190,626 users have contributed to 42,756 threads and 257,207 posts.
In the past 24 hours, we have 5 new thread(s), 13 new post(s) and 45 new user(s).
The performance of the T7 isn't good for small file access.
Accessing small non-sequential chunks of data is similar to accessing millions of small files. The smaller the chunks, the more non-sequential reads must be carried out per gigabyte, which slows things down.
Of course none of this explains why the T7 is slower than the T5 in this regard.
I hope it's a fixable bug rather than a design limitation.
I hooked my 2013 Samsung SSD via USB 3 to my 2012 Samsung i5 dual-core laptop. According to the Windows version of Synchron Pianos I'm getting 600 MB/s.
The same drive shows around 150 MB/s in the Mac version with my 2018 MBP 15.
The performance is better on the Mac though with the T7 with a 64 sample buffer. I can still manage 2 mic pairs with a 128 buffer on my old laptop with the older SSD. 😀
For my next trick, I'll try to get everything up and running on a 1970s calculator! 🤡
Is there a consensus on whether the Samsung T7 would be a better choice for VSL than the T5? From what I've read, apparently the older T5 is faster at random reads. Am I right to assume that VSL libraries are stored as tons of small files, so when a preset is loaded random reads determine the loading speed? If so, this would imply that not only is the T7 not worth the price difference, but the older T5 would actually perform faster for this application. (I'm using only Synchron Pianos.) Thanks in advance for any insights.
Johnstaf, your Windows computer + T5(?) Synchron speed result is most interesting - many thanks for that.
Now we just need someone who streams VSL libraries from a T7 to a Windows computer (with the correct type of USB port for the T7) to tell us what the PC-version Synchron speed meter says about the T7.
Alkan, I really wouldn't get too attached yet to the idea that the T5 is better than the T7 for sample libraries. If it turns out to be true - and there seems to be only one piece of highly dubious evidence so far - then Samsung would lose face and probably be hit by several class actions from thousands of miffed customers claiming they've been scammed. But I seriously doubt it's true, and we don't want anyone here ending up looking like a myth-loving chump, do we, Alkan.
I'd love to know, for starters, what sector size is used by the T5 and T7 internal controllers. I doubt that they have the variable sector option, and I seriously doubt that they're each different. Anyway, both these products seem aimed at the fiercely competitive mass market rather than enterprise and high-end professionals, and so Samsung understandably keep a lot of the engineering design detail out of the public eye.
Am I right to assume that VSL libraries are stored as tons of small files, so when a preset is loaded random reads determine the loading speed?
The actual files are several gigabytes each, but the parts of the file are not necessarily accessed sequentially, so it's similar to reading small files.
I'd imagine (but I don't know) that the player and files are designed in such a way that some large chunks can be read sequentially.
Thank you and well done, John. We sorely needed that crucial piece of evidence. I think we can rest our case there.
I've been playing various full Sy Dimension Strings presets live with two big handfuls of notes (playing all 5 sections together - 32 individual instrument players), without the slightest hint of any streaming starvation from my T7. Previously trying to do that while streaming from the iMac's internal HDD was pretty hopeless - sample streams plopping-out all over the place - yet Synchron Player's speedometer says my T7 is slower than the HDD.
Dear VSL, I hate to be a thorn in your side in these difficult times, but in this case I wonder if you would very kindly acknowledge that it appears the ball is now in your court.
Martin, many thanks for your comment.
I know woefully little about how MacOS actually works and had a vague notion about the SSD part of a Fusion Drive as being simply extremely fast static storage for much of the OS itself as well as various apps and stuff favoured by Apple as being important for "the Apple Experience".
So does MacOS prefetch from the HDD and cache stuff in SSD, somewhat like CPU prefetch caching? (It's ok, Martin, I'm not asking for a MacOS tutorial, lol.) Whatever, I'm certainly prepared to agree that the Fusion Drive may seriously complicate the issue here.
When I get time I'll try moving a big VSL library back into one of my bootable clone backup hard disk drives (LaCie D2 Thunderbolt with 7200rpm Barracuda delivering up to 220 MB/s best read speed), then boot the iMac from that drive and see what Synchron's speedometer says. And while booted in that HDD I could also try switching the Synchron Player's library path back to the external T7 SSD and check Synchron's speedometer readings again. I'll get back to you with test results asap.
Martin, here are the new test results I promised.
After booting iMac from external LaCie-enclosed Barracuda 7200.14 cloned High Sierra backup HDD, formatted as Mac OS (Journaled), connected via Apple T3/T2 Thunderbolt adapter to Thunderbolt 3 port, I ran 2 tests.
Test setup same as posted above - Clarett 2pre Thunderbolt audio buffer: 256; LPX 10.4.8: 1 instrument; Synchron Player: 4 threads loading; 4096 preload; force all slots enabled.
VSL preset selected:
15 SYzd Dimension Strings/Dimension Strings | Full/01 1st Violins/1st Violins full - All players (2.64 GB)
• Test 1: Synchron library path: external boot drive LaCie/Barracuda.
Time to load: 21 mins 38 secs (including initial cursor beachball for 15 secs). (Roughly 2x slower than path from iMac Fusion Drive.)
Synchron speedometer after preset loaded: 50.9 MB/s (About 4x slower than path from iMac Fusion Drive.)
• Test 2: after changing Synchron Library path to external T7 SSD connected via thunderbolt 3 port, preload 4096, then rebooting again from the external Barracuda.
Time to load: 53 secs (including initial cursor beachball for 13 secs)
Synchron speedometer after preset loaded: 200.5 MB/s
So ok, the case is altered. In the previous tests it appears my Fusion Drive made substantial differences to loading time and Synchron speedometer readout in the case of Fusion Drive library path. I hadn't understood that was happening, having assumed that only the iMac's internal HDD was involved in that path.
However, I'm still extremely puzzled by @johnstaf's Synchron speedometer showing 600 MB/s for his external T5 streamed into a Windows machine.
Moreover, I don't know how to account for the fact that Synchron speedometer showed only a 4x increase in speed when using my T7 instead of the Barracuda, yet the loading times imply a 24x increase in speed for T7 versus Barracuda.
Moreover, I don't know how to account for the fact that Synchron speedometer showed only a 4x increase in speed when using my T7 instead of the Barracuda, yet the loading times imply a 24x increase in speed for T7 versus Barracuda.
When you load a library into RAM, a lot of the data will be read sequentially from the SSD. The Synchron player tests non-sequential 64k reads, which have a lower throughput of data.
John, I'm well aware of implications of the distinction between 'sequential reading' and 'random access' from bulk data storage, thanks all the same. (The point I was making to Martin was somewhat rhetorical.) My point is the Synchron speedometer seems counterintuitive. If it showed the result of actual reading from storage (say like in the Mac's Activity Monitor) then at least users could get the hang of the performance of their various storage drives under various different conditions. After all, some computer-based music productions can get to be pretty hairy, wing-and-a-prayer, seat-of-the-pants stuff, and in those circumstances it can be very, very useful to have some idea of what's actually going on under the bonnet.
But no, even when loading a big, nicely-ordered factory preset it seems the Synchron speedometer ignores what's actually going on and stubbornly reports the result of its worst-case testing. That's what I call counterintuitive. It's worse than trying to mix with only PPMs on the meter bridge. Well, theory and practice often tend to diverge quite a bit in the real world.
