Ways to prevent clock speed throttling

[Art Todesco]

I am using a Window 7 machine with an 8 channel audio card to provide
multichannel outputs from a dedicated software application. The
application plays sampled pipe organ audio in response to Midi commands
from musical keyboards. Basically it's a virtual pipe organ. The
problem I'm having is that the processor (quad core AMD) speed changes
due to various things that can happen, such as overheating, etc. This
results is the audio board putting out a click or pop when the processor
speed is adjusted. I have change many parameters and have improved the
situation considerably, however, 1 pop or click is not acceptable. I
know many of the parameters must be adjusted in the BIOS, but I'm sure
there are some things that W7 does that affects this problem. Anyone
have any ideas?
Thanks.

 

[Paul]

I am using a Window 7 machine with an 8 channel audio card to provide
multichannel outputs from a dedicated software application. The
application plays sampled pipe organ audio in response to Midi commands
from musical keyboards. Basically it's a virtual pipe organ. The
problem I'm having is that the processor (quad core AMD) speed changes
due to various things that can happen, such as overheating, etc. This
results is the audio board putting out a click or pop when the processor
speed is adjusted. I have change many parameters and have improved the
situation considerably, however, 1 pop or click is not acceptable. I
know many of the parameters must be adjusted in the BIOS, but I'm sure
there are some things that W7 does that affects this problem. Anyone
have any ideas?
Thanks.

Are you using the "Always On" power schema ? Visit the Power
control panel, and look for a schema that keeps the CPU at
full speed all the time.

Even if the processor was running at its lowest speed under Cool N' Quiet,
it would probably be enough to service the sound card. But what is
bad though, is P-state changes, which might take 100 microseconds or so.
I don't know if the CPU is available to execute instructions during
that transition. During a P-state change, VCore is adjusted and the
multiplier is changed, and the two changes are sequential. (And it
depends on whether the CPU is speeding up or slowing down, as to which
change is done first in the sequence.) When playing a video, I think
AMD systems have been known to make 30 P-state changes per second.

I don't know if that's the nature of where your click of pop is
coming from - a buffer underrun due to that 100 microsecond outage
sounds pretty unlikely to be enough to do it. The buffer is probably
a lot bigger than that, and the threshold should leave plenty of
time for it to get serviced when it needs to be filled up.

*******

Other possibilities, are activities on the computer which are
not even visible as such, from the OS. Such as System Management
Mode or SMM. You can check for SMM activity, in an indirect
measurement way, using DPCLat. Certain Gigabyte boards show
spikes in DPCLat, implying long periods of time spent in SMM.
And, SMM could not be disabled. SMM is used for things, like
adjusting multi-phase VCore designs, while the system is running.
The change to the number of phases being employed, is done via
BIOS code running under SMM. The OS is blissfully unaware it
has been booted from the processor by SMM. (It's possible if you
were in SMM long enough, you could miss a clock tick interrupt.)
And DPCLat, uses the service time of Deferred Procedure Calls, to
indirectly determine something like SMM is happening.

http://www.thesycon.de/deu/latency_check.shtml

http://en.wikipedia.org/wiki/Deferred_Procedure_Call

I have no idea what OSes that program supports. It's a pretty
old program.

If you test out that program, and verify you don't have an SMM
problem, look in Task Manager, and see if you've acquired a
"LtcyCfgSvc.exe" process. It's possible that got installed
on my system when using DPCLat. I'm not really sure, but that's
about all I can associate it with on my main machine.

Typically, people who build audio workstations, test with DPCLat
to see if the motherboard is going to be a problem or not.
Gigabyte has released updates to the BIOS, that reduce spikes in
DPCLat, so it is possible to make improvements of the worst
cases (like, blowing a clock tick), caused by long SMM runtime.

