Do you want to share your patch? I could test it on my machine with 0.46 and 0.47
-----Ursprüngliche Nachricht----- Gesendet: Sonntag, 26 Juni 2016 um 13:27:23 Uhr Von: "Dario Sanfilippo" sanfilippo.dario@gmail.com An: pd-list pd-list@iem.at Betreff: [PD] Experiencing a higher CPU load with 0.47-0 and 0.47-1. Hi, list.
I'm loading the same patch with 0.46-7, 0.47-0 and 0.47-1 - all 64bit. The last two have a significantly higher CPU load. I'm on OSX 10.11.5.
Has any of you experienced anything similar?
I haven't changed my [vd~] objects into [delread4~], are they calling the same piece of code?
The patch is almost exclusively using signal objects, have some of these been modified in 0.47-0 and 0.47-1?
Thanks for your help.
Dario _______________________________________________ Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> https://lists.puredata.info/listinfo/pd-list
hello,
how are you doing cpu load measurement?
I find it very hard to do reliable measurement of cpu load nowadays, since computer have a variable cpu speed depending on load.
For exemple, pd CPU load can be at 75%, with CPU frequency at 800MHz. When increasing the patch complexities, the CPU frequency increase, and the apparent load reported by pd decrease.
On linux, you can bloc the processor to a fixed frequency, and then make reliable load measurement. But i don't know how to do than on OSX. Did you find a way? otherwise, your measurement are useless.
cheers c
Le 27/06/2016 10:44, christof.ressi@gmx.at a écrit :
Do you want to share your patch? I could test it on my machine with 0.46 and 0.47
-----Ursprüngliche Nachricht----- Gesendet: Sonntag, 26 Juni 2016 um 13:27:23 Uhr Von: "Dario Sanfilippo" sanfilippo.dario@gmail.com An: pd-list pd-list@iem.at Betreff: [PD] Experiencing a higher CPU load with 0.47-0 and 0.47-1. Hi, list.
I'm loading the same patch with 0.46-7, 0.47-0 and 0.47-1 - all 64bit. The last two have a significantly higher CPU load. I'm on OSX 10.11.5.
Has any of you experienced anything similar?
I haven't changed my [vd~] objects into [delread4~], are they calling the same piece of code?
The patch is almost exclusively using signal objects, have some of these been modified in 0.47-0 and 0.47-1?
Thanks for your help.
Dario _______________________________________________ Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> https://lists.puredata.info/listinfo/pd-list
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> https://lists.puredata.info/listinfo/pd-list
Hi, Christof.
It is a rather large project and relatively new, so I'd prefer not to share it at this point as it still kind of a work in progress. I will try putting together some test patches isolating some of the most used objects and see if there's any significant change in the different PD versions when instantiating many of them.
Cyrille: I'm just using PD's Load Meter patch. The test I performed had had just the patch on, without me doing anything. In 0.46-7, the average CPU load when turning DSP on is around 40-50%, with peaks at about 60-70% when acting on the patch. No dropouts experienced. In 0.47, the initial CPU load is around 60% or more and it gets to the point of producing audio dropouts when acting on the patch. So, empirically, 0.47 does seem to have a different CPU load.
I can see the same behaviour by looking at Activity Monitor on OSX. I wouldn't know how else to measure the CPU load, though.
Thanks for your help, guys.
Dario
On 27 June 2016 at 10:00, cyrille henry ch@chnry.net wrote:
hello,
how are you doing cpu load measurement?
I find it very hard to do reliable measurement of cpu load nowadays, since computer have a variable cpu speed depending on load.
For exemple, pd CPU load can be at 75%, with CPU frequency at 800MHz. When increasing the patch complexities, the CPU frequency increase, and the apparent load reported by pd decrease.
