October 18th Apple Event - Unleashed - Apple Silicon, MacBook Pro upgrades, HomePod mini, AirPods 3rd Generation

[hishnash]

46 minutes ago, leadeater said:

NAND write is never faster than NAND read,

It is possible that the limits on read speed are not due to the NAND at all. All data read/write on the internal storage has full disk encryption turned on. (even if you do not have FileVault enabled the data on disk is encrypted its just encrypted with a key that is not a function of your pw).

It is possible that the dedicated decryption pathways are more limited than the encryption (this would depend on the type of encryption schema they are using and the die area they are spending on this.. they do not use the cpu cores for IO encryption/decryption).

The back magic test is a valid test if your going to be working with their products (like resolve), it's just has valid as Cinebench is a valid cpu benchmark (very useful for people doing CPU rendering in Cinema 4d but otherwise not that relevant at all).  For most users of these devices the majority of heavy extended IO will be in applications like resolve.  

For sorter random busts the effective disk perfomance is much higher than this as the SSD controller is using LPDDR5 memory (and even on die SLC) as its cache. Happy to do some tests (i will be getting a 1TB version soon).

[Dracarris]

1 hour ago, leadeater said:

As a literal storage expert and engineer you are wrong if you think Apple's SSDs are faster than 980 Pro lol. The only pigs flying are those believing burst speeds mean anything at all, they do not.

I think he was more referring to the combination of 8TB, its speed and being in a laptop. The 980 Pro in the M.2 form factor caps out at 2TB, you'd need to put 4 of them in a laptop to match it.

 

Also, the price of 4 980 Pro 2TB is not that far off of Apples 8TB storage upgrade asking price (yes still higher but not sth crazy like 2x).

[saltycaramel]

1 hour ago, leadeater said:

As a literal storage expert and engineer you are wrong if you think Apple's SSDs are faster than 980 Pro lol. The only pigs flying are those believing burst speeds mean anything at all, they do not.

Faster and/or blessed with some kind of advantage before the socketed counterparts catch up (like the 8TB offering). That’s what happened in the last 2 years, we’ll see in the coming years if it was an exception or the norm.

[leadeater]

13 minutes ago, saltycaramel said:

Faster and/or blessed with some kind of advantage before the socketed counterparts catch up (like the 8TB offering). That’s what happened in the last 2 years, we’ll see in the coming years if it was an exception or the norm.

I agree that what Apple has done has allowed them to get much larger capacities, that's a benefit, but it's not any faster. You are aware you are trying to claim Kioxia TLC NAND is faster than Samsung MLC NAND right? It's not, it's quite a lot slower so you need a lot more to get the same seq performance however small block I/O is always going to be slower. Once you run any real actual workload on it and get past the burst performance, which is all Blackmagic etc report, then it's not quite as fast. Maybe the 8TB one would with all 8 NAND modules but until a proper storage benchmark shows it to be the case then it is not the case.

 

If you want to say Apple's SSDs are faster than a 980 Pro then you're going to need to prove it and Blackmagic Disk Benchmark is not proof and neither is spec sheet burst rate seq proof either.

 

Samsung 3bit MLC is vastly superior to Kioxia BG4 TLC.

 

Each one of those 1TB TLC modules is only good for a sustained 4k write of 75 MB/s, so the 8TB implementation is good for 600 MB/s. Contrast that to the 980 Pro where both the 250GB and 1TB model can do 680 MB/s. So I'm willing to say Apple's 8TB of Kioxia TLC is roughly about the same performance as a Samsung 980 Pro, Apple's 4TB and lower not.

 

36 minutes ago, Dracarris said:

I think he was more referring to the combination of 8TB, its speed and being in a laptop. The 980 Pro in the M.2 form factor caps out at 2TB, you'd need to put 4 of them in a laptop to match it.

He was saying both, he's totally out of his mind and gushing all over Apple over multiple things with zero understanding and no technical proof that it's any better than anything else, or if it is why and in what aspect. Forgive me for countering blind bluster but I'm going to do it. Apple only deserves credit for actual real things and for the specific things that are good, nothing more and nothing less and which I've already done.

 

Anyone saying MacBook Pro SSD is faster than 980 Pro is living in total fantasy land.

 

1 hour ago, hishnash said:

It is possible that the limits on read speed are not due to the NAND at all. All data read/write on the internal storage has full disk encryption turned on. (even if you do not have FileVault enabled the data on disk is encrypted its just encrypted with a key that is not a function of your pw).

Encryption would still apply to read and write. The write is showing as faster because all it's really showing is the write cache, of both the DRAM and the SLC portion of the Kioxia BG4. Once you actually start copying your videos files to the devices, open it in w/e editor of your choice and start doing real things caches start to run out and the backend NAND starts to matter more.