(Some background on SMM, here.)

http://en.wikipedia.org/wiki/System_Management_Mode

*******

This would be an example of a motherboard with a serious problem.
The green bits are good. The red spikes, are not. The original
photo is no longer available, and all I can get is this crappy
thumbnail.

http://ts1.mm.bing.net/th?id=H.4728833101990412&pid=15.1

*******

Note that, there are some transitions on a PC, where there
are unavoidable long delays. I've noticed, entering or
exiting a 3D game, causes a large spike in DPCLat. So if you're
working in a recording studio, with an audio workstation,
don't run off and play Quake while you're recording a live act
Stick to playing Solitaire.

Paul

 

[Dominique]

I am using a Window 7 machine with an 8 channel audio card to provide
multichannel outputs from a dedicated software application. The
application plays sampled pipe organ audio in response to Midi commands
from musical keyboards. Basically it's a virtual pipe organ. The
problem I'm having is that the processor (quad core AMD) speed changes
due to various things that can happen, such as overheating, etc. This
results is the audio board putting out a click or pop when the processor
speed is adjusted. I have change many parameters and have improved the
situation considerably, however, 1 pop or click is not acceptable. I
know many of the parameters must be adjusted in the BIOS, but I'm sure
there are some things that W7 does that affects this problem. Anyone
have any ideas?
Thanks.

You could try to give the processors priority to background services,
Right-click Computer icon, select properties, click on advanced system
parameters, in the box that opens click the Parameters button in the
performances section, in the next box, click the Advanced tab and the
parameter is there.

You could also set your ASIO buffer size a little bit higher, if you play
"live", the bigger the buffer size, the longer the latency will be. But if
actually you have a latency of,say, 5.0ms, maybe one of 8.5ms would be OK
for your application.

HTH

 

[Dominique]

I am using a Window 7 machine with an 8 channel audio card to provide
multichannel outputs from a dedicated software application. The
application plays sampled pipe organ audio in response to Midi commands
from musical keyboards. Basically it's a virtual pipe organ.

The
problem I'm having is that the processor (quad core AMD) speed changes
due to various things that can happen, such as overheating, etc. This
results is the audio board putting out a click or pop when the processor
speed is adjusted. I have change many parameters and have improved the
situation considerably, however, 1 pop or click is not acceptable. I
know many of the parameters must be adjusted in the BIOS, but I'm sure
there are some things that W7 does that affects this problem. Anyone
have any ideas?
Thanks.

Is your processor overheating???...

I don't know much of the internals of windows or bios but I think that when
windows loads, it kinds of takes over the bios and when you quit (windows),
it gives back control of the computer to the bios (or you), so with the
Default, Performance or AUTO BIOS parameters choices, I set everything to
AUTO in the BIOS (and I disable all that I don't need, Parallel ports,
serial ports, onboard audio(?), etc), while I'm there I set Plug'N Play OS
to YES, if I can.

My Win7 x64 Core2Quad has 2 PCI audio cards (old tech) 10ins-10outs each
and I don't have those kinds of problems. We have recently recorded drum
tracks for some musical pieces and we've recorded 16 tracks simultaneously.

HTH

 

[Art Todesco]

Is your processor overheating???...

I don't know much of the internals of windows or bios but I think that when
windows loads, it kinds of takes over the bios and when you quit (windows),
it gives back control of the computer to the bios (or you), so with the
Default, Performance or AUTO BIOS parameters choices, I set everything to
AUTO in the BIOS (and I disable all that I don't need, Parallel ports,
serial ports, onboard audio(?), etc), while I'm there I set Plug'N Play OS
to YES, if I can.

My Win7 x64 Core2Quad has 2 PCI audio cards (old tech) 10ins-10outs each
and I don't have those kinds of problems. We have recently recorded drum
tracks for some musical pieces and we've recorded 16 tracks simultaneously.

