42.5 GB file transfer from External SSD
M2 MBP 256GB: 2:17
M2 MBP 512GB: 0:23
This is the benchmark problem in a nutshell. This doesn’t jump out to anyone as weird? It was weird enough to me that I actually watched the video.
Long story short: The M2 MBP 256GB did the copy in 32 seconds versus the 512GB in 23 seconds, MaxTech is the hackiest of hacks, they are everything that's wrong with YouTube culture, they're being financially rewarded for spewing nonsense, and they can't even number their graphs and charts correctly, let alone calculate the data right.
Long story long, and remember the debate here isn't about whether the 256GB drive is at all slower (it is, it has to be), or why, or Apple's ethics, it's whether benchmarks are being used appropriately and meaningfully in the discussion:
[0:24] "Does the SSD speed matter when there are no other variables whatsoever?"
[0:50] "I am sick of making videos on this topic and this will be the last one."
[1:21] "I believe it is our job as reviewers to do high quality, detailed, real world testing."
[1:40] "Now we don't like any drama, any beef..."
[3:00] "At which point people are saying this is all true, but your titles are just going over the top. [...] I dunno, maybe we're just covering these topics in too much detail?"
So no drama, except yes drama, everyone on YouTube loves drama, maybe people don't like us because we're too detailed...
Blue MBP is the M2 256GB, Red is the M2 512GB.
On to the Blackmagic results:
Note, that the purpose of this benchmark is to determine if your drive is sufficient to read and or write video streams. Both drives show all green checkmarks except if you are streaming 12k resolution at 60fps. So if the fact there's insufficient space for 3 min of footage wasn't enough of a hint, it's worth knowing that you probably shouldn't use a base model MacBook Air with a 256GB disk for your 12k ProRes project.
The only thing this benchmark does is open a file and write to it procedurally (no work is being done to create the content) and then read that file back into a void (no work is being done with the data read)-- just a pure push disembodied data through the pipe test. Real world stuff.
Just to mark these numbers down: the 512GB drive is 85% faster on reads and 70% faster on writes compared to the 256GB.
Ok, now the test that brought me here. The external 42.51GB file copy, which we're told is completely normal for someone with a 256GB Air (?) and that shows the 512GB destroy the 256GB.
512GB M2 MBP: 0:23
256GB M2 MBP: 2:17
OMG!!!111 That's crazy! Why would anyone buy that?
The difference, we're told, is all to do with the SLC cache. The 256GB slows down because it fills its cache.
After that, it "is actually writing at about 200MB/s"
Now I don't know how they round, but doesn't that look closer to 300MB/s than 200? I mean, if you were going to round, wouldn't you round to 300?
We're told most reviewers don't talk about the cache, while he shows the read rate over time for a 1TB Samsung 980 Pro-- so I guess he's not going to talk about cache either. If it was really his job to provide us with high quality detail, wouldn't he show the read rate over time of the Apple SSD rather than some random Samsung?
Does Apple really use an SLC cache? How big is it? Wouldn't the 512GB also have a cache? Wouldn't you expect it to be twice as big at most? Tell us about the 512GB cache!
Wait... This looks a bit dicey, doesn't it?
How much space is left?
3 freaking GB remaining on the boot drive and the copy still isn't complete!
He's writing a 40+GB file right up to the last 0.8% of the boot drive capacity.
Does that seem like a speed test with "no other variables whatsoever"? Does that seem like a drive speed test that any sane person would do? Does it seem even close to real world? Was part of that 2+ minute copy spent clicking away incessant "your startup disk is almost full" messages?
Isn't that the type of detail someone should disclose?
[4:30] "We're not maxing it out because our external drive is slower"
Wait, so you're using an external drive to test an internal drive, but your external drive can't even keep up? How fast is it? Shouldn't you have started with a Blackmagic test on that external drive so we know the test you're doing is even relevant? Let me guess (I don't have to guess, I know): you're going to mention that a few times to imply without proving it that the results you're presenting aren't even as spectacular as they could be. No drama, just high quality detail. Maybe so much detail that people are complaining...
He points out that some annoying people are sure to point out that the M2 still transfers faster than the M1:
"But there's a
big caveat: you have to have your hard drives filled similarly for it to be accurate".