 

The performance required however is so low it's never going to be an issue anyway, the whole thing is a total non issue and zero point. The only reason I care is about factualism, someone's opinion on something being faster is not equivalent to factual correctness of that belief. 

 

[leadeater]

1 hour ago, hishnash said:

The back magic test is a valid test if your going to be working with their products (like resolve), it's just has valid as Cinebench is a valid cpu benchmark (very useful for people doing CPU rendering in Cinema 4d but otherwise not that relevant at all).  For most users of these devices the majority of heavy extended IO will be in applications like resolve.  

No it isn't because that benchmark is not the same workload as is actually working with those software and file formats. It's only testing burst rate performance of certain block sizes and queue depths. Unless you are only video editing for a few seconds at a time then it is a valid test, otherwise not.

 

The issue is the test time length, not the block size or necessarily queue depth settings of the benchmark.

 

It's a "Is it probably good enough" tool and it's great for that, it's useless for trying to discern the performance differences of various SSDs to some proper meaningful way.

[BondiBlue]

1 hour ago, hishnash said:

The back magic test is a valid test if your going to be working with their products (like resolve)

That's not how that benchmark works.....

[Roswell]

6 hours ago, leadeater said:

Also Blackmagic disk benchmark is not a good benchmark tool.

It does exactly what it's supposed to do. It's a benchmark for sequential speeds for people working with large sequential writes and reads in video software. 

 

3 hours ago, leadeater said:

No it isn't because that benchmark is not the same workload as is actually working with those software and file formats. It's only testing burst rate performance of certain block sizes and queue depths. Unless you are only video editing for a few seconds at a time then it is a valid test, otherwise not.

 

The issue is the test time length, not the block size or necessarily queue depth settings of the benchmark.

 

It's a "Is it probably good enough" tool and it's great for that, it's useless for trying to discern the performance differences of various SSDs to some proper meaningful way.

You're misinformed. The default for the program is to write 5GB chunks and read them back, write another 5GB chunk and read it back, etc etc. Any "burst speed" as you put it is out the window in extremely short order unless you actually believe Apple SSDs are packing like 64GB of DRAM cache.

[saltycaramel]

How many times have we read “you’re benchmarking it wrong” since the M1 Macs have been released. That’s almost a meme by now.

 

Someone was too quick to say the 980 Pro is faster than any Apple SSD and now we’re splitting the hair to make that somehow work. But I’m “out of my mind”..(for those semi-trolling “thank you mommy Apple” joke posts..)

 

Apple’s 4TB and 8TB SSDs are crazy fast. Capacities are crazy high since 2019. This is a winning approach and has room for yearly improvements. Standardization of sockets and upgradeability are important for the outward facing I/O, i.e. thunderbolt ports, HDMI, etc. For the sealed off innards, it’s less important and I’ll take the design flexibility of Apple’s approach every day, if these are the results. 

[leadeater]

2 hours ago, Roswell said:

You're misinformed. The default for the program is to write 5GB chunks and read them back, write another 5GB chunk and read it back, etc etc. Any "burst speed" as you put it is out the window in extremely short order unless you actually believe Apple SSDs are packing like 64GB of DRAM cache.

I know how much data the benchmark writes, and you know it can also be changed as well? You know data size is not the same thing as block size? 5GB falls within the DRAM and SLC cache aka it's burst performance. And yes on the 8TB models SLC cache is that large, in fact lager. The SLC cache portion on a 250GB 980 Pro is 49GB so I'd hope an 8TB MacBook Pro has greater than 49GB of SLC cache.

 

A 2TB 980 Pro has 2GB of DRAM, industry standard is basically 1GB per 1TB. Apple I suspect is dynamically allocating based on total system DRAM utilization so it wont be any fixed size.

 

If you have a M1 Pro/Max device then I suggest installing fio and writing out 1TB of data with a block size of 128KB and QD of 4, watch for the point where the performance drops.

 

If people could stop telling me, a literal storage expert, about my own job that would be great. But this is the internet so not going to happen... 

 

1 hour ago, saltycaramel said:

Someone was too quick to say the 980 Pro is faster than any Apple SSD

No it is faster, I said it doesn't matter and that you'd need 8TB to be equivalent. The difference is irrelevant, but it's faster than Apples Kioxia TLC for sure 100%, in every single way. If you could kindly stop being so insufferable, thank you.

 

If you're going to post utter nonsense don't be surprised to be ignored, I don't at all mind somebody liking a product but you're talking so far off in to space I worry for you sanity. 

