Peak Peek - Apple officially announces new SE, new iPad Air, M1 Ultra, Mac Studio and Studio Display

[AlTech]

20 hours ago, Roswell said:

 

 

VRAM doesn’t exist anymore, it’s been out of the picture for 20+ years since SDRAM took over.

 

If you’re going to argue semantics for no apparent reason you should at least have a rudimentary grasp on the subject you’re talking about.

 

Literally everybody refers to what you're calling SDRAM as VRAM including the GPU manufacturers.

 

I wasn't arguing semantics. I was arguing that combining RAM and VRAM doesn't mean you magically get everything turning into VRAM. I was also arguing if the current system ain't broke, why does Apple feel the need to fix it with Unified Memory.

 

VRAM being soldered is unfortunately nothing new and has existed since time immemorial. Soldering RAM in a computer is a big no no to me and some other people. Apple trying to normalize it is not okay IMO.

[igormp]

6 minutes ago, AluminiumTech said:

VRAM being soldered is unfortunately nothing new and has existed since time immemorial. Soldering RAM in a computer is a big no no to me and some other people. Apple trying to normalize it is not okay IMO.

Soldering ram brings benefits that you wouldn't be able to get otherwise, which IMO are acceptable for a laptop, such as lower power consumption, higher speeds and lower latency.

A PC with the RAM bus size of the M1 Ultra would need a shit ton of channels, which isn't doable even in full ATX form factors.

[saltycaramel]

13 minutes ago, AluminiumTech said:

Soldering RAM in a computer is a big no no to me and some other people. Apple trying to normalize it is not okay IMO.

 

The other day I was listening to a podcast (“Decoder” by Nilay Patel) where the creator of the Raspberry Pi mentioned the rise of appliance-PCs and fall of general purpose PCs. 

 

Well not even his Rpi, supposedly the champion of the latter kind of PCs, has non-soldered RAM…just like some of the most popular appliance-PCs (smartphones, game consoles, tablets, ultrabooks of many brands, and face-mounted PCs later this decade). The only thing Apple is normalizing nowadays is doing it on desktops..will be interesting to see what they’ll do on the Mac Pro. On mobile, that ship has sailed, not much left to normalize.

[WolframaticAlpha]

I would've loved to see a surface studio like touch screen monitor, into which you can dock in the Mac brick or the Mac mini

[WolframaticAlpha]

19 minutes ago, saltycaramel said:

 

The other day I was listening to a podcast (“Decoder” by Nilay Patel) where the creator of the Raspberry Pi mentioned the rise of appliance-PCs and fall of general purpose PCs. 

 

Well not even his Rpi, supposedly the champion of the latter kind of PCs, has non-soldered RAM…just like some of the most popular appliance-PCs (smartphones, game consoles, tablets, ultrabooks of many brands, and face-mounted PCs later this decade). The only thing Apple is normalizing nowadays is doing it on desktops..will be interesting to see what they’ll do on the Mac Pro. On mobile, that ship has sailed, not much left to normalize.

>Comparing 35 dollar device for iot to 4000 dollar machine for prosumers.

The studio will sell on the sole weight of the amazing M1 ultra. While the advantages of UMA are one of the major reasons for M1 platforms speed, for prosumers, expandability is a key concern. It will depend on the user for deciding whether they want the best perf or expandability.

 

What would be stupid is if the PC manufacturers normalize it, cuz they are using NUMA, and soldered RAM will have 0 benefits.

[Kilrah]

31 minutes ago, AluminiumTech said:

I was arguing that combining RAM and VRAM doesn't mean you magically get everything turning into VRAM.

Except it does, since firstly all the amount not used by the CPU is available to the GPU, and by the benefits of being soldered and having massive bus width actually has throughputs comparable to... well, VRAM or "the RAM on a normal GPU".

 

32 minutes ago, AluminiumTech said:

I was also arguing if the current system ain't broke, why does Apple feel the need to fix it with Unified Memory.

See above, it's just a league above in potential performance. Your entire system has RAM that's many times faster than anything existing. 

[AlTech]

45 minutes ago, igormp said:

such as lower power consumption, higher speeds and lower latency.

I'm aware. I just don't see that being an acceptable trade off, even for laptops. The ship has sailed on the phone and mobile tablet market but I disagree, it has not on laptop and it cannot be allowed to be normalized imo.

45 minutes ago, igormp said:

A PC with the RAM bus size of the M1 Ultra would need a shit ton of channels, which isn't doable even in full ATX form factors

[AlTech]

26 minutes ago, Kilrah said:

Except it does, since firstly all the amount not used by the CPU is available to the GPU, and by the benefits of being soldered and having massive bus width actually has throughputs comparable to... well, VRAM or "the RAM on a normal GPU".

