Mac: Near CPU Capacity with Nothing Playing

[plowman]

These are newbie issues. Recently I upgraded to 6.5 GB, and I was happy to load more EXS samples and VI's to my Logic files.

But as they accumulated, I noted a drop in performance. Eventually I realized that my dual 1.8 CPU's in Logic's CPU Monitor were registering about 75% capacity each with Transport running (Play) but no instruments performing. Needless to say, when I actually hit music, it didn't take much for a Core Audio overload.

I started deleting Audio Instruments, and the CPU load gradually diminished. Finally, I had no Audio Instruments and the CPU was zero (again, with "play" and nothing performing).

This is news to no one but me, I'm sure. But could someone explain why a song with nothing performing requires this kind of CPU? It's quite a catch-22: more RAM enables greater memory loads, but bigger loads require a Logic song to use more CPU even when nothing is sounding.

No need to tell me about Core Audio overloads with smaller processors. I'm a past master at those, given my humble equipment. But right now I'm actually less productive with higher RAM because even tacit portions of the song require more CPU than they did with lower RAM.

[JWL]

You are right-- more RAM means more samples (with an understandable limit), but what remains is the hard drive speed and bus bandwidth. Didn't mean to harp on an old issue, but that's sort of what's happening here. I've added a 1.2TB RAID recently and am now looking to add at least four of the Raptor 150GB 10k drives on a PCI host in the near future.

With your desired numbers of samples and instances loaded, check Virtual Memory in your Activity Viewer and view the numbers of pages in and out. Also take note of how much RAM has been assigned and how much is "available". The available RAM refers to the amount of RAM available to the computer and not to the user. Anything beyond a certain point loaded by the user is automatically sent over to Virtual Memory while the alleged available RAM will remain the same. The more data to be paged in and out of VM, the harder the CPU will work even when idle.

I just recently added an eSATA II RAID, but am now considering building a stack of 150GB Raptors spinning at 10k each.

But the amount of concurrent MIDI data running at once can easily push the CPU load over the edge. This is where running a few instances of standalone mode can help get a bit more out of a single tower with dual processors.

I know that running buffers at 512 is a borderline pain for tracking, but if you must use a lower buffer setting, take note of the Activity Viewer for just how much data VI and Logic can handle and still work well. For me, once VM reaches 10GB (with roughly 1.2GB marked as "available") I know I've reached a certain limit and must be careful when entering lots of fast notes lest my dual 2.5 goes into meltdown. Others may experience different numbers, but that's sort of where I am presently.

[plowman]

Your response is substantive and much appreciated.

"...What remains is the hard drive speed and bus bandwidth. Didn't mean to harp on an old issue, but that's sort of what's happening here."

I did know that sample performance is an aggregate of these things. But I'd thought that both were issues only when Logic was fetching the end of samples whose heads had already been loaded. I didn't foresee that the mere presence of loaded RAM would affect essentially idle CPUs so much.

It's quite a caveat for slower towers. I mean, if you load 4GB of samples and that immediately translates to 70% CPU gone, that's not a real world improvement. You've pretty much traded the chance to have online samples for the inevitability of freezing and bouncing.

"With your desired numbers of samples and instances loaded, check Virtual Memory in your Activity Viewer and view the numbers of pages in and out."

In a RAM reduced song ("Learn" activated for VI's, and 4.23 GB still reported as "Free"), I've got a VM of 7.79 and with page ins/outs at 21505/0. Tacit portions of the song still require about 50% of both CPU's (that's a guess based on meter observation). My buffer is at 512.

"Anything beyond a certain point loaded by the user is automatically sent over to Virtual Memory while the alleged available RAM will remain the same." I had noticed this when topping out memory. But it seems strange for samples to be sent to VM with over 4GB available and neither Logic nor VI anywhere close to the 3 GB limit per program.

"I just recently added an eSATA II RAID, but am now considering building a stack of 150GB Raptors spinning at 10k each."

Yeah, it's beginning to make sense. Tell me if you can, with a reasonably full sample load in VI and EXS, and playing a song with no music, how high do your CPU's run with a dual 2.5? In other words, what's your "out of the gate" load on the CPU before you even write the first note?

On another board, I was told, "CPU needs to 'prepare' what's going to be played via Audio Inst... in layman's terms."

[JWL]

@Another User said:

On another board, I was told, "CPU needs to 'prepare' what's going to be played via Audio Inst... in layman's terms."

That sounds about right. All of the reference data for larger files are preloaded. According to either Paul or CM, the '.dat' files of the Cube decompress as they stream. If that's true, it's an interesting concept and one I'd never guess would/could work so well on the fly. I still wonder, honestly, because of how soon the CPU goes into the red with this stuff.

Hey-- something just hit me:

You were talking about EXS here in the Vienna Instruments forum. Are you still using Pro Edition?

[JWL]

Another question:

How much stuff do you have plugged into your Firewire ports? You could be experiencing FW log jam if you have HDs and an FW interface going at once. Also, if you have an HD in your FW 800 port and anything in the 400 port, the bus will run no faster than 400.

I can't recall off hand, but I'm pretty sure that all the FW ports are on a single bus (they are on the dual 2.5). That could prove to be a big strain on the CPU as well.

[plowman]

"I'd do a test for you now, except I just dumped Pro Edition to open up HD space only last month."

Thanks for the offer, but it isn't worth your time either. Clearly I'm facing hardware issues that will only resolve when I throw more money at them. Now I see the bittersweet retrospect of how far I managed to travel with so little RAM. In the past, tutti meant overload; anything less could be fudged, wheedled and coaxed.