[Forbidden Wafer]

9 minutes ago, saltycaramel said:

Standardization of sockets and upgradeability are important for the outward facing I/O, i.e. thunderbolt ports, HDMI, etc. For the sealed off innards, it’s less important

Nope, it's just as important for cheaper and easier repair plus reducing e-waste.

 

Apple on the other hand charges a new complete motherboard (usually a badly refurbished one) for a single broken screen cable that they made too short.

 

All of that while subcontracting sweat shop repair centers to refurbish stuff, some of which have workers stealing and reselling user data (apple decided to settle that case instead of taking the lawsuit to court).

[saltycaramel]

At the risk of sounding insufferable and pedantic: if we were comparing “final SSD product vs final SSD product” (comprising of controller, NAND modules, layout of said modules and how many of those modules you could potentially use in parallel, among other things), and that’s what the Apple’s approach I nonsensically praise is about (not the NAND modules per se, who’s stopping Apple from using faster modules next year?), who decided to move the goalpost to “NAND modules vs NAND modules in a vacuum”? 

 

Blackmagic may not be a sound definitive proof of Apple’s 4TB/8TB being faster, but it’s not like super-definitive proof of the opposite was provided either. 

 

As for how many modules Apple is using, have we double checked that? I vaguely remember a youtube video where 14” vs 16” had a different number of modules (and an odd number at that) for the same capacity, we should look at disassemblies of all the different capacities to be sure about what’s going on. Hopefully LTT review (with all the configs they’ve ordered) will shed more light on this and provide real-life benchmarks against the fastest m.2 SSDs one could fit inside a non-Apple laptop.  

[leadeater]

58 minutes ago, saltycaramel said:

At the risk of sounding insufferable and pedantic: if we were comparing “final SSD product vs final SSD product” (comprising of controller, NAND modules, layout of said modules and how many of those modules you could potentially use in parallel, among other things), and that’s what the Apple’s approach I nonsensically praise is about (not the NAND modules per se, who’s stopping Apple from using faster modules next year), who decided to move the goalpost to “NAND modules vs NAND modules in a vacuum”? 

 

I brought up the NAND modules as technical proof it is impossible to be faster, if you cannot understand that then that is a you problem. The final product Samsung 980 Pro is faster than all Apple sizes except for the 8TB where I'd equate them based on the specifications of the NAND being used to be the same.

 

Unless a NAND manufacturer releases a new generation of NAND then Apple's not going to be replacing them next year with faster ones, using faster but lower capacity modules I doubt is an option. Kioxia BiCS 6 NAND technology should be available for use next year though.

 

58 minutes ago, saltycaramel said:

As for how many modules Apple is using, have we double checked that?

I have not only checked that I know the exact model number of the Kioxia NAND being used.

 

58 minutes ago, saltycaramel said:

Blackmagic may not be a sound definitive proof of Apple’s 4TB/8TB being faster, but it’s not like super-definitive proof of the opposite was provided either. 

Until there is proper storage benchmarks of these MacBook Pros there is no point posting any reviews because there is nothing to compare it against. You can go to storagereview.com and see the 980 Pro review, or any other good reviewer in the storage industry, but unless these MacBook Pros go through the same testing then you've got nothing to compare to.

 

It is on you, or anyone else, claiming Apple's is faster to prove that case and Blackmagic is not acceptable proof.

 

58 minutes ago, saltycaramel said:

had a different number of modules (and an odd number at that) for the same capacity

They don't, the 14" has modules on the back side of the mainboard as well. The odd number you are thinking of was for a lower capacity in the 16" model where 5 NAND modules were being used whereas based on the capacity only 4 would be required. I would take that to mean Apple is using 5 for extra performance and endurance which is nice.

[Video Beagle]

12 hours ago, leadeater said:

M1 Max die itself is about the size of a Nvidia GA104 die (RTX 3070), package size a lot bigger since it's the SoC die and 4 memory stacks. But essentially just link of it like a larger GPU package, around the size of a Vega 56/64, so not so big it would cause any problems getting it in to a Mac Mini.

,,,,,,

 

Could you put that in terms of CD's/45's,/33 1/3 records?

 

EDIT: Did it myself

you and @Spindelwere really underestimating my lack of understanding of the size and scale of these things

[leadeater]

54 minutes ago, Video Beagle said:

you and @Spindelwere really underestimating my lack of understanding of the size and scale of these things

Hmm, M1 Pro about the size of a Credit Card I guess, close enough anyway.

 

 

 

[Video Beagle]

21 minutes ago, leadeater said:

Hmm, M1 Pro about the size of a Credit Card I guess, close enough anyway.

About 1/4th the size.

 

[leadeater]

6 minutes ago, Video Beagle said:

About 1/4th the size.

 

Oh I meant the entire package with the LPDDR and everything. The SoC die is in the center and yea much smaller than the whole thing.