On linux, you can bloc the processor to a fixed frequency, and then make reliable load measurement. But i don't know how to do than on OSX. Did you find a way? otherwise, your measurement are useless.
cheers c
Le 27/06/2016 10:44, christof.ressi@gmx.at a écrit :
Do you want to share your patch? I could test it on my machine with 0.46 and 0.47
-----Ursprüngliche Nachricht----- Gesendet: Sonntag, 26 Juni 2016 um 13:27:23 Uhr Von: "Dario Sanfilippo" sanfilippo.dario@gmail.com An: pd-list pd-list@iem.at Betreff: [PD] Experiencing a higher CPU load with 0.47-0 and 0.47-1. Hi, list.
I'm loading the same patch with 0.46-7, 0.47-0 and 0.47-1 - all 64bit. The last two have a significantly higher CPU load. I'm on OSX 10.11.5.
Has any of you experienced anything similar?
I haven't changed my [vd~] objects into [delread4~], are they calling the same piece of code?
The patch is almost exclusively using signal objects, have some of these been modified in 0.47-0 and 0.47-1?
Thanks for your help.
Dario _______________________________________________ Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> https://lists.puredata.info/listinfo/pd-list
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> https://lists.puredata.info/listinfo/pd-list
Le 27/06/2016 11:58, Dario Sanfilippo a écrit :
Hi, Christof.
It is a rather large project and relatively new, so I'd prefer not to share it at this point as it still kind of a work in progress. I will try putting together some test patches isolating some of the most used objects and see if there's any significant change in the different PD versions when instantiating many of them.
Cyrille: I'm just using PD's Load Meter patch. The test I performed had had just the patch on, without me doing anything. In 0.46-7, the average CPU load when turning DSP on is around 40-50%, with peaks at about 60-70% when acting on the patch. No dropouts experienced. In 0.47, the initial CPU load is around 60% or more and it gets to the point of producing audio dropouts when acting on the patch. So, empirically, 0.47 does seem to have a different CPU load.
different cpu load: yes, but since you don't know the cpu frequency, you can't know if it's a higher load, a lower load, and if it's a significative change.
I can see the same behaviour by looking at Activity Monitor on OSX. I wouldn't know how else to measure the CPU load, though.
i'm afraid it's the same problem with activity monitor.
cheers c
Thanks for your help, guys.
Dario
On 27 June 2016 at 10:00, cyrille henry <ch@chnry.net mailto:ch@chnry.net> wrote:
hello, how are you doing cpu load measurement? I find it very hard to do reliable measurement of cpu load nowadays, since computer have a variable cpu speed depending on load. For exemple, pd CPU load can be at 75%, with CPU frequency at 800MHz. When increasing the patch complexities, the CPU frequency increase, and the apparent load reported by pd decrease. On linux, you can bloc the processor to a fixed frequency, and then make reliable load measurement. But i don't know how to do than on OSX. Did you find a way? otherwise, your measurement are useless. cheers c Le 27/06/2016 10:44, christof.ressi@gmx.at <mailto:christof.ressi@gmx.at> a écrit : Do you want to share your patch? I could test it on my machine with 0.46 and 0.47 -----Ursprüngliche Nachricht----- Gesendet: Sonntag, 26 Juni 2016 um 13:27:23 Uhr Von: "Dario Sanfilippo" <sanfilippo.dario@gmail.com <mailto:sanfilippo.dario@gmail.com>> An: pd-list <pd-list@iem.at <mailto:pd-list@iem.at>> Betreff: [PD] Experiencing a higher CPU load with 0.47-0 and 0.47-1. Hi, list. I'm loading the same patch with 0.46-7, 0.47-0 and 0.47-1 - all 64bit. The last two have a significantly higher CPU load. I'm on OSX 10.11.5. Has any of you experienced anything similar? I haven't changed my [vd~] objects into [delread4~], are they calling the same piece of code? The patch is almost exclusively using signal objects, have some of these been modified in 0.47-0 and 0.47-1? Thanks for your help. Dario _______________________________________________ Pd-list@lists.iem.at <mailto:Pd-list@lists.iem.at> mailing list UNSUBSCRIBE and account-management -> https://lists.puredata.info/listinfo/pd-list _______________________________________________ Pd-list@lists.iem.at <mailto:Pd-list@lists.iem.at> mailing list UNSUBSCRIBE and account-management -> https://lists.puredata.info/listinfo/pd-list
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> https://lists.puredata.info/listinfo/pd-list
Yes, the whole thing is baffling, but I gather something changed from 0.46 to 0.47 ... I've gt a coupld of benchmark patches I can try to see if I can see what's going on.
