12elapsed  +wt2
> 42.68user 0.02system 0:22.12elapsed  +wt2
> 43.44user 0.02system 0:22.54elapsed  +wt2
> 128.8
> 
> 3.8.0
> -------------------
> 67.51user 0.04system 0:17.98elapsed +wt4
> 67.33user 0.08system 0:17.93elapsed +wt4
> 134.84  +2.28% slower
> 
> 44.05user 0.04system 0:22.90elapsed +wt2
> 42.90user 0.04system 0:22.27elapsed +wt2
> 43.10user 0.02system 0:22.38elapsed +wt2
> 130.05  +0.97% slower
> 
> Perhaps related: Early this year - I think - Dick Balaska reported 
> seeing much, much slower results for a 3.7.1 version on linux for 
> frames of his animation hitting a certain linux machine in his 
> collection of render machines.
> 
> Bill P.

Tanks for your report!
The slowdown is related to the new compiler version (Xcode 9), somehow 
the compiler isn’t optimizing the code for this project.

I’ll revert back to Xcode 8 until I know what is causing this.

-- 
Yvo

-- 
-------------------------------------------------------------------------------------------


POV-Ray 3.7 unofficial: http://megapov.inetart.net/povrayunofficial_mac/
UberPOV Mac: http://megapov.inetart.net/uberpov_mac/index.html#Mac
MegaPOV: http://megapov.inetart.net
E-mail: yvo(DOT)s(AT)gmx.net

Post a reply to this message

Attachments:
Download 'utf-8' (7 KB)