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

[saltycaramel]

Flash storage on Apple Silicon Macs is part of the experience and I think it will get exponential improvements in the coming years.

 

Think of it as “L5 cache”. (it’s not PCIE-based btw)

 

If one asks for an M.2 slot and the ability to install a cheap 3rd party SSD in an Apple Silicon Mac, one doesn’t understand what Apple Silicon Macs are about.

[hishnash]

1 hour ago, StDragon said:

BS. The SSD is capped at PCIe 3.0 speeds at best. SED drives are just as fast on a PC.

Im talking about perfomance given the power budget.
Using a PCIe based SSD would draw a LOT more power even if it were an NVMe solution that was just PCIe3. Do you want your SSD to draw more power that multiple Cpu cores under full load? 

 

 

1 hour ago, HenrySalayne said:

If Apple could outpace the 3080 (mobile) "massively" they should just start building graphics cards.

They will not build graphic cards (and even if they did they would be useless for windows systems depending on DX and VK as apples gpu IP is tightly linked to Metal features.. see some of the blogs from the teams working on getting linux on M1).    What apple will do (for the macPro) is sell additional (optional) add in MPX cards that have full SoCs on board to use up the SoC dies that have defective CPU cores but fully working GPU cores. Key is these will be optional as the SoC will be the same (just will working CPU cores as well). 

 

 

1 hour ago, Blademaster91 said:

unless they were forced to make laptops easier to repair,

The right to repair movement does not require you to make things pocketable, all they  want and all they need is the schematics for diagnostics the software to re-pair and the parts to replace. The BGA soldering needed to replace SSD dies is not that difficult and any board level repair location can do this with ease.  In fact making things modular might be worse for repair as it requires that these modules are manufactured and provided, apple currently uses off the shelf NAND dies these are easy to get hold of, all apple really needs to do (from the SSD right to repair perspective) is provide specs on what SSD dies are compatible with their SSD controller (they already provide software for re-setting the controller using the DFU mode).  There are benefits to using the on die SSD controller vs a PCIe attached option version. 

 

 

[hishnash]

1 hour ago, HenrySalayne said:

How? They have to fit the entire SoC including CPU and GPU on a similar or even smaller die compared to the 3080. The 3080 (GA104) has almost 400 mm², the current M1 has just 120 mm². If Apple could outpace the 3080 (mobile) "massively" they should just start building graphics cards.

To be clear im saying in professional tasks were a mobile 3080 will end up memory limited (due to having at most 16GB of memory) and having a high latency (low bandwidth) connection to the cpu memory. The 32Core Gpu in the 16" MBP will have full bandwidth (low latency) connection to the full 32GB or even 64GB memory of the system and for pro tasks that are mixed CPU/GPU (NPU etc) the fact that the GPU and CPU can share memory in the same way 2 CPU cores can share memory (eg point to the exact same location and modify and read bytes from that locations) can result in a massive performance boost for mixed compute tasks. 

Games are a different story, a well optimised game engine for apples GPU arc could well also run better on this Gpu due to the much better culling supported by TBDR GPUs. In most games 50%+ of the work your GPU is doing ends up being rendering pixels that are not visible due to other objects obscuring them. TBDR gpus (due to doing the blending in pipeline) are much better at skipping the rendering of obscured fragments this really makes a big differences to power draw, memory bandwidth but also total frame speed. But to take advantage of this fully you really need to massively re-write/think your render pipeline so very few (if any) games take this much effort when porting to apples platforms. 
 

The are other things game devs (like e-sports) could do on apples platforms that could reduce input latency massively (the shard memory interface between CPU and GPU is just one of the reasonse for this. But it would require a large! very large! re-write of the core engine.  
 

So i think the beating the 3080 will really be something we see in professional workloads (that have some optimisation for apples platforms) rather than Games (that never really get any attention) 

[Arika]

I feel like there's an Apple event every alternate month.

[StDragon]

1 hour ago, Arika S said:

I feel like there's an Apple event every alternate month.

For me, it's every time there's an iOS update 15.0.2 plugs another zero day. It's a doozie.

[WolframaticAlpha]

I can only dream of airpods and apple watch working with my android(I know airpods can bluetooth but still). 

[Blademaster91]

6 hours ago, BondiBlue said:

The 2015 MacBook Pro has a removable SSD, but it absolutely does not have the butterfly keyboard. The 2017 13" non-TB MacBook Pro is actually the last to have a removable SSD. 