cheers Miller
On Mon, Jun 27, 2016 at 12:14:56PM +0200, cyrille henry wrote:
Le 27/06/2016 11:58, Dario Sanfilippo a écrit :
Hi, Christof.
It is a rather large project and relatively new, so I'd prefer not to share it at this point as it still kind of a work in progress. I will try putting together some test patches isolating some of the most used objects and see if there's any significant change in the different PD versions when instantiating many of them.
Cyrille: I'm just using PD's Load Meter patch. The test I performed had had just the patch on, without me doing anything. In 0.46-7, the average CPU load when turning DSP on is around 40-50%, with peaks at about 60-70% when acting on the patch. No dropouts experienced. In 0.47, the initial CPU load is around 60% or more and it gets to the point of producing audio dropouts when acting on the patch. So, empirically, 0.47 does seem to have a different CPU load.
different cpu load: yes, but since you don't know the cpu frequency, you can't know if it's a higher load, a lower load, and if it's a significative change.
I can see the same behaviour by looking at Activity Monitor on OSX. I wouldn't know how else to measure the CPU load, though.
i'm afraid it's the same problem with activity monitor.
cheers c
Thanks for your help, guys.
Dario
On 27 June 2016 at 10:00, cyrille henry <ch@chnry.net mailto:ch@chnry.net> wrote:
hello,
how are you doing cpu load measurement?
I find it very hard to do reliable measurement of cpu load nowadays, since computer have a variable cpu speed depending on load.
For exemple, pd CPU load can be at 75%, with CPU frequency at 800MHz. When increasing the patch complexities, the CPU frequency increase, and the apparent load reported by pd decrease.
On linux, you can bloc the processor to a fixed frequency, and then make reliable load measurement. But i don't know how to do than on OSX. Did you find a way? otherwise, your measurement are useless.
cheers c
Le 27/06/2016 10:44, christof.ressi@gmx.at mailto:christof.ressi@gmx.at a écrit :
Do you want to share your patch? I could test it on my machine with 0.46 and 0.47 -----Ursprüngliche Nachricht----- Gesendet: Sonntag, 26 Juni 2016 um 13:27:23 Uhr Von: "Dario Sanfilippo" <sanfilippo.dario@gmail.com <mailto:sanfilippo.dario@gmail.com>> An: pd-list <pd-list@iem.at <mailto:pd-list@iem.at>> Betreff: [PD] Experiencing a higher CPU load with 0.47-0 and 0.47-1. Hi, list. I'm loading the same patch with 0.46-7, 0.47-0 and 0.47-1 - all 64bit. The last two have a significantly higher CPU load. I'm on OSX 10.11.5. Has any of you experienced anything similar? I haven't changed my [vd~] objects into [delread4~], are they calling the same piece of code? The patch is almost exclusively using signal objects, have some of these been modified in 0.47-0 and 0.47-1? Thanks for your help. Dario _______________________________________________ Pd-list@lists.iem.at <mailto:Pd-list@lists.iem.at> mailing list UNSUBSCRIBE and account-management -> https://lists.puredata.info/listinfo/pd-list _______________________________________________ Pd-list@lists.iem.at <mailto:Pd-list@lists.iem.at> mailing list UNSUBSCRIBE and account-management -> https://lists.puredata.info/listinfo/pd-list
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> https://lists.puredata.info/listinfo/pd-list
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> https://lists.puredata.info/listinfo/pd-list
I just loaded a nice fat benchmark patch (based on smeck, the guitar processor) in a few different versions of Pd. I got no difference between Pd-0.46-7 and pd-0.47-1 ... however, in each version the "64 bit" compile ran in about 85% of the CPU load that the non-64-bit version did. Perhaps you're comparing 0.46 634 bit with 0.47 32 bit?
