Hardware Techies look here!!! (Mystery question)

[Thomas_J]

Okey folks, I\'m scratching my head here.

Gigasampler is limited to 64 voices of
polyphony. (this is the equalient of 64 mono
samples playing simultaneously).

Now, this is a limitation set by your hardware. Mainly the speed of your harddrive, cpu and amount of ram.

I\'m questioning the specs of Ultra ATA/66 drives.
Do they in fact transfer 66mb/sec at a sustained rate, or is that just some misleading spec? Because if they did....

...i\'ve been doing some calculation here:
A one second 44khz 16bit mono wavfile takes up 88,2k (without a RIFF4 header). This means that you in theory can have 766 voices playing simultaneously on an UATA 66 drive (provided your CPU is up to it, and it probably isn\'t http://www.northernsounds.com/ubb/NonCGI/images/icons/smile.gif.
That is 383 stereo voices.

Gigastudio features 160 voices of poly.
For the record, that is 80 stereo samples. All at once.

For this amount of voices you would (technically) only need a harddrive with a sustained transfer rate of exactly 13.7mb/sec.

I can see the reason why people want to get SCSI drives (as SCSI\'s don\'t use any cpu, unlike IDE\'s). But I fail to understand why people get the drives with 160mb/sec and above transfer rates?? I just don\'t get it.
In theory you would be able to get 1857 voices of polyphony out of a harddrive-beast like that. If it only was as simple as keeping a sustained transferrate at 160mb/sec. Of course it isn\'t. All those 1857 voices would have to be mixed (processed), in realtime (to keep the latency below 10ms), and any fool can see that it would require a MONSTER cpu to mix 1857 mono channels in realtime.

So how can it be that scsi drives are called for with gigasampler, when UTA66 drives already are overspecced for use with Gigastudio?

Also, if a p2 233mhz is sufficient for mixing 64 mono samples in gigasampler, then I would expect a p2 450mhz (twice the speed) to be capable of 128 voices, so an overclocked celeron 450 @600mhz should be able to handle Gigastudio\'s 160 voices of poly.

Maybe?

Someone please enlighten me!

[John Matrix]

ATA 66 and SCSI Ultra160 have nothing to do with the speed of the hard drives themselves. The numbers refer to the speed of the bus to which the drives are connected. Thus, the theoretical max transfer rate from the read cache of an ATA 66 drive to the ATA controller is 66 megs/sec. Whereas the equivalent from an Ultra160 drive to the SCSI host is 160 megs/sec.

In terms of sustained data rate, the real bottleneck is the drive itself - usually in the range of 10 to 30 megs/sec. Unless you connect e.g. 10 SCSI drives to the same bus, which -in theory- could mean that the bus would be the bottleneck.

As far as I know however, the seek time is equally (if not more) important than the sustained data rate when using Gigasampler/Studio. Please comment anyone.

Exsqueeze my English

[Bob Campbell]

My guess would be that seek time/access would be the most important thing for Giga, not sustained transfer. For recording say 90 audio tracks in VST/Logic you need massive transfer rates, like the 160/80 of scsi.

But for Giga, you need very quick access, because if you are playing lots of different samples together on different midi channels, the drive is seeking lots of different files and locations within the Gig files, so you need sub 9ms access times for performance. For this reason theres not much difference between ATA66 and ATA100 for Giga apart from saving and loading times! The RPM is much more important.

Bob

[Bill]

<BLOCKQUOTE><font size=\"1\" face=\"Verdana, Arial\">quote:</font><HR>Originally posted by Thomas_J:
Okey folks, I\'m scratching my head here.

Gigasampler is limited to 64 voices of
polyphony. (this is the equalient of 64 mono
samples playing simultaneously).

Now, this is a limitation set by your hardware. <HR></BLOCKQUOTE>

Actually, that limitation is built into GS. There is a voice expansion upgrade.

[IOComposer]

Thomas (and gang),
No single drive is capable of a sustained transfer rate of more that about 20Megs. I own 2 U2W Seagate Cheetahs on my Mac and they transfer about 30-35M per second paired together in a RAID config. The only reason I use these drives in a raid config is because I have to work with uncompressed video from time to time where sustained transfer rate is important. It\'s not as important with audio apps like GS. More important (as previously posted) is seek time. SCSI seek times blow UDMA away. I\'ve got a 7200rpm IBM Utlrastar UW in one of my GS computers and it has about a 6.5 second transfer rate. UDMA/66 drives max out around 8.5 - 9.

I\'ve come to the conclusion that no matter what Nemesys pulls off with GigaStudio, there\'s no way any normal computer with a single UDMA/66 drive will be able to handle 160 voices at a time. If it does, than they\'ve pulled off a modern miracle. I would expect that you\'ll need at least dual U2W/LVD scsi drives for consistant playback with a huge amount of voices. Here\'s a better plan: use 2 computers dedicated for GStudio. That\'s what I do with GS. Works like a charm...I rarely lose polyphony.

Just my $.02.
-Jamey

[Viktor]

IOComposer,

I dont agree with you. Gstudio is capable of getting the 160 poly with udma drives. Nemesys specified the min config for 160 voice GS:
PIII 700+
128 MB+
fast udma/scsi disk

The trick is that gs dont read 160 voices directly from disk. It use RAM to minimize latency and to make a disk\'s life a little bit easier. Seek time IS most important thing in GS poly, and udma66 is capable of burst transfer of 59MB/s. I guess GS reads small chunks of data (less then 100kB) from disk and transfer them to cache found on the disk itself (modern udma disks have 2MB cache), then these 100kB data can move trough the udma66 chanell in burst mode directly to RAM (DMA enabled). After that point there is no more bottlenecks in PC (CPU-chipset-RAM bandwidth is around 200MB/s)

btw 160 voice poly requires around 16MB/s bandwidth. Hard disk itself (with so many random reads) cant handle that throughput, but using the RAM as cache it can.


hope this helps

[abi]

clarification on the bandwidth for 160 voice: a mono 16 bit 44100 Hz sample rate is 75kb/second, therefore 160 * 75 kb is around 12megabytes / sec. Seagate medalist ultra ata66 can do 13 megabytes / sec average sustained throughput, but since giga reads small chunks here and there, medalist wouldn\'t be able to do 13 megs. It\'s best to spread the wave files into several harddrives.