What's new

NVMe vs SATA: Will it make Kontakt faster?

While we are at the disk debate, how crucial is it really to keep the OS drive separated from samples if you use that particular machine exclusively as a sample slave?

Does it make that much of a difference?
 
Yeah there's no background loading in UVI engine (and it doesn't even support multicore for voice processing like Kontakt does), and Bohemian has quite a lot of samples to load...

While we are at the disk debate, how crucial is it really to keep the OS drive separated from samples if you use that particular machine exclusively as a sample slave?

Does it make that much of a difference?

IMHO it's a "piece of mind" kind of thing. If something happens to the OS, you only reinstall it on that drive, you don't have to touch the sample drives. Plus, OS will constantly want to read or write to its drive, so why should that interfere with DFD performance (no matter how fast our SSDs are)?
 
While we are at the disk debate, how crucial is it really to keep the OS drive separated from samples if you use that particular machine exclusively as a sample slave?

Does it make that much of a difference?

If the machine is just a sample slave with lots of RAM, I'd assume there isn't much that the OS would even need to read/write onto the system disk all the time. I doubt you'd feel the difference between one shared and two separate SSDs here. EvilDragon has a point about system reinstalls. But how often do you really do that? My rig has a roughly 8 year old Windows installation right now.


RAID 0 is just another way to spread your libraries across multiple drives. :)

It should perform pretty similar to placing them directly on separate drives. Personally I would avoid RAID-0 just so as not to have the headache of losing the entire volume when a drive fails (or having to deal with specialized recovery tools).

(I still centralize all my libraries in one place by way of junction points.)

I use junctions as well, very handy for organizing stuff accross a growing number of drives, but easy to lose track over the years.
I would expect raid 0 to be slower than two separate drives in certain high usage scenarios, at least on HDDs, where you're capped on seek-time instead of bandwidth, because imho raid 0 should roughly double the number of reads needed for the same amount of data. I would stay away from raid 0 for the reasons you mention.
 
  • Like
Reactions: TNM
If the machine is just a sample slave with lots of RAM, I'd assume there isn't much that the OS would even need to read/write onto the system disk all the time.

Oh but it does happen all the time, regardless of what the machine is supposed to do. Check it out with Process Monitor. :P
 
I would expect raid 0 to be slower than two separate drives in certain high usage scenarios, at least on HDDs, where you're capped on seek-time instead of bandwidth, because imho raid 0 should roughly double the number of reads needed for the same amount of data.
That depends on your stripe size, but even with a relatively small stripe size those read operations are distributed over multiple drives in parallel, which after all is kinda the point of raid 0's performance benefit :)

I think it might be slower for different reasons. For example if you had lopsided access times between the drives in the array, where you would end up with a lowest common denominator effect. Another scenario where raid 0 with spinning rust would suck pretty bad is two parallel sets of sequential reads (e.g. preloading two different libraries in parallel): with the libraries on separate drives the heads would advance in one direction while with raid 0 they'd be all over the place.
 
Last edited:
There is a huge sale on samsung T3 like I never seen at an aussie store.. he has already sold over 600..

the 2TB for 825 AU, minus 20%, so minus 165$.. therefore $660.. I can't even see them that cheap in the US.. that's about 450 USD for the 2TB.
I think i will get 2, simply connect one each to usb 3 ports on my caldigit thunderbolt 3 hub, and raid 0 them...
T3 sustains 450 read even over USB 3.. that should definitely take it to 700.. Believe it or not no one else has done this on video, anywhere. all people have done is raid some flash drives for fun, like a bunch of 8GB ones.

Thanks for this heads up TNM! I just ordered one T3 for 660 AUD to ship here to Adelaide, coming in a day or two. I'll let you know how it works. Did you get them?
 
That depends on your stripe size, but even with a relatively small stripe size those read operations are distributed over multiple drives in parallel, which after all is kinda the point of raid 0's performance benefit :)

I think it might be slower for different reasons. For example if you had lopsided access times between the drives in the array, where you would end up with a lowest common denominator effect. Another scenario where raid 0 with spinning rust would suck pretty bad is two parallel sets of sequential reads (e.g. preloading two different libraries in parallel): with the libraries on separate drives the heads would advance in one direction while with raid 0 they'd be all over the place.
Well, I was talking about 4 1G 860 EVOs over thunderbolt 3. Still a little confused, should I raid 0, or leave as separate drives? Any guidance would be appreciated as I am setting up a new storage/streaming solution.
 
Maybe if a key use-case was doing heavy streaming of a single library I could see an argument for RAID 0, but otherwise it's not worth the hassle of losing the array when a drive fails. And even in the single library case there's nothing preventing you from manually spreading it across multiple drives with the junction trick.
 