cheers Miller
On Mon, Jun 27, 2016 at 09:19:35AM -0700, Miller Puckette wrote:
Yes, the whole thing is baffling, but I gather something changed from 0.46 to 0.47 ... I've gt a coupld of benchmark patches I can try to see if I can see what's going on.
cheers Miller
On Mon, Jun 27, 2016 at 12:14:56PM +0200, cyrille henry wrote:
Le 27/06/2016 11:58, Dario Sanfilippo a écrit :
Hi, Christof.
It is a rather large project and relatively new, so I'd prefer not to share it at this point as it still kind of a work in progress. I will try putting together some test patches isolating some of the most used objects and see if there's any significant change in the different PD versions when instantiating many of them.
Cyrille: I'm just using PD's Load Meter patch. The test I performed had had just the patch on, without me doing anything. In 0.46-7, the average CPU load when turning DSP on is around 40-50%, with peaks at about 60-70% when acting on the patch. No dropouts experienced. In 0.47, the initial CPU load is around 60% or more and it gets to the point of producing audio dropouts when acting on the patch. So, empirically, 0.47 does seem to have a different CPU load.
different cpu load: yes, but since you don't know the cpu frequency, you can't know if it's a higher load, a lower load, and if it's a significative change.
I can see the same behaviour by looking at Activity Monitor on OSX. I wouldn't know how else to measure the CPU load, though.
i'm afraid it's the same problem with activity monitor.
cheers c
Thanks for your help, guys.
Dario
On 27 June 2016 at 10:00, cyrille henry <ch@chnry.net mailto:ch@chnry.net> wrote:
hello,
how are you doing cpu load measurement?
I find it very hard to do reliable measurement of cpu load nowadays, since computer have a variable cpu speed depending on load.
For exemple, pd CPU load can be at 75%, with CPU frequency at 800MHz. When increasing the patch complexities, the CPU frequency increase, and the apparent load reported by pd decrease.
On linux, you can bloc the processor to a fixed frequency, and then make reliable load measurement. But i don't know how to do than on OSX. Did you find a way? otherwise, your measurement are useless.
cheers c
Le 27/06/2016 10:44, christof.ressi@gmx.at mailto:christof.ressi@gmx.at a écrit :
Do you want to share your patch? I could test it on my machine with 0.46 and 0.47 -----Ursprüngliche Nachricht----- Gesendet: Sonntag, 26 Juni 2016 um 13:27:23 Uhr Von: "Dario Sanfilippo" <sanfilippo.dario@gmail.com <mailto:sanfilippo.dario@gmail.com>> An: pd-list <pd-list@iem.at <mailto:pd-list@iem.at>> Betreff: [PD] Experiencing a higher CPU load with 0.47-0 and 0.47-1. Hi, list. I'm loading the same patch with 0.46-7, 0.47-0 and 0.47-1 - all 64bit. The last two have a significantly higher CPU load. I'm on OSX 10.11.5. Has any of you experienced anything similar? I haven't changed my [vd~] objects into [delread4~], are they calling the same piece of code? The patch is almost exclusively using signal objects, have some of these been modified in 0.47-0 and 0.47-1? Thanks for your help. Dario _______________________________________________ Pd-list@lists.iem.at <mailto:Pd-list@lists.iem.at> mailing list UNSUBSCRIBE and account-management -> https://lists.puredata.info/listinfo/pd-list _______________________________________________ Pd-list@lists.iem.at <mailto:Pd-list@lists.iem.at> mailing list UNSUBSCRIBE and account-management -> https://lists.puredata.info/listinfo/pd-list
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> https://lists.puredata.info/listinfo/pd-list
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> https://lists.puredata.info/listinfo/pd-list
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> https://lists.puredata.info/listinfo/pd-list
Hi, Miller.
