New dual-core system: performance optimal or not?

Hi (and sorry for the long post),

Here it goes: I made the switch during the Holidays and goes from an Athlon XP 1600+ to an Athlon X2 3800 based-system. To test my new rig, I made a simple project in Sonar 5.2 with VSampler 3.5 loaded with a Vintaudio C7 piano (Halion format, 1.6 Gb) and a Sonitus Reverb as insert on the track. Then I played/recorded about 45 seconds of full chords with a lot of notes and a bunch of pedal and... to my surprise, I wasn't able to play/record the track at 1.5 ms of latency without one or two odd clicks (no dropouts in WDM though!). Also, the CPU use is only at 40-45% so it seems strange. Playing or recording at 2.9 ms (WDM or ASIO) is glitch free...

So, I ask: was I expecting too much of my new system? Is this a normal performance for a system of this kind? If this isn't normal, is there some solutions I didn't think about? Could replacing the SATA built-in controller by an add-on PCI SATA card cure the (presumed) problem?

Data that could be useful:

System:

- Gigabyte K8U-939 with latest BIOS revision
- 2 Gb of Kingston ValueRam DDR400
- Windows XP SP2
- Maxtor 40 Gb IDE as OS drive on IDE0 primary
- Pioneer DVR-109 on IDE0 secondary
- Seagate 160 Gb IDE as Audio drive on IDE1 primary (not involved in this test)
- LG CD-RW drive on IDE1 secondary
- Seagate 250 Gb SATA drive on SATA primary. This is the HDD where the Vintaudio samples are stored and streamed from...
- Echo Audio Mia (not midi) with 6.08 drivers (has its own hardware IRQ as per the BIOS screen. APIC IRQ in Windows isn't shared neither)
- Midisport USB 2x2 with latest drivers
- Matrox G450 with 5.86 WHQL drivers. Bus master off (on or off doesn't seems to make a difference anyway...)


About VSampler: the samples are loaded in a mode called DRC (a kind of disk-streaming thing I think...) and take 900 mb in RAM. When it plays the first time, the HDD led flickers like crazy and it is when I hear the clicks. Second time it plays, no led activity and no clicks at the same latency (1.5ms). Upon verification with the Windows performance tool, the samples are read from the SATA drive and paged to RAM the first time. Second time, no indication of paging... Please note that the page file (on the OS drive: 1 Gb) remains at zero so it is not the cause of the problem.

Things that were done so far:

- AMD Cpu latest drivers installed
- Uli latest integrated drivers installed
- Uli latest SATA controller drivers installed (I tried without them and there's no real difference. The drive is then seen by WinXP as an UDMA133 normal drive)
- MS dual-core hotfix with the correct registry settings (strangely, it makes no difference in CPU consumption!)
- UAA (888111) MS hotfix
- I tried with the latest Echo drivers (6.14) and the performance is the same
- I tried the ASIO drivers and the first click occuring cause a dropout
- Minimum of services running
- DAW dedicated partition (no Internet, NIC or Office software)
- Hardware not used disabled in control panel
- I tried to stream the samples from the Seagate 160 Gb IDE and there are more clicks at the same latency...
- The three HDD's were tested several times and they perform the way they're supposed. Obviously, the SATA drive beats the two other ones in all aspects...
- I tweaked the PCI latencies with PCI latency tool and found that if I push the Mia to the limit (255), the SATA controller at 128 and the Ali busmaster controller at 200, there's less clicks (1 or 2 during the project plays instead of 5-6). I tried a bunch of different values here and these are the best settings I've found
- I tried different flavors of Uli, Echo and Matrox drivers without a noticeable gain
- I tried a demo of Kontakt 2 and the performance was worst than with VSampler (there was a nki patch made for the piano and it also make more clicks when I load it in VSampler instead of the Halion patch)
- I made a vs3 patch (VSampler format) of the same piano: no difference
- I tried ASIO4ALL 2.7: no change excepts that it does not dropout like the Echo ASIO driver.
- VSampler 3.51 or 3.52 instead of 3.5: no change
- I tried to enable/disable the hard drive cache on OS drive without change. It is not possible to put it off on the SATA drive...
- Memory timing set to 1t :performance boost in SANDRA but no change regarding the clicks
- All drives thoroughly defragged.

Any advices, hints or confirmation that my new computer works as it should and that I'm paranoid will be appreciated...

Thanks for your reading (and your patience).

i have a very similar system and this is normal. the occasional clicks or dropouts could be due to a number of things , including the performance of your synths, and your soundcard driver. i dont think there is anything wrong with your system. 1.5 ms is quite low for demanding soft synths, particularly disk streaming synths. try setting your processor scheduling to background services. i have also found that performance is slightly better if i close the soft synth display.

ORIGINAL: Patrice Brousseau

Thanks a lot Duojet,

I'll follow your advice and set it to "background". Also, I noticed that I do not have a Venice core (2.4ghz) but a "Manchester" CPU (2.0ghz). Is yours a Manchester also?

In my quest of perfection, I've found that ASIO4ALL drivers (yes, I know they're only WDM drivers wrapped) perform better because I can set "Hardware buffers" and run at 1.5 ms. Not sure however at which point the hardware buffers settings (15ms) adds to the ASIO latency... Hard to tell with latencies as low.

I'll wait impatiently too for Echo Audio and Midiman (USB 2x2) to publish their 64 bit drivers as I want to try Sonar under XP64...

venice i believe is single core only. i have the dual core manchester 2GHZ 512k cache per core. i'm not so sure x64 will help you with latency.

BTW, I've just discovered that CNQ could cause troubles too. If Power Options is set to "always on", am I correct to assume that CNQ will never kick in?

Thanks.

ORIGINAL: Duojet

i'm not sure, but i think you can disable CNQ in the bios

No options for this in BIOS. The only way, as I read on different places, seems to set CPU multiplier to 10x instead of default. For now, setting "Always on" does the trick in my case also as I discovered (CPU-Z reports no variation in frequency).

Thanks.

ORIGINAL: Jim Roseberry

Hi Patrice,

Nothing wrong with your hardware...
If you were streaming those same samples from Kontakt2 (or GigaStudio), I'd bet you wouldn't hear any glitches whatsoever.
FWIW, Vsampler isn't nearly as efficient as K2 or GS3 when it comes to disk-streaming.

I tried to stream the same sampleset (Vintaudio C7 Player 6 velocities in .nki format ) with a demo of Kontakt 2.2 and it was worst... There were a lot more clicks and pops and this, even after I tried to tweak the DFD Kontakt engine.

Maybe I've done something wrong or the .nki patch made for the Piano is bad. As I said in the beginning of this topic, the Kontakt patch also produce a lot more clicks and pops when loaded in VSampler. Go figure...