52 comments

[ 4.4 ms ] story [ 101 ms ] thread
I'm happy that I stayed on M1 MacBook Pro.

I wouldn't be surprised if these numbers translated to slower app startup times, a slightly faster CPU won't balance it out.

Correction: somebody already noticed that Lightroom got slower.

That makes sense sadly. I imagine most people use LR on their desktops though. I can't fit half of my library on one of these SSDs.
Lot of people need it while travelling, and the cloud version downloads only the data you ask for.
Hmm, I may give that a try, thanks for the tip..
Note that using Adobe products with their cloud services grants them permission to use the images in AI training.

Not everyone cares, which is valid, but it's also kinda hidden.

Let’s dwell a little on the insanity that Lightroom gets slow on an drive with ONLY 2.8GB/s read speeds
Its probably more iops effecting performance than transfer speed, imagine going from a 4 lane highway to a 2 lane, of course reality is a bit more nuanced.
(comment deleted)
Most likely. But these SSDs do well in iops too.
Except in this case, we've gone from 5 billion lanes to 2.8 billion, and I'm fairly certain that this should be more than enough lanes to load a UI and some cached images.
I'm just happy that I finally found the courage to ditch LR and Adobe a few years ago. Seeing that their performance issues are still plaguing them doesn't surprise me in the least.

Switched out to Photo Mechanic for cataloguing and Capture One / Affinity Photo for editing if anyone is interested.

For ipad, Capture One subscription is way costlier than Lr.
True, hence the Affinity Photo app. Good for on-the-go editing and syncing back to my laptop
What do you mean by "happy you stayed on"? Do you usually upgrade your laptop immediately to the next one the moment it's released? Or do you mean you opted to buy an m1 after the m2 pro was already available?
I was planning to upgrade, but I won't spend money for staying on the same TSMC process node (I hoped Apple waits for 3nm with M2).

Right now though I feel that the new laptop is not even superior, it's similar category (still awesome).

One thing to remember is that random I/O did not get slower: the benchmark they used is a very simple one simulating writing a single 16GB stream as quickly as possible — unless your job is something like copying high-resolution video files around, you're unlikely to see it slower.

That does affect a few people but consider what it means: you can fill the drive up in a little over 2 minutes on the M1 and a little over 3 minutes on the M2. The only way that's relevant to you is if you have some other high-end hardware which can stream data that fast _and_ you're doing relatively little processing on that data _and_ you don't actually store that data for very long (otherwise you need some kind of attached storage array).

Lightroom is a terrible app so to the extent that it needs a high-end SSD to start up, that's telling us what we've already known for a decade: Adobe doesn't care about quality and it's a shame the very noticeably faster Aperture was discontinued.

Kind of expected more than a tweet with two screenshots and "more to come."
Loved the "BREAKING: ..." as if we should all drop what we're doing and pay attention to "more to come".
well, that is the definition of "breaking news", more is to come. You don't have to drop anything if the topic is not of interest to you, but if you're buying a laptop right now you might care.

when dawn breaks, we have a whole day ahead of us.

Really all we need is a chart with 4 or 5 scenarios and throughput/latencies for 90% of readers I think. I wish more blogger thought that way.
Has OWC or some other company come up with SSD upgrades for these machines yet? Would adding another chip even increase performance?
Afaik the storage devices are directly soldered to the 'logic' board and are unable to be replaced by the end user.
The SSDs in these machines consist of NAND chips soldered directly to the motherboard; the controller aspect is handled within the main M* SOC. As such there's no way that a company could offer an internal 'upgrade' or expansion to these drives.
This is not true . The SSDs on this are not soldiered but are in distinct chips that can be removed but none of them have a discrete controller on them instead the controller is on the SoC itself. Basically they are NAND modules not full SSDs at all.
The NAND chips are soldered with ball grid arrays, requiring a reflow oven to add a new chip, not to mention how the board itself would probably not even detect the extra chips.

Here's an M2 Air logic board with 1 of 2 NAND chips soldered on (in yellow), there's an empty spot on the board for another chip: https://guide-images.cdn.ifixit.com/igi/1A2NHYEY6dJcUcd4.hug...

Most likely from halving the number of SSD chips and no longer striping.

Though to be honest in day to day use, I only notice SSD speeds when transferring/loading video files in the tens/hundreds+ of GBs.

Otherwise, as long as the latency is low and the machine can load applications quickly from the disk, I never can perceive the difference. And with M1/M2 machines being able to start applications significantly quicker than my intel apple machines ever could, I'm not sure I could tell a perceivable difference in day-to-day use.

But I am still disappointed that they are not using faster SSDs.

Well, I just tried that tool on my M1 Max and I got read speed of 2900MB/s. So I ran it again and finally put it in continuous mode: it did not keep an average and quartiles. I saw a read value over 4000MB/s. I would say that this tool is not a valid measure.
They should use AmorphousDiskMark. But I don't doubt the results in general. It looks like Apple has a hard time sourcing new 128 GB Nand chips.
(comment deleted)
Same as the M2 MacBook Air with 1 256 GB Nand chip. The 512 GB M2 Pro has 2 256 GB Nand instead of the previous 4 128 GB.

Probably no big deal for most users.

No wonder I was having deja vu.
These machines are still insanely fast. If you actually notice anything get substantially slower, you are either running very specialized workloads or you can forward the blame right to the developer of said application.
It seems pretty unlikely that many people have a workload that is highly sensitive to read throughput, has a disk footprint so small it fits on a 512GB SSD, and where the cost of slower disk outweighs a 20% faster CPU.
If I select 200 files and make a zip, that's probably an impacted workload.
As in 20% faster on the M2? At least for gzip defaults the per core throughput on my M1 is around 50MBps.
As in, possibly slower on the M2. Zipping a complex file tree can be disk-constrained, particularly if there are many small files.
For scale we're talking > 2TBps linear read and probably > 500MBps random 4k read in both cases, compared to on the order of 50-500MBps limit on the CPU side depending on threading etc. So still CPU limited. If there's any problem with big file trees it's probably macOS's relatively slow filesystem so again, advantage to the faster CPU. It sounds like LightRoom startup is a little slower but if you use it for editing you're going to get that back in faster workflow. In general modern NVMe has removed disk as a significant bottleneck for normal peoples' workflows, even on Mac. That includes backups, zipping up collections of files, git compaction, etc. What scenario are you imagining here?
Brain fart, 2GBps not 2TBps.
I'm imagining the large filetree zip. I've got an M1 mini at home, so I can run a test on that later today. Maybe some parts of Xcode builds are limited by the disk too.
Xcode depends on random I/O, which is reportedly faster.
The Twitter OP elaborates further down the thread that they're seeing real-world impact on Lightroom. That would be a bummer for me...but I'd probably buy a build with a larger drive. Waiting for confirmation that this is only on the base model.
This kind of "Breaking" news is creating drama where there isn't one.

The 'slower' SSD is still pumping out gigabytes of data per second. Gigabytes!

The real drama is that they focus on the wrong metric: they should focus on random IOPs and latency performance, throughput above certain thresholds is meaningless.

I repeat, nobody is going to notice ever.

Somehow, I'm not surprised at all to see Apple users openly defending the fact that they get worse hardware when they buy the newer, more expensive model.
(comment deleted)
I am a bit surprised. Do they think they are getting points from apple to redeem for a lunch with steve jobs or something?
If apple allows their ssds to get any slower I’m afraid those off the shelf nvmes might start being the faster option…
Is there any way to tell how many SSD lanes you have without opening up the device?