The comparison was between 64bit versions of the software. Like a mentioned in another email, there was ~15% higher CPU load (however accurate that estimation is) in .47 when running 512 instantiations of a simple patch with an [osc~]-driven [vd~] and [delwrite~]. I can try putting together a list of the most used objects in my project to narrow down any potential problem.
Cheers, Dario
On Wed, 3 Aug 2016 at 01:21 Miller Puckette msp@ucsd.edu wrote:
I just loaded a nice fat benchmark patch (based on smeck, the guitar processor) in a few different versions of Pd. I got no difference between Pd-0.46-7 and pd-0.47-1 ... however, in each version the "64 bit" compile ran in about 85% of the CPU load that the non-64-bit version did. Perhaps you're comparing 0.46 634 bit with 0.47 32 bit?
cheers Miller
On Mon, Jun 27, 2016 at 09:19:35AM -0700, Miller Puckette wrote:
Yes, the whole thing is baffling, but I gather something changed from
0.46
to 0.47 ... I've gt a coupld of benchmark patches I can try to see if I
can
see what's going on.
cheers Miller
On Mon, Jun 27, 2016 at 12:14:56PM +0200, cyrille henry wrote:
Le 27/06/2016 11:58, Dario Sanfilippo a écrit :
Hi, Christof.
It is a rather large project and relatively new, so I'd prefer not to
share it at this point as it still kind of a work in progress. I will try putting together some test patches isolating some of the most used objects and see if there's any significant change in the different PD versions when instantiating many of them.
Cyrille: I'm just using PD's Load Meter patch. The test I performed
had had just the patch on, without me doing anything. In 0.46-7, the average CPU load when turning DSP on is around 40-50%, with peaks at about 60-70% when acting on the patch. No dropouts experienced. In 0.47, the initial CPU load is around 60% or more and it gets to the point of producing audio dropouts when acting on the patch. So, empirically, 0.47 does seem to have a different CPU load.
different cpu load: yes, but since you don't know the cpu frequency,
you can't know if it's a higher load, a lower load, and if it's a significative change.
I can see the same behaviour by looking at Activity Monitor on OSX. I
wouldn't know how else to measure the CPU load, though.
i'm afraid it's the same problem with activity monitor.
cheers c
Thanks for your help, guys.
Dario
On 27 June 2016 at 10:00, cyrille henry <ch@chnry.net <mailto:
ch@chnry.net>> wrote:
hello,
how are you doing cpu load measurement?
I find it very hard to do reliable measurement of cpu load
nowadays, since computer have a variable cpu speed depending on load.
For exemple, pd CPU load can be at 75%, with CPU frequency at
800MHz. When increasing the patch complexities, the CPU frequency increase, and the apparent load reported by pd decrease.
On linux, you can bloc the processor to a fixed frequency, and
then make reliable load measurement.
But i don't know how to do than on OSX. Did you find a way? otherwise, your measurement are useless.
cheers c
Le 27/06/2016 10:44, christof.ressi@gmx.at <mailto:
christof.ressi@gmx.at> a écrit :
Do you want to share your patch? I could test it on mymachine with 0.46 and 0.47
-----Ursprüngliche Nachricht----- Gesendet: Sonntag, 26 Juni 2016 um 13:27:23 Uhr Von: "Dario Sanfilippo" <sanfilippo.dario@gmail.com <mailto:sanfilippo.dario@gmail.com>>
An: pd-list <pd-list@iem.at <mailto:pd-list@iem.at>> Betreff: [PD] Experiencing a higher CPU load with 0.47-0 and0.47-1.
Hi, list. I'm loading the same patch with 0.46-7, 0.47-0 and 0.47-1 -all 64bit. The
last two have a significantly higher CPU load. I'm on OSX10.11.5.