Thanks for this heads up TNM! I just ordered one T3 for 660 AUD to ship here to Adelaide, coming in a day or two. I'll let you know how it works. Did you get them?
I can't believe I didn't..

When I ordered my imac pro and she gave me the 500 discount, i then added the g tech 1TB SSD-R.. i have 18 months interest free so i thought it was worth the extra over the T3 cause it's about 100 mb/s faster.
However in hindsight, i should have gotten at least the T3 1TB.. One can never have enough storage. Too late now!

I do have a brand new crucial MX500 1TB Sata so I'll just get a nice usb 3.1 fanless enclosure for it and use that instead, which i was originally planning to put in a windows laptop, but i decided against getting one for now (for gaming).
 
I can't believe I didn't..

Well, if they go back down in price, I recommend it. I got the 2TB Samsung T3, and ran it through a test, and it reads and writes at 450, which is the same as my internal SSD. What surprised me the most (happily) is how small it is. It's length is as small as the width of normal externals, and it is much thinner too. I'd' say overall you could fit four of them in the same space as a one normal compact external drive.
 
  • Like
Reactions: TNM
Hi, I would like to know regarding your nvme test. You write that on purge there is NO difference from ssd in project loading times, But what about latency (For real time midi playing)? You know nvme latency is lower so in purge instruments wouldn't you get advantage with real time midi playing? I am planning on building a template where all the samples/tracks are purge and all the tracks ready to play on the fly. (Very Heavy libraries)
 
But what about latency (For real time midi playing)? You know nvme latency is lower so in purge instruments wouldn't you get advantage with real time midi playing?
Yes, streaming is going to be better with NVMe. However I didn't specifically test and measure DFD, so I can't quantify by how much (and therefore whether the cost differential is warranted).
 
Well my friend your test document will never be complete without the DFD measurements ;), if you can do a quick observation on this matter it would be super helpful.
 
Well my friend your test document will never be complete without the DFD measurements ;)
It was complete enough for me. :) I had a very specific question: would NVMe help with the part of loading projects I found most annoying.

if you can do a quick observation on this matter it would be super helpful.
That's kind of the rub. To do it properly is definitely not a quick observation. Or at least, the quick observation is this: NVMe will help DFD streaming. I just can't tell you by how much. :)
 
My opinion about NVMe is: If I am building a new system and will have to buy either SSD or NVMe, then I will absolutely get NVMe, even if its maybe up to 30% more expensive. I don't expect to see much difference in load times or even streaming frankly...there are just too many people on the internet saying that the difference is performance is that not great in real life practical terms due to other bottlenecks. But still, might as well get the faster stuff in case other things down the road change better.

But like for my existing system that has $600 worth of SSD's in it now, should I consider changing over to NVMe? Absolutely not worth the cost. IMHO.
 
there are just too many people on the internet saying that the difference is performance is that not great in real life practical terms due to other bottlenecks.
The outcome from my testing was this: while initial load times (where Kontakt blocks the UI) were equal, unpurged patches preloaded all samples into memory 2x faster with NVMe with the documented configuration. Objectively NVMe will handle more voices for DFD before dropouts. With the caveat that I haven't measured it, my WAG on that is about 2-3x more voices in the tested configuration, but possibly up to 4-5x times more, since DFD is sensitive to latency and the drives I benchmarked had about a 4.5x disparity in latency at the relevant block size for Kontakt.

Whether or not your projects need that extra headroom in voices is another question entirely. And probably the most relevant question too.
 
And those are the reasons if I was building a brand new system with no spare SSD's laying around, I would definitely use NVMe. No way I can justify changing from my SSD's to that.
 
Hello Guys
I recently bought a sabrent rocket pro 2 TB with USB-C connection to my 2019 iMac, iMac is i9 with 64 GB RAM. I got the sabrent to have a faster load times for my VI libraries on Kontakt and Omnisphere and other VIs. However, i am noticing that the load times i nsome cases are super fast, and other cases are slower by far than the spinning disks (8 TB G-tech on thudnerbolt 2). I think with big VSTs the nvme drive is loading much slower, sometimes it reaches to 8-10 minutes, same is happening when i want to close the project, sometimes it takes 10 minutes until the project closes. i did not have this issue before with the spinning disks. i understood from sabrent, there is the ssd cache, that might be a cause, but i want to ask do u guys face a similar issue? and what solutions u suggest, u think if i get a thunderbolt 3 enclosure and put the sabrent drive inside the th3 enclosure instead of the usb-c enclosure (that comes as original with sabrent) will solve the issue?
 
Top Bottom