Ok I must be thinking of the 2016 macbook, I also didn't see any teardowns for either the 2015 macbook pro with a removable SSD, or the 2017 non-touchbar version on iFixit. But still years since Apple offered a easily serviceable macbook, and I don't expect Apple to offer one, as soldered in everything maximizes their profit, and keeps customers buying another laptop instead of swapping in a new SSD.

5 hours ago, hishnash said:

Im talking about perfomance given the power budget.
Using a PCIe based SSD would draw a LOT more power even if it were an NVMe solution that was just PCIe3. Do you want your SSD to draw more power that multiple Cpu cores under full load?

Thats a fine tradeoff IMO over everything soldered in, and a Samsung 980 pci-e 4.0 ssd consumes about 3.5W at load, I personally greatly prefer the convenience of being able to buy a cheaper laptop then upgrade later if i need more storage.

5 hours ago, hishnash said:

The right to repair movement does not require you to make things pocketable, all they  want and all they need is the schematics for diagnostics the software to re-pair and the parts to replace. The BGA soldering needed to replace SSD dies is not that difficult and any board level repair location can do this with ease.  In fact making things modular might be worse for repair as it requires that these modules are manufactured and provided, apple currently uses off the shelf NAND dies these are easy to get hold of, all apple really needs to do (from the SSD right to repair perspective) is provide specs on what SSD dies are compatible with their SSD controller (they already provide software for re-setting the controller using the DFU mode).  There are benefits to using the on die SSD controller vs a PCIe attached option version. 

Not true, right to repair includes independent repair, which means anyone can obtain parts, and ideally companies wouldn't pull crap like serializing everything to the motherboard because offering parts for sale would be completely useless. The SSD being BGA soldered to the motherboard makes repairs or replacement significantly more costly for the consumer, and good luck getting anyone to swap the NAND chips on a macbook. The only one that benefits from using BGA soldered SSD and RAM is the manufacturer.

[suicidalfranco]

Going to be hard for apple to show something better than a framework

[AldiPrayogi]

I'm not well versed with Apple stuff, but wasn't there an Apple event already this year? Or was that only on iPhones and mobile stuffs? Or is it normal for Apple to have two separate events in a year?

[Lightwreather]

50 minutes ago, AldiPrayogi said:

I'm not well versed with Apple stuff, but wasn't there an Apple event already this year? Or was that only on iPhones and mobile stuffs? Or is it normal for Apple to have two separate events in a year?

2 events yes... But it is usual for them to have it one month apart. But they broke the mold last year too, so.....

[saltycaramel]

1 hour ago, AldiPrayogi said:

I'm not well versed with Apple stuff, but wasn't there an Apple event already this year? Or was that only on iPhones and mobile stuffs? Or is it normal for Apple to have two separate events in a year?

 

The normal number of events would be 1 software event (WWDC) and 1-3 hardware events per year. Nowadays I wouldn’t rule out even a Services (AppleTV+, gaming, etc.) event for next month.

 

On top of that, the pandemic made it more viable to make shorter fast paced pre-recorded virtual events instead of a bigger in-person live event. Last year the watch and the iPhone had separate (virtual) events: that would never have happened in the “before” days. Once events became pre-recorded videos Tim Cook can one-click publish any day any time, there’s less incentive to make a big 2 hours event filled to the gills with as many products as possible. Just make smaller events and maximize press coverage and customer attention span.

[hishnash]

3 hours ago, Blademaster91 said:

Not true, right to repair includes independent repair, which means anyone can obtain parts, and ideally companies wouldn't pull crap like serializing everything to the motherboard because offering parts for sale would be completely useless.

I was just talking about the SSD NAND apple does not serializing these, people have upgraded these in the past on T2 macs, M1 macs and iPhones (these all use the same SSD controller embedded within the SOC).
 

3 hours ago, Blademaster91 said:

The SSD being BGA soldered to the motherboard makes repairs or replacement significantly more costly for the consumer, and good luck getting anyone to swap the NAND chips on a macboo

Any board level repair shop can do this, the difficult part for them is finding out what spec of SSD dies apple is using. the BGA grid used on SSD NAND is not that dense and if you are used to replacing other BGA parts (like anyone doing board level repair will need to do all the time to fix broken charging systems) this is not a hard task.  As I said with respect to the SSD repair all apple really needs to do (that they currently are not doing) is provide the spec for the SSD dies so that people can with confidence go out and source them (you don't need the exact model numbers just the protocol version, voltages etc).  Since when you replace them you do not end up throwing away the costly SSD controller (as you would with a NVMe upgrade) depending on the cost of labor the upgrade might not cost much more than buying a high speed NVMe drive.  