Has any of you experienced anything similar? I haven't changed my [vd~] objects into [delread4~], are theycalling the
same piece of code? The patch is almost exclusively using signal objects, havesome of these
been modified in 0.47-0 and 0.47-1? Thanks for your help. Dario _______________________________________________ Pd-list@lists.iem.at <mailto:Pd-list@lists.iem.at> mailinglist
UNSUBSCRIBE and account-management ->https://lists.puredata.info/listinfo/pd-list
_______________________________________________ Pd-list@lists.iem.at <mailto:Pd-list@lists.iem.at> mailinglist
UNSUBSCRIBE and account-management ->https://lists.puredata.info/listinfo/pd-list
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management ->
https://lists.puredata.info/listinfo/pd-list
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management ->
https://lists.puredata.info/listinfo/pd-list
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management ->
https://lists.puredata.info/listinfo/pd-list
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> https://lists.puredata.info/listinfo/pd-list
Aha... I tried again with a patch with lots of vd~/delwrite pairs and got 47 taking about 10% more CPU than 0.46. (That I didn't get 15 could be that I had a different mix of objects than yours.) So something is wrong... I don't know what yet. (I did fix a small bug in delay reading/writing that could be affecting this someho but I can't imagine how :)
M
On Wed, Aug 03, 2016 at 09:17:28AM +0000, Dario Sanfilippo wrote:
Hi, Miller.
The comparison was between 64bit versions of the software. Like a mentioned in another email, there was ~15% higher CPU load (however accurate that estimation is) in .47 when running 512 instantiations of a simple patch with an [osc~]-driven [vd~] and [delwrite~]. I can try putting together a list of the most used objects in my project to narrow down any potential problem.
Cheers, Dario
On Wed, 3 Aug 2016 at 01:21 Miller Puckette msp@ucsd.edu wrote:
I just loaded a nice fat benchmark patch (based on smeck, the guitar processor) in a few different versions of Pd. I got no difference between Pd-0.46-7 and pd-0.47-1 ... however, in each version the "64 bit" compile ran in about 85% of the CPU load that the non-64-bit version did. Perhaps you're comparing 0.46 634 bit with 0.47 32 bit?
cheers Miller
On Mon, Jun 27, 2016 at 09:19:35AM -0700, Miller Puckette wrote:
Yes, the whole thing is baffling, but I gather something changed from
0.46
to 0.47 ... I've gt a coupld of benchmark patches I can try to see if I
can
see what's going on.
cheers Miller
On Mon, Jun 27, 2016 at 12:14:56PM +0200, cyrille henry wrote:
Le 27/06/2016 11:58, Dario Sanfilippo a écrit :
Hi, Christof.
It is a rather large project and relatively new, so I'd prefer not to
share it at this point as it still kind of a work in progress. I will try putting together some test patches isolating some of the most used objects and see if there's any significant change in the different PD versions when instantiating many of them.
Cyrille: I'm just using PD's Load Meter patch. The test I performed
had had just the patch on, without me doing anything. In 0.46-7, the average CPU load when turning DSP on is around 40-50%, with peaks at about 60-70% when acting on the patch. No dropouts experienced. In 0.47, the initial CPU load is around 60% or more and it gets to the point of producing audio dropouts when acting on the patch. So, empirically, 0.47 does seem to have a different CPU load.
different cpu load: yes, but since you don't know the cpu frequency,
you can't know if it's a higher load, a lower load, and if it's a significative change.
I can see the same behaviour by looking at Activity Monitor on OSX. I
wouldn't know how else to measure the CPU load, though.
i'm afraid it's the same problem with activity monitor.
cheers c
Thanks for your help, guys.
Dario
On 27 June 2016 at 10:00, cyrille henry <ch@chnry.net <mailto:
ch@chnry.net>> wrote:
hello,
how are you doing cpu load measurement?
I find it very hard to do reliable measurement of cpu load
nowadays, since computer have a variable cpu speed depending on load.
For exemple, pd CPU load can be at 75%, with CPU frequency at
800MHz. When increasing the patch complexities, the CPU frequency increase, and the apparent load reported by pd decrease.