HTH

Lots of good suggestions, and I'll try them later today. Looking at
the tools provided, I don't see any overheating. But, if I watch the
processor clock speed, it is constantly changing. It used to change
quite a bit, but now after changing many BIOS settings, it only changes
a little bit. BTW the audio board is an M-Audio Delta 1010LT PCI card.
Maybe it card was a bad choice as googling the problem on this board,
pulls up all sorts of similar problems. I actually ran this software on
a pretty slow, W7 laptop, using the 2 internal sound channels, and (of
course) never saw (or heard) this problem. BTW, HTH, what type of
boards were you running? Actually, from the description, it sounds like
the same M-Audio board.

 

[Paul]

Lots of good suggestions, and I'll try them later today. Looking at
the tools provided, I don't see any overheating. But, if I watch the
processor clock speed, it is constantly changing. It used to change
quite a bit, but now after changing many BIOS settings, it only changes
a little bit. BTW the audio board is an M-Audio Delta 1010LT PCI card.
Maybe it card was a bad choice as googling the problem on this board,
pulls up all sorts of similar problems. I actually ran this software on
a pretty slow, W7 laptop, using the 2 internal sound channels, and (of
course) never saw (or heard) this problem. BTW, HTH, what type of
boards were you running? Actually, from the description, it sounds like
the same M-Audio board.

You may have more than one way to reach the card. ASIO versus DirectSound.
Maybe the buffer depth is set differently for those or something.

http://en.wikipedia.org/wiki/Audio_Stream_Input/Output

Paul

 

[Dominique]

Lots of good suggestions, and I'll try them later today. Looking at
the tools provided, I don't see any overheating. But, if I watch the
processor clock speed, it is constantly changing. It used to change
quite a bit, but now after changing many BIOS settings, it only changes
a little bit. BTW the audio board is an M-Audio Delta 1010LT PCI card.
Maybe it card was a bad choice as googling the problem on this board,
pulls up all sorts of similar problems. I actually ran this software on
a pretty slow, W7 laptop, using the 2 internal sound channels, and (of
course) never saw (or heard) this problem. BTW, HTH, what type of
boards were you running? Actually, from the description, it sounds like
the same M-Audio board.

I run 2 M-Audio Delta 1010, the model with a separate rackmount
connectors box but I think it's basically the same engine than the
1010LT.

By the way HTH means "hope that helps".

Dominique

 

[Dominique]

Paul <[email protected]> écrivait

You may have more than one way to reach the card. ASIO versus DirectSound.
Maybe the buffer depth is set differently for those or something.

http://en.wikipedia.org/wiki/Audio_Stream_Input/Output

Paul

Usually, ASIO is preferred for musical live work because of the lower
latency, when you press a key on a MIDI keyboard to drive a software
synthesizer you want to ear the sound instantly without "pops and clicks".

The wiki article talks about the ASIO4ALL driver for WDM but the OP has an
M-Audio interface with native ASIO support.

Trying the WDM driver might work but it may have a longer latency. Musical
applications usually allow to choose the type of driver to use.

Dominique

 

[Paul]

Art Todesco <[email protected]> écrivait


If by major change you mean changing computer, I would go Intel PC
(processor AND chipset), not MAC. Be sure that the motherboard support
"true" PCI. I've read that on some modern MB there is "emulated" PCI slot
that add some kind of layer over PCIe which can affects latency; it can be
OK for all kinds of add-on cards but not that good with older PCI audio
cards. Real techs can jump in and correct me if I'm wrong.

In the past, some AMD systems had troubles dealing with pro audio. I don't
know if that has changed, the last AMD system that I've used was a 386
running Windows 3.1 a long time ago.

Another change you could do is change your audio interface but your 1010LT
should be good for the task.

Dominique

Computers contain many busses, so there are already bridges present inside
the chipset.

Bridges are inefficient when bit-banging registers on the target.

To see what impact that can have, consider the low bandwidth achieved
during PIO mode disk transfer. Doing things one-at-a-time is slow to
very slow.

Modern brdge products, usually don't contain a pathological analysis of
what can happen with poorly chosen interfaces. But a local company,
a company making bus bridges, they did offer application notes,
and took the trouble to display bus traces for pathologically bad
conditions. Their bridge did around 1MB/sec under those conditions.

