Tonverk SD Card Performance Test (Stock microSD vs Verbatim SD)

I was curious whether changing the SD card in Tonverk would actually affect loading times, so I ran a quick comparison.

Compared:

– Stock 64GB microSD (included)
– Verbatim 32GB SD

Measured:

– Power On → Factory Project Ready (DONE)
– Full Factory Project + all samples loaded into RAM
– Sequential read speed
– Random 4K read speed

Results:

Power On → Factory Project Ready
36.37 s (stock)
33.34 s (Verbatim) ~8% faster

Factory Project + all samples into RAM
6:16.57 (stock)
2:54.38 (Verbatim) ~53% faster

Sequential read was almost identical (~90 MB/s).
Biggest difference was 4K random read:
6 MB/s vs 10 MB/s (~66% faster).

So it seems random read matters much more than sequential speed for sample-heavy projects.

Has anyone else experimented with different SD cards?

2 Likes

Interesting follow-up:

Another Tonverk user tested a Samsung card with even better random read performance and reported almost identical results for:

– Power On → Factory Project Ready

– Factory Project + all samples into RAM

So it might indicate there’s a performance ceiling somewhere in the system once a certain SD speed threshold is reached.

Curious what exactly becomes the bottleneck at that point — controller, filesystem, CPU, internal buffering?

3 Likes

Excellent post, curious to read more as you experiments develop.

Do you feel the difference in loading time has to do with the difference in storage? i.e. Does the Verbatim load faster just due to being 32 vs. the original’s 64?

I don’t think it’s about 32 vs 64GB.

Sequential speeds were nearly identical, but random 4K read was significantly different — and that’s probably what matters when loading many small sample files.

So it seems more related to actual random access performance than capacity itself.

It’s a shame Elektron didn’t go for a UHS-II interface.

1 Like

Quick update after a clean re-test.

I fully reformatted the Verbatim card (MBR + ExFAT), copied only the Factory Content (no user packs), and repeated the measurements.

Results:

Power On → Factory Project Ready (DONE):

33.95 s (no real change)

Factory Project + all samples into RAM:

1:55.46

Previously I was getting ~2:54 with additional user content on the card.

So after a completely clean format and factory-only structure, load time dropped by almost a minute.

This suggests that overall file count / directory structure may also influence load time — not just raw sequential speed. Random 4K performance still seems important, but card state definitely plays a role.

I’ll continue testing with user content added back to see where the slowdown starts.

2 Likes

If random read is what Tonverk needs then you should try the Kingston Canvas Go! Plus (128gb or smaller). This is the card used in M8, which also depends on random read.

For such a machine with GB of samples in mind, an SD express would have been best. Maybe for an mkii…

1 Like

Just to add — another Tonverk user tested a Samsung card with even stronger random read performance and reported very similar loading times.

So it might suggest that beyond a certain level of SD performance, the gains become minimal.

Here’s the reference for context: Tonverk SD Card Performance Test (Stock microSD vs Verbatim SD) - #2 by MusicYet

The Dirtywave M8 uses cards that have a good sustained read speed, because it streams all its samples from the card. The peak speeds usually tested don’t apply.

I wonder if something similar is happening here? The TV is loading a whole bunch of stuff in one go, which might call for a good sustained read speed?

Dirtywave has a list of recommended cards, but they’re all microSD, not SD. Still, perhaps the full-sized versions of those are worth testing if anybody has them:

Brand Size Tested
Kingston Canvas Go Plus 512GB
256GB
128GB
64GB Best*
Lexar 1800x (discontinued) 128GB Best
Kingston Canvas Select Plus 32GB Good
SanDisk Ultra 32GB Great

*On firmware 3.2.0 or newer.

Elektron has given a recommendation for what SD card to use. Users understood it like: don’t go below these specs, but if you go above, nothing will improve.

1 Like

Why wouldn’t they word it as “at least 104 MB/s“ instead of “as close as possible to 104 MB/s” if that’s what they actually meant?

very good thread , actually i’m getting mad about save times. my project is loaded with only 600mb of samples but i start having lots of patterns and saving the project takes more than 1 minute each time ( i like saving each time i do a significant change in sound design or composition ) . can you give us the entire reference for that verbatim card you tested ?

Imo the issue is all about indexing, not speed. Common sense would dictate to create a master index and if the control sum changes to re index again. Meanwhile, I feel like the device is indexing files from scratch with each new project.

1 Like

toneverk has a lot of horsepower but it eats crayons rn

1 Like

You can just run a microsd in an adapter!