That's a caveat? I thought that's what "no variables" meant...
[5:20] "This one wasn't filled up all the way."
Do you believe the cajones on this guy? He finishes a transfer with less than 1% free and claims it's not full. But that's realistic, we're told. If you're "getting along with a base model" that 98GB of documents is less than one video project, or if you go out and do a photoshoot with RAW files, the kinds of things the base model is meant for, then it's going to fill fast.
58GB of apps, but "a lot of these came with it", except for the ones that didn't because it looks like all of MacOS requires just
26GB for a clean install.
But he obeyed his caveat, right? He did his test with the hard drives filled similarly?
Not a chance:
Plenty of space on that 512GB machine.
Ok, so what if we got really picky and insisted that he delete the PC test folder from Documents and try the transfer again?
So the actual benchmark comparison is:
512GB M2 MBP: 0:23
256GB M2 MBP: 0:32
Which I will bold and italicize because the video makes a strong effort to hide that fact.
For example, the full screen graphic shown when he announces the comparison:
Wait, what? Talk about staying on message! Oh, and nobody forget the external drive sucks!
But, we're told, this is why some people falsely claim the transfers are quick-- because people are doing copies when they still have space. If you fill the drive, let's make sure we show the bogus number again, this time with an arrow for emphasis:
If you don't want the 2:17 time, "make sure at least half of it is free [...] which is a major bummer"? No drama there! You can only use half your drive if you want it to perform!
Now by this point we'd almost forget that the whole reason we were told the 256GB machine was slower was because of the SLC cache being too small and the 512GB having a bigger one. But delete 100GB of files and free up more than 47GB of space for your 45GB transfer and suddenly it's not a cache problem anymore? Hmmm... I'm starting to wonder if this guy understands how technology works...
Turns out the 512GB does slow down eventually when you need to make it look bad against a 16" MBP, "to about 350 or so":
which jumps around a bit, but I find it humorous they draw the arrow when it says 304 and call it 350 but on the other drive draw the arrow when it says 283 and call it 200. And now they're testing the 512GB with a 1TB external drive and less free space on the local drive and are now transferring a folder of files for some reason which is always slower, but at least they're testing with no variables whatsoever.
While the 256GB version we're again told slows down into the "200MB/s range":
Which I suppose would be technically true even if it were 299, but still seems egregious to me.
Now he "gets" that the 128GB might not be available, but still refers to people point that out as "Apple apologists". No beef.
"But take a look a this!" as he brings us to
Apple's supplier database Amazon: There's a random SSD there we can talk about for $21. "It's not the price point that's getting them in supply chain"? While that's almost certainly an accidentally true if grammatically odd statement, finding a random drive on Amazon is not the Apple supply chain.
[7:55] "And this isn't a sucky one, this is faster than what Apple is using."
No it is not. It's SATA III. If he'd read the reviews below he'd find that it maxes out with a read speed of about 450MB/s, not 1500MB/s and certainly not 2700MB/s. Even two of those in a RAID aren't faster than what Apple is using.
Even the seller says it tops out at 500MB/s
Perhaps it's worth a reminder here that the point he's making is about what part Apple should put here:
Maybe that's why he leaves Amazon to bring us to the other well known Apple tier one supplier ChinaFlashMarket (??) to show "the market price for one of these 256 gig TLC chips". He doesn't reveal any technical data for this particular chip except exactly one thing, the capacity.
Yeah, that's right, the only technical detail given is that the part he's referring to as "one of these 256 gig TLC chips" is a 256 gigabit part, not a 256 gigabyte part.
Is anyone still going to argue that MaxTech knows anything about technology, let alone operations and supply chain management?
But we're not done here. Now it's time to run Lightroom where he shows the performance with virtual memory with swap. Note the SSD capacity shown in the menubar for the 256GB model:
It is still using a completely full drive, and now thrashing whatever small remaining drive space there is between write/erase cycles in the SSD. He doesn't show the available drive space here, but that bar looks a lot like it did when there was 3GB free
and here he's using almost 4GB of swap space:
This is the absolute worst case scenario for an SSD. Unlike RAM or even an HDD, a flash block needs to be erased before being rewritten and erasing is a much, much slower exercise than reading or writing. Usually this is solved by always writing to unused portions of the drive while the system erases pages in the background in a batch, but with no free space there's nowhere to write while erasing.