On linux, you can bloc the processor to a fixed frequency, and
then make reliable load measurement.
But i don't know how to do than on OSX. Did you find a way? otherwise, your measurement are useless.
cheers c
Le 27/06/2016 10:44, christof.ressi@gmx.at <mailto:
christof.ressi@gmx.at> a écrit :
Do you want to share your patch? I could test it on mymachine with 0.46 and 0.47
-----Ursprüngliche Nachricht----- Gesendet: Sonntag, 26 Juni 2016 um 13:27:23 Uhr Von: "Dario Sanfilippo" <sanfilippo.dario@gmail.com <mailto:sanfilippo.dario@gmail.com>>
An: pd-list <pd-list@iem.at <mailto:pd-list@iem.at>> Betreff: [PD] Experiencing a higher CPU load with 0.47-0 and0.47-1.
Hi, list. I'm loading the same patch with 0.46-7, 0.47-0 and 0.47-1 -all 64bit. The
last two have a significantly higher CPU load. I'm on OSX10.11.5.
Has any of you experienced anything similar? I haven't changed my [vd~] objects into [delread4~], are theycalling the
same piece of code? The patch is almost exclusively using signal objects, havesome of these
been modified in 0.47-0 and 0.47-1? Thanks for your help. Dario _______________________________________________ Pd-list@lists.iem.at <mailto:Pd-list@lists.iem.at> mailinglist
UNSUBSCRIBE and account-management ->https://lists.puredata.info/listinfo/pd-list
_______________________________________________ Pd-list@lists.iem.at <mailto:Pd-list@lists.iem.at> mailinglist
UNSUBSCRIBE and account-management ->https://lists.puredata.info/listinfo/pd-list
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management ->
https://lists.puredata.info/listinfo/pd-list
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management ->
https://lists.puredata.info/listinfo/pd-list
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management ->
https://lists.puredata.info/listinfo/pd-list
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> https://lists.puredata.info/listinfo/pd-list
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> https://lists.puredata.info/listinfo/pd-list
Hi, Miller. I'm glad to see that we're getting there.
If the same "object-problem" could result in different CPU load differences between .46 and .47 depending on the context they're in, then maybe delwrite~ and/or vd~ alone could explain the ~20% difference I had loading the same patch with the two versions of the software.
Most of the objects in my project include audio math objects (arithmetic, pow, cos, pow, abs, wrap, samphold), but I doubt that these have changed. I also have quite a few tabwrite~ and tabread4~ objects, how about these ones?
Thanks, Dario
On 3 August 2016 at 19:47, Miller Puckette msp@ucsd.edu wrote:
Aha... I tried again with a patch with lots of vd~/delwrite pairs and got 47 taking about 10% more CPU than 0.46. (That I didn't get 15 could be that I had a different mix of objects than yours.) So something is wrong... I don't know what yet. (I did fix a small bug in delay reading/writing that could be affecting this someho but I can't imagine how :)
M
On Wed, Aug 03, 2016 at 09:17:28AM +0000, Dario Sanfilippo wrote:
Hi, Miller.
The comparison was between 64bit versions of the software. Like a
mentioned
in another email, there was ~15% higher CPU load (however accurate that estimation is) in .47 when running 512 instantiations of a simple patch with an [osc~]-driven [vd~] and [delwrite~]. I can try putting together a list of the most used objects in my project to narrow down any potential problem.
Cheers, Dario
On Wed, 3 Aug 2016 at 01:21 Miller Puckette msp@ucsd.edu wrote:
I just loaded a nice fat benchmark patch (based on smeck, the guitar processor) in a few different versions of Pd. I got no difference
between
Pd-0.46-7 and pd-0.47-1 ... however, in each version the "64 bit"
compile
ran in about 85% of the CPU load that the non-64-bit version did.
