Hello,
I'm considering a Creamware card purchase to run a reverb or two from Timeworks.
Anybody know how many dsp's I need to run one instance of the P-100 at 88.2khz? Or the other A-100?
Thanks,
Bobby
88.2khz and DSP
first you'd need an external clock source and the 'sync plate' hardware extension to run 88.2 khz.
I guess you want to simplify 'upsampling' by choosing an even multiple of 44.1, but it doesn't work this way.
Proper upsampling is a rather demanding process which is independant from the source and destination rates. It's always the same effort - don't ask me for the math background, but there is one
by just doubling (or dividing) the number of samples you'd introduce a constant jitter which will distort the signal.
Simplyfied the axes don't match: time is linear while amplitude follows a sine curve.
that's why it's preferable to work in the destination sample rate and to restrict conversions to the unavoidable.
Trig functions have no fixed results and always introduce (some) error.
Anyway - with those reverbs you'd probably be limited rather by the PCI bus bandwidth, than by calculation power of the DSPs.
At 44k you should be able to run one instance of each simultaneously if you have a proper setup(!) - I never considered 96k for that reason.
yet both are great and extremely affordable, so it would even make sense to dedicate them their own box.
Imho the only reason for higher sample rates is audio processing in DVD postproduction, but that's for convenience (as they have that standard), not for quality sake.
Finally everyone got to make up his own mind about the (neverending) samplerate discussion
welcome btw and cheers, Tom
<font size=-1>[ This Message was edited by: astroman on 2005-08-09 03:07 ]</font>
I guess you want to simplify 'upsampling' by choosing an even multiple of 44.1, but it doesn't work this way.
Proper upsampling is a rather demanding process which is independant from the source and destination rates. It's always the same effort - don't ask me for the math background, but there is one
by just doubling (or dividing) the number of samples you'd introduce a constant jitter which will distort the signal.
Simplyfied the axes don't match: time is linear while amplitude follows a sine curve.
that's why it's preferable to work in the destination sample rate and to restrict conversions to the unavoidable.
Trig functions have no fixed results and always introduce (some) error.
Anyway - with those reverbs you'd probably be limited rather by the PCI bus bandwidth, than by calculation power of the DSPs.
At 44k you should be able to run one instance of each simultaneously if you have a proper setup(!) - I never considered 96k for that reason.
yet both are great and extremely affordable, so it would even make sense to dedicate them their own box.
Imho the only reason for higher sample rates is audio processing in DVD postproduction, but that's for convenience (as they have that standard), not for quality sake.
Finally everyone got to make up his own mind about the (neverending) samplerate discussion
welcome btw and cheers, Tom
<font size=-1>[ This Message was edited by: astroman on 2005-08-09 03:07 ]</font>
I mix to 1/2" tape before converting to 44.1k so my choice of sample rate is not based on any resampling math. I just like higher sample rates and recorded lots at 44k and 48k before making the decision.
My outboard EQ's both have bell shapes that extend above 25k and I always found reaching for the air of my final mix 2-buss to sound a bit strange @ 48k. Kinda like taking the anti-aliasing shelf and shoving it closer to the listener!
At 88.2k I think it's more natural on it's way to tape. It's subtle though, I'll admit, and there's a big DSP price for this. But I have to crank this music at my head everyday!
As for the PCI buss, there are around 320 mono channels at 48k/24bit in real-world bandwidth and DSP takes one each way (Universal Audio techs discussed this with me) so I can't see a project of less than 40 tracks @ 88.2 khz ever touching the PCI buss limit with 3 or 4 PCI slots of DSP.
I've streamed 24 channels of 48khz audio, and used about 30 channels of Powercore and UAD-1 through the PCI buss before without problem. I doubt, if there is a problem running 88 and 96k reverbs on the Creamware it would be due to PCI overload. Two 88k true stereo reverbs is equivalent to only 16 mono channels @ 48k. Only 5% of buss!
I agree about the dedicated box for the verbs, no matter what the platform. I don't like running the DSP for reverbs at 88.2k as I think it's a waste. The air I like at 88.2k is more useful in the program than in the effects, IMO. I probably will take the time to install whatever I get for reverbs in another box so I can take advantage of 44.1k there.