Meanwhile the 512GB model:
But of course he claims "right now the only thing different is the SSDs, everything else is identical." which is embarrassingly false.
The results of exporting 50x 42MP images?
512GB M2 MBP: 1:45
256GB M2 MBP: 2:00
Which he points out "is a performance difference of
15%!"
He says it like that's proof how bad the SSD is, but remember the raw SSD write performance way up at the Blackmagic test showed the 512GB was 70% faster.
So, 100% more chips leads to a raw performance difference of 70% and a real world performance difference of 15% when pushing 2.5GB of data to a drive that appears nearly full while also swapping. That sounds like the real world performance is better than the benchmark. But that's not the story being told by MaxTech because it doesn't fit the narrative.
Then there's 45 seconds of drama about how disappointing it is that the export took 15 seconds longer.
Then he starts opening 10 browser tabs in Chrome (not Safari) to show multitasking performance. This pushes swap up toward 6GB, so we know the swap is active. No ad blockers, so both machines are running different ads in their browsers.
Same story. Running the 256GB at the edge of capacity.
Of course even the 512GB took twice as long in this scenario, not that the chart they published makes any sense:
Yay for high quality details! (That lower bar should be labeled 4:20 and that's a number not a comment on the presenters state of mind)
His explanation for why even the 512GB is slower? The M2 needs more ram than the M1. Wut?! Same OS, same instruction set, same workload, but the M2 for some reason needs more RAM? And what does this have to do with the SSD?
In this test, unsurprisingly given the drive architecture, available space, additional need for swap, and uncontrolled variables in the testing the 256GB takes 80% longer (7:43). That difference is, for some reason, not attributed to needing more RAM, but entirely to the SSD.
And, of course, that wasn't wasted time, because he's multitasking and doing other work while the export is happening because that's what multitasking means.
And I can't even make sense of the next set of comments about some test with the M1 MBP-- something about doubling the RAM, more tabs, some other apps open (ie. no variable whatsoever) giving these results that I can't interpret:
It's at 12:30 in the video if anyone else wants to try to make sense of it (leave your comments below!).
Then there's some weird sidebar about how their audience is really into tech and is buying 3x as many base model Airs, so imagine how sad it is that people who don't know anything about tech are also buying the same hardware that power users are perfectly happy with.
Now a FinalCut export while multitasking:
What's interesting here is that not only is the SSD near capacity on the 256GB again, we're used to that now, but the 256GB has consistently and significantly higher CPU and GPU utilization throughout the portions of the test we can see. How does one explain that if the export is being bottlenecked by SSD accesses for both export and swap? Doesn't that warrant an explanation and ideally an isolation of that variable?
The 256GB is again slower, no surprise, but he explains that's with "nothing else open", which is straight up wrong because he had a bunch of Chrome tabs open and was playing 4k 60fps YouTube content.
Then what I think is a great comment to close with: "I don't know what would have happened if I had 4 pro apps open and 25 tabs instead".
Yeah, if only there was a way for a reviewer focused on high quality, detailed results to know that... 🤔
His conclusion after trying to run production workflows through a base model Air? It doesn't deserve to exist, and it's only there to get people to upgrade to a 14" pro.
The stated point of the video wasn't to demonstrate that a 256GB SSD 99% full will slow down, it was to, and I quote:
with, ahem, no drama or beef.
The testing here is so flawed and wrapped in hype, to be useless even as an indication of how this will perform in the real world. Any result contrary to the message is quickly buried. It's just a stream of consciousness from someone who fancies themself a pundit.
So can we please all agree that whatever people think about benchmarks, sites claiming to use them like MaxTech are a complete joke? They don't know what they're talking about, they don't know how to do proper testing, they have an agenda behind their data, they're extrapolating beyond what their data can possibly support and it's feeding junk science to a bunch of people who then go into the world and preach it like gospel?
Somebody else will have to tell me if this video from July 1 was the last video they made about SSD speeds as promised, because I refuse to click on another MaxTech video to check.