Perhaps
you're comparing 0.46 634 bit with 0.47 32 bit?
cheers Miller
On Mon, Jun 27, 2016 at 09:19:35AM -0700, Miller Puckette wrote:
Yes, the whole thing is baffling, but I gather something changed from
0.46
to 0.47 ... I've gt a coupld of benchmark patches I can try to see
if I
can
see what's going on.
cheers Miller
On Mon, Jun 27, 2016 at 12:14:56PM +0200, cyrille henry wrote:
Le 27/06/2016 11:58, Dario Sanfilippo a écrit :
Hi, Christof.
It is a rather large project and relatively new, so I'd prefer
not to
share it at this point as it still kind of a work in progress. I will
try
putting together some test patches isolating some of the most used
objects
and see if there's any significant change in the different PD versions
when
instantiating many of them.
Cyrille: I'm just using PD's Load Meter patch. The test I
performed
had had just the patch on, without me doing anything. In 0.46-7, the average CPU load when turning DSP on is around 40-50%, with peaks at
about
60-70% when acting on the patch. No dropouts experienced. In 0.47, the initial CPU load is around 60% or more and it gets to the point of producing audio dropouts when acting on the patch. So, empirically,
0.47
does seem to have a different CPU load.
different cpu load: yes, but since you don't know the cpu
frequency,
you can't know if it's a higher load, a lower load, and if it's a significative change.
I can see the same behaviour by looking at Activity Monitor on
OSX. I
wouldn't know how else to measure the CPU load, though.
i'm afraid it's the same problem with activity monitor.
cheers c
Thanks for your help, guys.
Dario
On 27 June 2016 at 10:00, cyrille henry <ch@chnry.net <mailto:
ch@chnry.net>> wrote:
hello,
how are you doing cpu load measurement?
I find it very hard to do reliable measurement of cpu load
nowadays, since computer have a variable cpu speed depending on load.
For exemple, pd CPU load can be at 75%, with CPU frequency at
800MHz. When increasing the patch complexities, the CPU frequency
increase,
and the apparent load reported by pd decrease.
On linux, you can bloc the processor to a fixed frequency, and
then make reliable load measurement.
But i don't know how to do than on OSX. Did you find a way? otherwise, your measurement are useless.
cheers c
Le 27/06/2016 10:44, christof.ressi@gmx.at <mailto:
christof.ressi@gmx.at> a écrit :
Do you want to share your patch? I could test it on mymachine with 0.46 and 0.47
-----Ursprüngliche Nachricht----- Gesendet: Sonntag, 26 Juni 2016 um 13:27:23 Uhr Von: "Dario Sanfilippo" <sanfilippo.dario@gmail.com<mailto:
sanfilippo.dario@gmail.com>>
An: pd-list <pd-list@iem.at <mailto:pd-list@iem.at>> Betreff: [PD] Experiencing a higher CPU load with 0.47-0and
0.47-1.
Hi, list. I'm loading the same patch with 0.46-7, 0.47-0 and 0.47-1
all 64bit. The
last two have a significantly higher CPU load. I'm on OSX10.11.5.
Has any of you experienced anything similar? I haven't changed my [vd~] objects into [delread4~], arethey
calling the
same piece of code? The patch is almost exclusively using signal objects, havesome of these
been modified in 0.47-0 and 0.47-1? Thanks for your help. Dario _______________________________________________ Pd-list@lists.iem.at <mailto:Pd-list@lists.iem.at>mailing
list
UNSUBSCRIBE and account-management ->https://lists.puredata.info/listinfo/pd-list
_______________________________________________ Pd-list@lists.iem.at <mailto:Pd-list@lists.iem.at>mailing
list
UNSUBSCRIBE and account-management ->https://lists.puredata.info/listinfo/pd-list
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management ->
https://lists.puredata.info/listinfo/pd-list
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management ->
https://lists.puredata.info/listinfo/pd-list
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management ->
https://lists.puredata.info/listinfo/pd-list
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> https://lists.puredata.info/listinfo/pd-list
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> https://lists.puredata.info/
listinfo/pd-list
-
christof.ressi＠gmx.at -
cyrille henry -
Dario Sanfilippo -
Miller Puckette