DP4.5 problems when UAD CPU hits 80%

[rush909]

I have on of the early G5 which are not ideal for the UAD, but what seems to be happening is that my host CPU goes into the red when load on the UAD goes over 80%... and goes up slightly with each UAD plugin I insert when the UAD start to approach 80%....

now I realize that this migh be due to the UAD \"stealing\" some CPU cyles at the kernel level, but this behaviour is very severe for what I think the UAD might be doing...

is there any settings I can do so that the UAD does NOT touch my host CPU!!!! I don't care if I can only get 2 plugs out of the UAD, having the UAD take cycles from my host CPU defeats the entire purpose of having a processor card... I was under the impression that there were some limitations to having one of the early G5, but never though It could mean my CPU load would increase as I add UAD plugins...

is this normal? is there a solution?

thanks!

r.

G5 dual 2Gig (first versions)
Dp4.5
UAD version 4.0

 

[UAJames]

Hi Rush,
You can try disabling 8131 mode from the UAD-1 Meter Configuration panel, but it's not guaranteed that you will even be able to get 1 with 8131 mode off. If you reduce the number of UAD-1 plugins running, that should cut down on the extra host CPU overhead for each plugin (it's the same amount for each plugin, regardless of how much DSP load the plugin requires). Using bus compression/mix groups is an effective way of doing this (and also how a lot of the vintage analog units were used in studios).

Also, what kind of audio interface are you using and what's the buffer size you are using?

 

[rush909]

thanks for the reply...

I am using motu 2408 MKI with buffer settings of 512 or 256... I can use less UAD plugs... or buss stuff... no problem... would another UAD card help out or am I just asking for more trouble???

r.

 

[UAJames]

I would suggest a buffer size of at least 512, preferably 1024 and one other thing to try would be to disable any I/O from the 424/2408 that you are not using from the MOTU PCI Console. The 424 card will reserve PCI bandwidth for any *enabled* I/O channels, even if you are not using them in a project.

Since this is an 8131 G5, most likely it's the throughput that's causing the problem; so adding a second card won't necessarily solve anything, it will just get you more available DSP.

 

[beubbo]

...and one other thing to try would be to disable any I/O from the 424/2408 that you are not using from the MOTU PCI Console.

Hi,

I own a dual G4 1.42GHz with 2 gig ram, OSX 10.3.9, DP 4.52 & 4 uad-1 cards inserted in the G4's PCI slots. Is there something similar to do with a Firewire audio card (actually, Motu 896 HD) to gain some CPU cycles ? (have already disabled all ADAT ins/outs in the Cuemix console)

In fact, I have a similar issue than rush909 and can *only* use 80% of UAD's DSPs if and ONLY if there is zero native plug-in inserted.

From that point, I have to remove usually 1 to 2 instances of UAD-1's plug-in each time I am inserting a native one to prevent DP's Performance meter getting bloody red !

At the end, I get slightly less than 2/3 of both CPU & UAD dsp ressources at the same time... And for some big mixs I really miss those 2 x 1/3+ !!!

I remember being able to get almost 99% of ALL available ressources on my *old* mono G4 733 MHz running Mac OS 9.2.2 and 3 UAD-1's cards : so, is this an OSX-only issue ?

BTW, DP 4.61 update is out but Motu didn't released any list of bug fixes on their web site, so is there some improvements done in regard to UAD-1 plug-in's use (stability, peformance, bugs fixed...) ?

Thanks a lot in advance

 

[Trace]

I may be wrong, but from what I remember FW interfaces may use more host CPU than PCI interfaces do. This might be one of the things you're seeing.

I have a dual G5 1.8Ghz, FW Yamaha O1X and 4 UAD-1's in a Magma Chassis. From my own tests, I can get up around 95% of my UAD-1 DSP. I can't get more than 70-80 Instances depending upon the plugins used, but I can get almost all of the available DSP when using the bigger plugs.

I noticed that the host CPU seems to be taxed a little more than it should using the FW interface. I use 8 in/outs. My buffer is at 512, it was actually worse at 1024 for some reason.

I got a smoother performance with my UAD-1 PCI latency set a little lower. I set the MIN-GNT PCI Latency to 48 with AMD 3131 Comp DMA Mode set to OFF. The more instances I use the higher the Host CPU got. The highest the Host CPU got wa when I had 80 instances of cambridge EQ with 5 bands. The G5's were at 58% while the UAD was only at 89%. I got similar results from 1176SE. I had 72 instances and the G5 was at 50% and UAD at 91%.