"But, I recall getting about 35 instances of EXS loaded before things started to cave. CPU hovered over 80%, and once midi note data entered the picture, there was no predicting which track would drop out first."

We know how specious instrument load totals can be, since RAM varies per instrument. But your numbers -- with a dual 2.5 -- match mine with dual 1.8 where I can get about 15 to 18.

"Bascially, it boiled down to one single instrument per session before bouncing to audio became a necessity." Untenable, really.

",,,A user epiphany regarding which software more readily revealed itself more as a smoother personal creative extension of the workflow process." I'm choosing ignorance of the alternatives. I'm sure if I ventured into DP, there'd be that "What? It can be this easy?" revelation. I console myself wth Logic's power, which I imagine to be peerless.

"You were talking about EXS here in the Vienna Instruments forum. Are you still using Pro Edition?" I mentioned EXS for complete disclosure. But the only Pro Edition patches I run are basic, long notes. I never run any Performance Tool patches. And I can low-ram the simple stuff (cherry-pick the samples and thin out the total size). Also, Project SAM requires EXS. I'm at about 60% VI, 40% EXS. In itself, this is needed because of the 3 GB limit for each. Perhaps I should shift totally to VI wherever I can and see how that helps CPU.

"I can't recall off hand, but I'm pretty sure that all the FW ports are on a single bus (they are on the dual 2.5). That could prove to be a big strain on the CPU as well."

Hmm. This is what I have indeed -- a LaCie 250 hooked up to my FW800 (nothing on the FW400 port). My VI's and EXS samples are divided between them and an internal Htachi (not the system drive).

EXS and VI's lead me to set aside my Gigastudio. I think Giga may be making a comeback in the short term.

[JWL]

I wouldn't worry about DP, especially since you are at home with Logic and must use EXS for Project SAM (nice collection, that is). Logic's audio engine is great, and I have quiet suspicions that because its now so closely connected to OSX that it is about as efficient as one can get on a Mac without going to a PT HD system.

You are in a great position with not having to use any of the Perf patches. That spares RAM and CPU right there.

This is such an odd era in computer technology. We've got more horsepower than ever, and the late model PPCs are the slowest to fade of any discontinued Macs I've seen. Yet, these VIs ask so much of the fastest computers-- with libraries the size of VSL and the Cube, it's so easy to hit a brick wall.

Someone is now trying to talk me into an X-Serve setup with CPUs and a RAID, but for me presently it is beyond cost prohibitive. I'm waiting to see if rumors of this fabled 8 Core Intel-Mac will pan out next month. If nothing else, current MacPros and PPCs will drop in price as new models are released.

I love VSL and the Cube especially, but I'm *still* reeling from sticker shock for both software and hardware needed to really make the most of them.

Every time I 'throw money' at my system, it's hardly enough to do much more than to get carried off on a gentle breeze before it fulfills its intended purpose. I could run out today (before holiday shopping ends) to buy all that I need for mega farm-- but somehow I believe as computers get faster, software will get more efficient. I'm still waiting to hear more from software developers about the benefits of Leopard-- with everything going UB now, there will be yet another transition and teething period when Apple and its third-party software developers will address the entire 64-bit pipe dream.

Does one drop $15k+ on a 32-bit farm now at the risk of having everything change over sooner than later? RAM, KVM switches, monitors, hard drives, OS network licensing-- These things don't pay for themselves as quickly as they once did, especially where farms are concerned.

How nice it would be to have the CPU equivalent of a sleigh and eight reindeer to carry the load.

[plowman]

Good (and Silent) night, are we on parallel tracks. You have detailed verbatim the reasons why I am dawdling with this dual 1.8 computer. I've come within one tap of the"Enter" button for a new Mac Mini. Yet I choose to wait.

I think the bouncing baby Clovertown is due, but who knows how long it will "teeth"? Conceivably we could be well into '08 before that power is harnessed, and in the meantime, we're paying top dollar for the exorbitant RAM of MacIntels.

But, as you have said, some fog will clear by Macworld. I do encourage you to wait until then. It's not a question of preventing an ill-timed purchase. All computer purchases are ill-timed. But at least Macworld lets us *know* what we'll be missing in the short and mid-term. And we're all a little smarter after NAMM.

Ecclesiastes says, "As riches increase, so do the mouths that consume them." I've always thought of CPU when I read that verse.

Happy Holidays to you.

[JWL]

Thanks for citing that verse. It's truly universal binary wisdom!

Oh, I'm going to continue to be patient. I nearly jumped into a Mac farm this time last year and then Intels appeared. Glad I waited because whatever I bought last year I'd be stuck with for a while longer-- and it would be impossible to walk into an Apple Store for a while without crying.

The patience required does not necessarily mean that hardware upgrades need to be made slowly. It is probably better to upgrade incrementally, but to some extent it is necessary to upgrade at a pace that balances one's budget with the rate of change in tech world. Otherwise, its forever a game of catch up in which the consumer never wins. Tomorrow's software updates will eventually challenge today's hardware, and for all the waiting we're doing such changes still happen faster than many can keep pace with. Before I really got anything appreciable done in PE it was time for the Cube, for example.

But, I'm grateful to be making music again. With that as my #1 priority, it can only get easier as hardware is augmented to accommodate today's needs while leaving some space for a few needs of tomorrow.

Peace for a safe and happy Holiday Season!