USB 3 external HD compared to internal HD for samples?

[Phryq]

Hey Scott: I love my ASK laptop

From what I've been reading, PCIe SSDs are the best for sample, but I don't know if you can put a PCIe Hard drive in a regular machine (that had been running Sata, for example). You would need a special motherboard for PCIe hard drives, right?

[BertKoor]

I think you need to read all that again...

Afaik harddisks nowadays are all SATA. SSD is just another type of very fast harddisk, and hence uses SATA. Sure, internal disks are always faster than external ones. But maybe USB3 is fast enough for you, it depends...

Who dropped the word PCIe, and in what context? That's got nothing to do with it.

[Phryq]

AUTO-ADMIN: Non-MP3, WAV, OGG, SoundCloud, YouTube, Vimeo, Twitter and Facebook links in this post have been protected automatically. Once the member reaches 5 posts the links will function as normal.

I've been reading about the new PCIe SSDs.

Like here, http://ocz.com/consumer/pci-express-ssd (http://ocz.com/consumer/pci-express-ssd), it shows you can get Random reads at 252,000 IOPS. The highest I've seen was over 9 mill.

Sata III caps out around 100,000. So I'm wondering if it's possible to run PCI hard drives in a laptop, or maybe through USB3 (and if the latter case would be faster than an internal SATA.

But I think you'd need a laptop custom build to handle PCI hard drives, yes?

[mkdr]

Lots of talk.. No real facts.

The fact is that on spinning disks it depends more on the HD itself than the port it uses. USB3 is fast enough for spinning disks to transfer all they can with maximum speed. How can i claim such? Cos i have both and in my case the internal Sata one isn't faster than the external USB3 one. Which of course suggests that it's not that much about the port, but the HD.

Proof:

internal Sata


external USB3


(But yes if you wan't to nitpick about it, the disk inside the USB3 enclosure could be marginally faster if put inside the computer and on a Sata port. But the fact is that you CAN upgrade your current machine with USB3..)


Btw. SSD on a regular Sata is still the best option for streaming sample content (Kontakt instruments for example). It does cost a lot more but the speed benefit is so great it pays for itself (if you compare it to running your samples from RAM.. which is filthy expensive). For OS it isn't that great. It makes the system only marginally faster(tried that). PCI-E SSD is just waste of money unless you need to stream the crap out of it. That means thousands and thousands of Kontakt voices.

A quick math example: 1 second in 24/48 or 3 bytes/48 (usual sampling depth of Kontakt instruments) is: 3 bytes/sample * 48000 samples/second = 144000 bytes /second = 140.625 kbytes / second = 0.137 mb/s
A regular SSD moves 500mb/s => 500 / 0.137 = 3649.64

Which all means you can play sustainably around 3600 mono or 1800 stereo voices of "different" sample material. Why i say different is that my machine can run around 6000 mono voices when the sample material is closely related and Kontakt can share the stuff it's read to memory and my SSD can read around 500mb/s. So a PCI-E SSD isn't looking that good of an investment... unless you need to stream a SHHIT load of samples. Which you don't.


For a laptop an external USB3 SSD might make sense. For desktops it's safer bet to go for internal Sata if possible. Even on USB3 it will be a lot faster than spinning disks though.

[Phryq]

But if you're running the new Berlin Strings (4gb per instrument / articulation) and you're limited to 32 gb ram, then having an SSD that can direct stream entirely would be great. My first SSD doesn't seem to be able to do that (though having a second may help; I have about 900gb worth of sample content on the first drive, and usually use about a third of it at once).

Anyhow, I think it is impossible to use PCIe in a laptop; you would need a special motherboard that supports PCIe drives, right? So using a PCIe would mean buying a new laptop (which is out of the question in my case).

[mkdr]

I have about 900gb worth of sample content on the first drive, and usually use about a third of it at once).

Realisticly you are not running even close to that "at once". As a third would be 300gb.. which would yield 300gb / 0.137 mb/s => 2242335 mono voices. You only have that 300gb selected for use and the starting points are buffered into memory, but no you are not "using" them at once. Your current SSD can surely stream them efficiently enough. Just try it.

As a side note, 32gb of ram is just silly. Load the instruments from SSD and minimize the buffers.

[egbert]

@mkdr
The convention is megabytes per second is MBps and megabits per second is Mbps. You confuse people if you use a lower case b for bytes.

As far as calculations about streaming from multiple files simultaneously, actual measurements carry more weight than back of the envelope calculations as some time is likely to be lost seeking for many separated files to read from - obviously not as bad as rotating media but still some loss compared to continuous read from a single file.

[mkdr]

@mkdr
The convention is megabytes per second is MBps and megabits per second is Mbps. You confuse people if you use a lower case b for bytes.

Seriously?

Well actually the official convention is Mebibyte or MiB by the international standard for quantities and units in information science. Or IEC 80000-13:2008 like most of us engineers know it.

Just wanted to keep things simple as musicians don't know what bits are(32bit vs 64bit.. you know what i mean). So everything around here is MegaBytes no matter how it's written.

As far as calculations about streaming from multiple files simultaneously, actual measurements carry more weight than back of the envelope calculations as some time is likely to be lost seeking for many separated files to read from - obviously not as bad as rotating media but still some loss compared to continuous read from a single file.

Have you checked the SSD seek times lately? 0.04 ms for the Samsung SSD i'm using for example. Totally random read speeds sure are slower because of it, but in the case of Kontakt instruments the reads are not totally random. Which is of course evidenced already by the fact that spinning drives can be used for streaming them too.

If you really need benchmarks you can check out this for a start:
http://www.tomshardware.com/charts/ssd- ... ,2786.html
or
http://www.tomshardware.com/charts/ssd- ... ,2788.html

[egbert]

Hard drive read speeds are often stated on the manufacturer's website in bits per second. It gets confusing if the same abbreviations/symbols are used for both bits and bytes.

The most useful benchmarks would be streaming voices in Kontakt on different drives.

Musicians don't know digital terminology on KVR? Online music sites are full of IT literate types in my experience - plenty of them here.

[mkdr]

Hard drive read speeds are often stated on the manufacturer's website in bits per second. It gets confusing if the same abbreviations/symbols are used for both bits and bytes.

What manufacturer does that? I've never seen any. Usually there's no indication of read speeds(spinning drives) or there's MegaByte per Second readings. They do state the interface speed in Bits though. Mostly because internal and external devices need to be comparable and USB is defined like all serial ports in Bit speeds.

The most useful benchmarks would be streaming voices in Kontakt on different drives.

Musicians don't know digital terminology on KVR? Online music sites are full of IT literate types in my experience - plenty of them here.

Yes maybe that would be better. Benchmarks on Kontakt voices. That test needs to be very well thought out though. There's too many factors that affect the result. And usually the bottleneck is going to be the CPU as many Kontakt instruments have wild scripts, modulations and effects on. And with too small samplebase there's going to be caching involved.

I've already tested some bit with a single instrument but that only shows the CPU and RAM efficiency. Windows is caching the instrument in the background.
I might try to whip something up that would give a truthful result for the SSD speeds.