Now, the sound card already has to consider the nature of the PCI bus.
The PCI bus isn't that efficient if you're bit-banging registers.
So the sound buffers would be transferred by DMA, the bursts would
translate from a burst of PCI bus words, into a single PCI Express
packet. That's an efficient mode of operation, preserving most
of the available bus bandwidth.

So I couldn't really make any statement, about what you might expect.
If most of the sound card setup is done out of the critical path,
and most all of the subsequent operations are the filling of the
sound buffer with DMA bursts, a bridged setup might not be that
bad at all. If on the other hand, the thing didn't have scatter gather
DMA, didn't pull buffer lists with DMA as well, then you might notice
some effect. If the card had an efficient design, to work with native
PCI, chances are a motherboard with bridged (fake) PCI bus would
not be that bad.

Paul

 

[Dominique]

Computers contain many busses, so there are already bridges present inside
the chipset.

Bridges are inefficient when bit-banging registers on the target.

To see what impact that can have, consider the low bandwidth achieved
during PIO mode disk transfer. Doing things one-at-a-time is slow to
very slow.

Modern brdge products, usually don't contain a pathological analysis of
what can happen with poorly chosen interfaces. But a local company,
a company making bus bridges, they did offer application notes,
and took the trouble to display bus traces for pathologically bad
conditions. Their bridge did around 1MB/sec under those conditions.

Now, the sound card already has to consider the nature of the PCI bus.
The PCI bus isn't that efficient if you're bit-banging registers.
So the sound buffers would be transferred by DMA, the bursts would
translate from a burst of PCI bus words, into a single PCI Express
packet. That's an efficient mode of operation, preserving most
of the available bus bandwidth.

So I couldn't really make any statement, about what you might expect.
If most of the sound card setup is done out of the critical path,
and most all of the subsequent operations are the filling of the
sound buffer with DMA bursts, a bridged setup might not be that
bad at all. If on the other hand, the thing didn't have scatter gather
DMA, didn't pull buffer lists with DMA as well, then you might notice
some effect. If the card had an efficient design, to work with native
PCI, chances are a motherboard with bridged (fake) PCI bus would
not be that bad.

Paul

Thanks for those infos. I checked the ressources of my Delta 1010s (10
years old) and they use 1 IRQ and 4 address ranges (DMA I guess) each.
I'm pretty sure the internal design of the engine of my audio card is
very similar to the OP 1010LT.

I could try my Delta 1010 in my more recent i3 (Windows8Pro) and see how
it goes; if I have the time.

Dominique

 

[Dominique]

Art Todesco <[email protected]> écrivait

<snip>

But, if I watch the processor clock speed, it is constantly changing. It
used to change

quite a bit, but now after changing many BIOS settings, it only changes
a little bit.

<snip>

Have you tried to disable those "system performances watching" utilities?

I think they can use some processing power when it's not the time to do it.

HTH

 

[Art Todesco]

Thanks for those infos. I checked the ressources of my Delta 1010s (10
years old) and they use 1 IRQ and 4 address ranges (DMA I guess) each.
I'm pretty sure the internal design of the engine of my audio card is
very similar to the OP 1010LT.

I could try my Delta 1010 in my more recent i3 (Windows8Pro) and see how
it goes; if I have the time.

Dominique

That's way too much to ask of you, however, it would be wonderful
information for me. Actually, I was going to try putting the 1010LT
into another older computer that I have here. But it's an AMD also, an
Athlon 64, 3200+. And, I'm not sure if the 1010LT will even fit,
physically. But, I'll check that out too.

 

[Dominique]

Art Todesco <[email protected]> écrivait email.me:


<snip>

But, if I watch the processor clock speed, it is constantly changing. It
used to change

<snip>

Have you tried to disable those "system performances watching" utilities?

I think they can use some processing power when it's not the time to do it.