 

My top picture the black squares is the LPDDR5 and the grey thermal paste is the SoC. The whole package is the larger square metal piece inside the marked section in yellow (that goes around it).

[Video Beagle]

3 minutes ago, leadeater said:

Oh I meant the entire package with the LPDDR and everything. The SoC die is in the center and yea much smaller than the whole thing.

You start to see how little I understand any of the stuff at this level of computers

[leadeater]

1 minute ago, Video Beagle said:

You start to see how little I understand any of the stuff at this level of computers

Meh, well if you want to think about it this way something the size of a credit card is faster than most entire gaming PCs and consoles lol.

[Paul Thexton]

1 hour ago, leadeater said:

Meh, well if you want to think about it this way something the size of a credit card is faster than most entire gaming PCs and consoles lol.

Exactly. There will always be faster PCs out there, especially full form factor rather than laptop based, but for the physical and power constraints, Apple have made a genuinely excellent SoC.

 

Some of the YouTube reviewers saying "OMG this specific i9 is faster than the M1 Max at this specific workload, lol!" seem to be entirely missing the point.

[Spindel]

1 hour ago, Video Beagle said:

You start to see how little I understand any of the stuff at this level of computers

 

1 hour ago, leadeater said:

Meh, well if you want to think about it this way something the size of a credit card is faster than most entire gaming PCs and consoles lol.

If we are still discussing of the M1 Pro/Max would fit into the current Mini case look at the pictures here with all the empty space: https://egpu.io/forums/desktop-computing/teardown-late-2020-mac-mini-apple-silicon-m1-thunderbolt-4-usb4-pcie-4/

 

It would be no problems fitting the Pro/Max variants into it.

[leadeater]

5 hours ago, Spindel said:

 

If we are still discussing of the M1 Pro/Max would fit into the current Mini case look at the pictures here with all the empty space: https://egpu.io/forums/desktop-computing/teardown-late-2020-mac-mini-apple-silicon-m1-thunderbolt-4-usb4-pcie-4/

 

It would be no problems fitting the Pro/Max variants into it.

I'd really like to see it happen, even if it requires a thick (slightly) boy Mac Mini to do it. I've always like the Mini because of how small it is but actually still being good.

[Spindel]

6 minutes ago, leadeater said:

I'd really like to see it happen, even if it requires a thick (slightly) boy Mac Mini to do it. I've always like the Mini because of how small it is but actually still being good.

The OG M1 Mini (one of which I'm an owner to) is a beast for its form factor. Before I got iMacs instead because to be frank the integrated graphics on intel just where to weak, I mean my 2013 iMac with GTX780M ran circles around the 2018 Mini with intel graphics and I do not have that high demands of the graphics (but higher than the integrated offerings from AMD/Intel). 

 

But as I said I would be damn tempted to get a M1 Pro or Max Mini if they released one. And I have to stress that I really don't need the performance above the OG M1 that the Pro/Max offer since most of the stuff I do only need ST performance so more cores do very little for me.

 

Better GPU is always nice but not needed, that paired with more RAM would only mean I would crank my Civ 6 graphics settings higher (since it's the only game I play on the computer currently and since it was released (before that it was X-Com UE/EW and Civ 5 but when 32-bit support was dropped it killed X-com for me )).

 

BTW I've noted that the biggest hurdle to crank graphics in Civ 6 on my M1 is that I run out of RAM at to high settings  game itself runs good. 

[Paul Thexton]

2 hours ago, leadeater said:

I'd really like to see it happen, even if it requires a thick (slightly) boy Mac Mini to do it. I've always like the Mini because of how small it is but actually still being good.

I’ve always been a Mac mini fan, I won’t need one now if they do put a Pro/Max in it but I really hope they do, it would be a kick arse little machine. 

[BondiBlue]

3 hours ago, leadeater said:

I'd really like to see it happen, even if it requires a thick (slightly) boy Mac Mini to do it. I've always like the Mini because of how small it is but actually still being good.

The Mac mini really is an excellent little thing. I've got a Late 2009 Mac mini that runs 24/7 as a utility machine, and the only issue I've had with it was with the optical drive. Other than that it's been super reliable. Newer Mac minis are even better considering they have integrated power supplies now. 

[Video Beagle]

8 hours ago, Spindel said:

10 hours ago, leadeater said:

 

If we are still discussing of the M1 Pro/Max would fit into the current Mini case look at the pictures here with all the empty space: https://egpu.io/forums/desktop-computing/teardown-late-2020-mac-mini-apple-silicon-m1-thunderbolt-4-usb4-pcie-4/

 

It would be no problems fitting the Pro/Max variants into it.

what you're missing is that I was imagining these things to be 100x and more bigger than they are.