Thanks for dicussing and glad to get to know Creamware a bit.
<font size=-1>[ This Message was edited by: wilcofan on 2005-08-09 09:34 ]</font>
My outboard EQ's both have bell shapes that extend above 25k and I always found reaching for the air of my final mix 2-buss to sound a bit strange @ 48k. Kinda like taking the anti-aliasing shelf and shoving it closer to the listener!
At 88.2k I think it's more natural on it's way to tape. It's subtle though, I'll admit, and there's a big DSP price for this. But I have to crank this music at my head everyday!
As for the PCI buss, there are around 320 mono channels at 48k/24bit in real-world bandwidth and DSP takes one each way (Universal Audio techs discussed this with me) so I can't see a project of less than 40 tracks @ 88.2 khz ever touching the PCI buss limit with 3 or 4 PCI slots of DSP.
I've streamed 24 channels of 48khz audio, and used about 30 channels of Powercore and UAD-1 through the PCI buss before without problem. I doubt, if there is a problem running 88 and 96k reverbs on the Creamware it would be due to PCI overload. Two 88k true stereo reverbs is equivalent to only 16 mono channels @ 48k. Only 5% of buss!
I agree about the dedicated box for the verbs, no matter what the platform. I don't like running the DSP for reverbs at 88.2k as I think it's a waste. The air I like at 88.2k is more useful in the program than in the effects, IMO. I probably will take the time to install whatever I get for reverbs in another box so I can take advantage of 44.1k there.
Thanks for dicussing and glad to get to know Creamware a bit.
<font size=-1>[ This Message was edited by: wilcofan on 2005-08-09 09:34 ]</font>
this is probably correct for (say) write an audio stream to disk, but it's definetely not what the CWA specific PCI bottleneck is about.On 2005-08-09 09:32, wilcofan wrote:
...As for the PCI buss, there are around 320 mono channels at 48k/24bit in real-world bandwidth and DSP takes one each way (Universal Audio techs discussed this with me) ...
You could call this a design flaw, but one has to keep in mind when those cards were designed.
Around 1996/97 Ram was extremely expensive compared to current prices, so the cards come with only the 500mbit high speed dual-ported Ram on each of the DSPs.
Reverb algorithms like those of the P100 and A100 need much more 'scratch memory' and use the main board's Ram for that purpose. The 'reverb' data is passed forward and backward between the card and mainboard and the data 'packets' (for obvious reasons) have to be precisely in time.
That's why it's such a bus hog.
As any other system SFP isn't perfect, but one can be absolutely sure that new hardware by CWA (in whatever shape) will not have this 'problem' again.
The DSPs in the latest ASB boxes already have a four times larger local memory, probably enough for those reverb algorithms.
On UAD the situation is not that critical as they are operated at much higher latency, and due to their graphic heritage the cards have enough local memory anyway.
cheers, Tom
<font size=-1>[ This Message was edited by: astroman on 2005-08-09 10:22 ]</font>
For info : the background DSP " you can't go under " is roughly half a DSP for 44.1 and one for 96 kHz ( a rough 2 X factor )
P-100 in a minimum project configuration is around 2.5 DSP at 44.1 and 5.5 at 96.
A-100 nearly the same ...
I-100 ( A and B ) slightly less like 2.2 at 44.1 and close to 5 at 96.
And I hope for next Scope update a more useful DSP use info for those that are part of the " Volk " and not developers ...
<font size=-1>[ This Message was edited by: decimator on 2005-08-09 10:46 ]</font>
P-100 in a minimum project configuration is around 2.5 DSP at 44.1 and 5.5 at 96.
A-100 nearly the same ...
I-100 ( A and B ) slightly less like 2.2 at 44.1 and close to 5 at 96.
And I hope for next Scope update a more useful DSP use info for those that are part of the " Volk " and not developers ...
<font size=-1>[ This Message was edited by: decimator on 2005-08-09 10:46 ]</font>
I don't think you need the sync plate for that. Doesn't CWA cards slave to 88.2 just fine without it? (though the sync plate is said to improve jitter stats or something like that)On 2005-08-09 03:06, astroman wrote:
first you'd need an external clock source and the 'sync plate' hardware extension to run 88.2 khz.