Since there is no slow diagnostic work involved these will be much cheaper labour wise compared to regular board level repair tasks. 

remember replacing an entire NVMe driver just to throw out the costly controller and attached ram is not great for e-wast either. What would be much better is if apple (and others) make it clear that the upgrade/repair path through indecent repair for Memory, SSD etc was possible and helped these repair shops with the specs they needed. (for memory and NAND dies apple does not need to supply these the reapir shops can get these from regular supply chain).  

[hishnash]

3 hours ago, suicidalfranco said:

Going to be hard for apple to show something better than a framework

In perfomance it will be easy, the M1 runs circles around the framework laptop on every spec point (even price). 

[leadeater]

16 hours ago, Commodus said:

This is the big test: we should find out how well Apple Silicon scales to pro-level machines.

I still have a feeling Apple isn't going to release their biggest planned chips just yet, like the big mac Pro replacement ones. So there's still a good chance there will be even higher end stuff coming later as well.

[leadeater]

28 minutes ago, hishnash said:

In perfomance it will be easy, the M1 runs circles around the framework laptop on every spec point (even price). 

MacBook Air man, best value thing out, if you need any amount of performance at all. Personally I couldn't stand to use anything ultra cheap that may or may not technically be better value for money but the Air boy that is a solid choice for "just a portable computer".

[leadeater]

11 hours ago, saltycaramel said:

Flash storage on Apple Silicon Macs is part of the experience and I think it will get exponential improvements in the coming years.

 

Think of it as “L5 cache”. (it’s not PCIE-based btw)

If you were talking about memory you'd have a point but Apple isn't doing anything special with SSD/Flash storage at all, it's not caching any CPU or GPU data in that way at all.

 

Also the SSD is NVMe and PCIe based btw.

[saltycaramel]

It’s not literally L5 Cache but I believe its performances are so intertwined with the M1 experience and so destined to be on the same yearly upgrade trajectory (at least the speed of the controller and the interconnect, i.e. the parts Apple is directly responsible for) as the SoC that asking for it to be removable would be akin to asking for L3 cache to be user removable.

 

I’ll double check the pcie/nvme thing. 

[saltycaramel]

We didn’t get to 1TB iPhones editing without a hiccup multi-hundred-gigabyte ProRes 4K video files by using removable storage.

 

Same applies to Apple Silicon Macs.

[Dracarris]

2 hours ago, hishnash said:

In perfomance it will be easy, the M1 runs circles around the framework laptop on every spec point (even price). 

Not only performance, but also build quality, screen, trackpad, keyboard, speakers.. Framework is a great machine, but the Windows world is still playing catch-up to the current Macbooks (including the last generation of Intel-based ones), with the recent XPS coming close (and very similarly priced).

[leadeater]

1 minute ago, saltycaramel said:

It’s not literally L5 Cache but I believe its performances are so intertwined with the M1 experience and so destined to be on the same yearly upgrade trajectory (at least the speed of the controller and the interconnect, i.e. the parts Apple is directly responsible for) as the SoC that asking for it to be removable would be akin to asking for L3 cache to be user removable.

 

I’ll double check the pcie/nvme thing. 

The M1 SoC has inbuilt NVMe controller.

 

Speed wise it's just as fast as any other top tier SSD, point however is that it's not being used in any special way. Next generation Apple silicon might start to use the SSD as part of the memory subsystem, we're starting to see industry standards coming in to adoption that have that feature capability in them and it makes a lot of sense for Apples smaller devices like the Air, Mini and smaller MacBook Pros to go this path rather than expensive LPDDR that might not actually be needed.

[leadeater]

5 minutes ago, saltycaramel said:

We didn’t get to 1TB iPhones editing without a hiccup multi-hundred-gigabyte ProRes 4K video files by using removable storage.

 

Same applies to Apple Silicon Macs.

Because Apple implemented NVMe in their phones as well, that's what gave them that speed increase. You can do it with removable storage but only if it's Thunderbolt PCIe NVMe external storage.

 

Quote

Furthermore, the report revealed that the iPhone 6s uses NVMe protocol with custom NAND memory which ruled out UFS or traditional eMMC. The iPad Pro and iPhone SE uses the same type of storage as the 2015 iPhones.