<snip)
That's way too much to ask of you, however, it would be wonderful
information for me. Actually, I was going to try putting the 1010LT
into another older computer that I have here. But it's an AMD also, an
Athlon 64, 3200+. And, I'm not sure if the 1010LT will even fit,
physically. But, I'll check that out too.

.... the text above from the last <snip> comes from the other part of this
thread

My comment:

I guess you are not only playing pipe organ on your system with the card
you have but for the pipe organ program, have you tried to <really>
increase the buffer size in the Delta control panel (number of samples in
the hardware section, if your control panel is like mine)?

Correct me if I'm wrong but, if I recall, a real pipe organ has latency
by nature with all that "compressed air". So, maybe a little latency when
you're playing might be acceptable while you troubleshoot the problem.

Actually trying one of my Delta 1010 in my i3 is quite a job, they're in
a rack with lots of wires around, but one of them needs capacitors
replacement in the external box. So when I'll replace them, I will test
it in my i3 before putting it back where it belongs (if my repair works).

Also, my i3 runs under Win8 and there's no Win8 drivers on the M-Audio
site but I'm confident the Win7 drivers should work; I hope because if
the test is positive, my i3 could become my "studio" computer with a
processor upgrade to i7.

Dominique

 

[Art Todesco]

... the text above from the last <snip> comes from the other part of this
thread

My comment:

I guess you are not only playing pipe organ on your system with the card
you have but for the pipe organ program, have you tried to <really>
increase the buffer size in the Delta control panel (number of samples in
the hardware section, if your control panel is like mine)?

If you are speaking of number of samples per second, that is pretty much
set by the program. I think there is a way to set the buffer size. I
can try that.

Correct me if I'm wrong but, if I recall, a real pipe organ has latency
by nature with all that "compressed air". So, maybe a little latency when
you're playing might be acceptable while you troubleshoot the problem.

Yes, there are actually 2 types of latency in a real pipe organ. First
you have mechanical latency of opening the valve to let air into the
pipe. Typically a key is depressed making an electrical connection that
opens a small valve, which in turn opens a bigger one, which may in turn
open the air to the pipe. Of course, there are purely mechanical
(Tracker) organs, but I won't talk about that. The second latency comes
from the pipe itself. It take a little time to get the vibrations
going. The bigger the pipe, the longer it may take.

Actually trying one of my Delta 1010 in my i3 is quite a job, they're in
a rack with lots of wires around, but one of them needs capacitors
replacement in the external box. So when I'll replace them, I will test
it in my i3 before putting it back where it belongs (if my repair works).

Also, my i3 runs under Win8 and there's no Win8 drivers on the M-Audio
site but I'm confident the Win7 drivers should work; I hope because if
the test is positive, my i3 could become my "studio" computer with a
processor upgrade to i7.

I've never liked AMD processors. I probably shouldn't have gotten this
one and just stay with the Intel stuff.

 

[Dominique]

Art Todesco <[email protected]> écrivait email.me:

If you are speaking of number of samples per second, that is pretty much
set by the program. I think there is a way to set the buffer size. I
can try that.

The control panel of my delta 1010s looks like this:

http://imageshack.us/a/img10/1708/1esx.jpg

It's the last icon at the right of the taskbar on the picture. You should
have something similar with your card. If it's not on the taskbar, it
must be in the Windows Control panel (create a shortcut if you haven't
already have).

I would try different numbers of samples starting at 512 and going down
and see (hear) how it goes (sounds)...

....and, I would restart the computer between every change in the (Delta)
control panel.

Yes, there are actually 2 types of latency in a real pipe organ. First
you have mechanical latency of opening the valve to let air into the
pipe. Typically a key is depressed making an electrical connection that
opens a small valve, which in turn opens a bigger one, which may in turn
open the air to the pipe. Of course, there are purely mechanical
(Tracker) organs, but I won't talk about that. The second latency comes
from the pipe itself. It take a little time to get the vibrations
going. The bigger the pipe, the longer it may take.

Thanks for this info, those beasts sound Great!

Dominique