How many instances are you getting when the problem starts and what plugs are you using?

TRACE

 

[rush909]

70 instances!!!! must be nice to have 4 UADs!!! 8)

I have never counted instances, but it never fails... as soon as UAD hits over 80%... and I have about 30-40% of host cpu used up... my host CPU shoots up to 100% If I add one more UAD plug...

r.

 

[Trace]

70 instances!!!! must be nice to have 4 UADs!!! 8)

I have never counted instances, but it never fails... as soon as UAD hits over 80%... and I have about 30-40% of host cpu used up... my host CPU shoots up to 100% If I add one more UAD plug...

r.

I've found that the Hostt CPU responds diffferently to different UAD-1 plug ins, so they are not all equal in how they efffect the Host CPU resources. I'm not sure what that is all about tho.

I think what is helping me to get more mileage out of my UAD-1's is the fact that I don't use a PCI Audio Interface. I have MOTU 1224, but I don't use it. I've been using my Yamaha FW O1X, so that leaves evey ounce of PCI Bandwidth for my UAD-1's

Here's my most recent Test:

Dual G5 1.8Ghz PCI-X
2.5 Gb of Ram
Yamaha O1X
7 Slot 33Mhz Magma Chassis
4 UAD-1's

OSX 10.3.9 - UAD vs 4.0
I set the MIN-GNT PCI Latency to 48
AMD 8131 Comp DMA mode is off
Running 24 Channels in Cubase SX 3.0.1, ASIO Buffer at 512, 44.1k
I've posted the most plug ins I could get to run stable with no Host CPU overload:


24 - Dreamverb - 20% G5 CPU & 95% UAD DSP (24 is max - so that's 100%)

40 - CS-1- 30% G5 CPU & 87% UAD DSP/92% mem (40 is max - so that's 100%)

16 - Plate 140 - 15% G5 CPU & 85% UAD DSP/43% mem (16 is max - so that's 100%)

8 - Nigel - 8% G5 CPU & 70% UAD DSP//48% mem (8 is max - so that's 100%)

28 - 1176LN - 25% G5 CPU and 95% UAD DSP (32 is max - 4 Less than max)

24 - Pultec Pro - 20% G5 CPU & 92% UAD DSP (28 is max - 4 less than max)

40 - Pultec - 28% G5 CPU & 93% UAD DSP (48 is max - 8 less than max)

40 - Fairchild 670 mono - 28% G5 CPU & 92% UAD DSP (48 is max - 8 less than max)

64 - LA2A - 45% G5 CPU & 95% UAD DSP (88 is max - so that's 24 less than max)

72 - 1176SE - 50% G5 CPU & 91% UAD DSP (128 is max - 56 less than max)

80 - Cambridge EQ's w/5 Bands - 58% G5 CPU & 89% UAD DSP (184 is max - 104 less than max)


From my own personal research I think its safe to say that its well worth it to add more UAD-1's even if you have a G5 with the AMD 8131. Yes it will totally kill your number of instances, but for the plugs that matter the most, you should get some very good use out of the available DSP.

This brings up an old request that many of us had made for a long while, the Chained Plugs or Chainer plug in shell. Either one would help tremendously to cut down on the traffic across the PCi bus and would help us to get more out of our UAD-1's no matter what G5 we have. I know its not a sexy plug or one that would make any money for UA, but it would be a nice gesture to many of us who got a G5 without knowing about the limitation. Even if it was a fixed suite like Nigel or CS-1 i'd take that. Just make a Analog Channel Strip using all the classic compressors and EQ's. They could be individually turned on or off as we wish. It wouldn't be perfect but it would be useful none the less. I use some combination of those plugs all the time.

If someone wanted to put a combo together on each track in a 24 track session it would ease the strain on the whole system and save on dithering as well. I know its probably not so easy to do using the Pultec along with other plugs due to the upsampling, but i'd like to see some movement on this front.

Perhaps they can just give us some sort of hybrid plug that isn't necessarily a 1:1 of all the classic compressors, but a useful higher quality Dynamics and EQ combo plug based on their classic emulations.


TRACE