 

Quote

Sure, some will point out that Apple uses its proprietary controller technology with PCI-E NVMe protocols for its iPhone memory, while Samsung uses the AHCI protocol for its UFS V-NAND memory. It’s all sounds too complicated with all the acronyms, but so far Apple’s solution is the faster one according to benchmarks

https://bgr.com/tech/iphone-vs-galaxy-note-storage-speed/

[hishnash]

11 minutes ago, leadeater said:

Speed wise it's just as fast as any other top tier SSD, point however is that it's not being used in any special way.

Thats not quite true, since the SSD controller users the same shared memory as the rest of the system cache backed writes (and reads) are no-op operations were the page pointer already in memory is just returned. (compared to a remote PCIe SSD controller were when you do a read/write it even if that data is in ssd local memory it needs to copy back to the system memory) this does have a miserable impact in random IOPS not in sustained loads. Given macOS uses APSF most operations (even file copies) are not large date copy operations but small meta data mutations so IOP latency is imports, while you could get the bandwidth with PCIe3/4 NVMe ssds getting that latency on cached IOPS would not be possible and even just getting the same perfomance would draw quite a bit power power as the SSD controller on your NVMe stick will not be on TSMCs latest 5nm node and there is the extra overhead of PCIe.

 

The efforts to get linux on M1 have shorn that this controller is not attached over PCIe as you might expect but rather attached to the main bus in the same was as the GPU/NPU/CPU etc it is just yet another (one of at least 12) co-prososoers within the SOC. 

 

[Roswell]

14 hours ago, Commodus said:

That I wouldn't count on, unfortunately, although the dropping costs of SSDs at least increase the chances that you'll get the capacity you want out of the box. I'm not counting on Apple starting at 512GB, but I wouldn't rule it out, either.

They already start at 512 on the 16. The rumors also say the 14 should be identical in all but size and battery. I suppose we’ll see.

[leadeater]

14 minutes ago, hishnash said:

Thats not quite true, since the SSD controller users the same shared memory as the rest of the system cache backed writes (and reads) are no-op operations were the page pointer already in memory is just returned. (compared to a remote PCIe SSD controller were when you do a read/write it even if that data is in ssd local memory it needs to copy back to the system memory) this does have a miserable impact in random IOPS not in sustained loads. Given macOS uses APSF most operations (even file copies) are not large date copy operations but small meta data mutations so IOP latency is imports, while you could get the bandwidth with PCIe3/4 NVMe ssds getting that latency on cached IOPS would not be possible and even just getting the same perfomance would draw quite a bit power power as the SSD controller on your NVMe stick will not be on TSMCs latest 5nm node and there is the extra overhead of PCIe.

 

The efforts to get linux on M1 have shorn that this controller is not attached over PCIe as you might expect but rather attached to the main bus in the same was as the GPU/NPU/CPU etc it is just yet another (one of at least 12) co-prososoers within the SOC. 

There is a decent benefit to having the NVMe controller within the SoC however it using the system memory as it's DRAM like a regular standalone NVMe SSD would doesn't mean it's actually faster within the OS or application. I've not seen any benchmarks of any M1 device showing any better performance than that of other top tier SSDs, however to be fair there is also a lack of good testing as all the tests you can find on M1 devices are far too basic.

 

It's still an NVMe protocol call with it's memory page living in a physically shared DRAM, it's still far as I know the same access method and data copy path just done over single SoC interconnects, it's not able to skip any memory copy operations.

 

M1 Device

http://blog.greggant.com/posts/2021/03/25/amorphousdiskmark-is-crystaldiskmark-for-mac.html

 

Samsung 980 Pro

 

If you can show me a better benchmark of the M1 storage in small I/O sizes I'll happily compare to better tests on the Windows/Linux side with IOmeter or equivalent. 

 

P.S. 3rd row cannot be compared, different tests.

[Video Beagle]

16 hours ago, Dracarris said:

Or, just maybe, do it properly this time and label 4 points on the axis and highlight some points where there is a lot of improvement.  This takes maybe 10s and won't halt anything, yet can be analyzed in more detail later by people that are interested.

 

There won't be, so might as well get your complaining about it done this week. This is not a presentation for tech geeks. It's not an engineering presentation to board partners like AMD does. This is a big shiny marketing show aimed at Wallstreet and the mainstream. They'll get press for the presentation, they'll get press with geeks whining about lack of details, they'll get press when LTT and Snazzy and Marcus and whoever put out tests.

 

2 hours ago, leadeater said:

I still have a feeling Apple isn't going to release their biggest planned chips just yet, like the big mac Pro replacement ones. So there's still a good chance there will be even higher end stuff coming later as well.

Very likely. We'll have a better idea of how things look when we learn if this will be called the M1X or the M2.