Can't play GigaPiano2...not liking Speedstep

[gkwok]

Hi, I've been lurking here for 8 months, waiting for GS3 to come out. Now that I finally got it, I stayed up until 4am last night trying to get things working. I use GS mostly for live piano playing. Here's my setup:

- Dell Inspiron 8600 Pentium M 1.6GHz (equivalent to about a P4 2.0GHz desktop)
- 1GB RAM, 60GB 7200rpm HDD
- M-Audio Firewire 410, buffer size set to 128 samples (2.7ms at 48KHz)
- Windows XP SP2, reinstalled from scratch just for GS3. I have best performance for "Background services" selected, and memory usage for "Programs." Note, I tried setting memory usage to "System cache," but GS3 would give me errors loading some gigs. Setting it back to Programs fixed it. I also have DisablePagingExecutive=1 in the registry.

GS2.54 worked perfectly under this setup. With GS3, I set the max voices to 512, and I was able to get the full version of PMI's Bosendorfer 290 to about 250 voices, then Windows crashed with a IRQ_NOT_LESS_THAN_OR_EQUAL blue screen in ew.sys. I loaded the full version of GigaPiano2, and the CPU sat at ~50% when GS is idle. I played a few notes, and then Windows crashed with the same error.

This being a laptop, the CPU throttles down to 600MHz when not in full use, and Windows manages its speed as necessary. At this point I installed SpeedswitchXP to get it to run at 1.6GHz all the time (which as a result makes it hotter and noisier because of the constant running of the fans). After running SpeedswitchXP, I can play GP2 freely as long as I don't use the sustain pedal. As soon as I press the pedal, I hear pops. The readme says to increase the audio device's buffer if this happens. I increased it to 256 > 384 > 512 > 1024 samples. At 1024 it eliminated the pops but still crashed if I repedalled enough, and with that kind of delay it was not playable any more. All this happens with GigaPiano2 LE too, but not with MegaPiano. I am guessing the root problem lies in GigaPulse's extreme use of the CPU.

So in summary I have two issues:
(1) GS3 crashes Windows when the CPU gets overloaded. I'd rather have voices stolen than having to reboot my system.
(2) It seems that my mobile CPU is getting overloaded too early, because when it's running at <1.6GHz, the crash occurs sooner. Perhaps GS is just "too efficient," for I never see the CPU pegged at 100% or pegged at 1.6GHz, because maybe Windows doesn't see the need to increase the CPU speed for an extended time. I could make the CPU not throttle and keep it at 1.6GHz, but problem 1 still occurs eventually.

Has anyone else seen similar things on a Pentium M? What about Pentium 4 Mobile?

Sorry for the long post, just trying to provide information and hopefully get some help too.

[JonFairhurst]

I've had some crashes with my AMD 2200+. It only happens when I'm doing a lot of GigaPulse and when I try to do something else at the same time. I've only had one crash when just playing music, but that was with 3 GP instances while recording MIDI into Sonar3 on the same machine. I got audio break up long before I got a crash.

As long as I don't push GPulse too hard, things are fine. I now turn off audio processing (or bypass GPulse) when loading instruments and doing other non streaming work.

I think that the problem is that GPulse has small buffers for low latency, and it has minimal fault checking for high efficiency. It's great for the ultimate performance, but it's a bit like working without a net. Too much GPulse on too slow a processor and you're bound to have problems.

The CPU meter shows very high deviation. The average may be around 50%, but even a short 100% load can overflow a buffer. Then you get a crash.

Oh, I also got a crash when I tried to shutdown the Msg32 thread so I could free things up when doing some non-GS3 work. I won't try that again.

I've found that if I don't overdrive GigaPulse (I can do two instances), and if I disable it while doing non-musical tasks, my 2200+ system is very stable. I'd lay off of GPulse, until you can get stronger hardware.

-JF

[Bruce A. Richardson]

Hi guys,

I would suggest working with the balance of buffers and polyphony limits to find the sweet spot for your particular setup.

This is the "flipside" of the no-limit polyphony. Giga WILL run your machine right into the ground if you push your machine into heatsville.

The laptop situation is, unfortunately, going to be what it is. Because you're dealing with an application that can literally take every last cycle of the machine and do something with it, you're dealing with a torture test for a laptop which cannot possibly cool itself as efficiently as a desktop. The solution is the same--experiment with the polyphony upper limit, and buffers, and balance the two to get the best possible performance.

Either way, you're going to get very good polyphony management with the level of polyphony your machine CAN support, so it should still get you a net improvement.

[JonFairhurst]

I'm pretty confident that my crashes were not due to too much poly. It was mostly a CPU being pushed past it's limits / GigaPulse thing. Aside from the time I tried to shut down the Msg32 thread in the Task Manager (hey, it says bad things may happen when you stop a process!), every crash I've had has been related to pushing the CPU past the point of making clean sounds.

It's likely that I would get better stability with a bigger audio buffer. The audio starts to melt down well before I get the crash. As long as I don't overdo GPulse the audio is stable, as is the program.

As it is, I'm pretty happy with the latency, performance and stability of GS3 right now, so I'll leave the audio buffers as they are. I'll upgrade the CPU soon and should be able to run more GPulse. My solution is to take it a bit easy with GPulse for now, and to turn off audio processing whenever I want to mess with stuff.

It was pretty simple to find the limits, and it's not too hard to stay within those limits. The program is more than stable enough for real world use. I wouldn't consider going back to GS2.5 for a second!

-JF

[Bruce A. Richardson]

If you're interested in playing the GPII with lower CPU overhead:

Save the GigaPiano II off to another file name, and convert the wavepool in the new instrument to 16-bit, full-acceleration, with dither. Save the new instrument, and load 'er up.

This will lower its CPU usage, and should let you play it just fine on your machine (which is pretty much the same as mine).

[JonFairhurst]

I started to do that process, but stopped it when I saw how long it was going to take. I had some video to render last night, so I haven't converted the piano yet. That's definitely a good job to do while sleeping!

-JF

[gkwok]

Thanks for the replies. I've been looking into replacement laptops all day, since I realize the core issue is the CPU. Right now I'm looking at the mobile Athlon 64 3700+...if there are any other fellow mobile GS3+GP2 users out there, please post your experiences.