Sounds like we need faster RAM standards, bigger memory buses on motherboards, and more VRAM on GPUs.

26 minutes ago, Kilrah said:

See above, it's just a league above in potential performance. Your entire system has RAM that's many times faster than anything existing. 

It's also a league above in e-waste creation, inflating product prices, reducing consumer choices and upgradability, as well as being generally anti-consumer.

[saltycaramel]

1 hour ago, WolframaticAlpha said:

>Comparing 35 dollar device for iot to 4000 dollar machine for prosumers.

The studio will sell on the sole weight of the amazing M1 ultra. While the advantages of UMA are one of the major reasons for M1 platforms speed, for prosumers, expandability is a key concern. It will depend on the user for deciding whether they want the best perf or expandability.

 

What would be stupid is if the PC manufacturers normalize it, cuz they are using NUMA, and soldered RAM will have 0 benefits.

 

I think the demise of user-serviceable parts consists of appliance-PCs being adopted by some users (and use cases) that would have shopped for a fully serviceable PC before. No matter the price or professional nature. 

So no one will “come for” your removable RAM, it’s just that more and more users will realize an appliance-PC is adequate for their needs. Some users will still need trucks and swappable RAM, no doubt. Even Apple may admit this, we’ll know when the MacPro will be revealed. 

[leadeater]

4 hours ago, LAwLz said:

I'm sorry Leadeater, but the Firestorm cores really does wipe the floor with Zen3, and does it at way lower power consumption.

Well other than single core I disagree and gave you the data that proves it, you can either believe the data and facts or not 

 

Being slightly better, at the benefit of using TSMC 5nm is not such a massive win. It's a win but it's not wiping the floor with anything bar that one thing. At least just talking pure CPU performance.

 

And if we want to compare kernel build times then use https://openbenchmarking.org/test/pts/build-linux-kernel, like I did.

[leadeater]

6 hours ago, Paul Thexton said:

Depends entirely on what modules you’re compiling as well of course. In this case I believe the thread ripper was cross compiling the kernel for his M1 Linux work.

Yep precisely why I go to something more trusted than twitter, like the above mentioned https://openbenchmarking.org/test/pts/build-linux-kernel&eval=2e091b8032ec35a1fdb1b678beec9531ea9c956b#metrics. Because the build configurations are checked and there are multiple submissions so I can be more confident of those that just singular claims. Downside, no M1 data from what I can see.

[igormp]

34 minutes ago, AluminiumTech said:

I'm aware. I just don't see that being an acceptable trade off, even for laptops. The ship has sailed on the phone and mobile tablet market but I disagree, it has not on laptop and it cannot be allowed to be normalized imo.

Well, you may not, nor maybe big companies with large IT depts (which usually like to repair their own machines), but most consumers don't care, and neither do I since we do get some benefits tbh.

Your opinion might just not be that relevant to most companies anymore (but do not stop complaining, otherwise the ship will sail for sure and having alternatives is always good).

 

29 minutes ago, AluminiumTech said:

Sounds like we need faster RAM standards, bigger memory buses on motherboards, and more VRAM on GPUs.

Sure! Then you apply those to soldered stuff and it'll still be faster, you can't just go against physics in this case.

 

1 minute ago, leadeater said:

Downside, no M1 data from what I can see.

The phoronix test suite isn't really ready to cross compile a kernel, a test done on the M1 would need some changes that would invalidate the tests. Anyhow, I'd still trust the tweets from 2 of really valuable kernel developers (Marcan and Axobe), which even Phoronix (the same guy who created openbenchmarking) has acknowledge before:

https://www.phoronix.com/scan.php?page=news_item&px=CPUFreq-Apple-M1-WIP 

[leadeater]

3 minutes ago, igormp said:

The phoronix test suite isn't really ready to cross compile a kernel, a test done on the M1 would need some changes that would invalidate the tests. Anyhow, I'd still trust the tweets from 2 of really valuable kernel developers (Marcan and Axobe), which even Phoronix (the same guy who created openbenchmarking) has acknowledge before:

https://www.phoronix.com/scan.php?page=news_item&px=CPUFreq-Apple-M1-WIP 

True, it's not like something like that is completely untrustworthy but it requires more checking to make sure. But anyone wanting to die on the hill of Linux kernel build times to prove the CPU performance alone is better by huge amounts always in every way is simply going to need a good reality check slap 

 

More so if comparing using a 1950X lol.

[igormp]

Just now, leadeater said:

True, it's not like something like that is completely untrustworthy but it requires more checking to make sure. But anyone wanting to die on the hill of Linux kernel build times to prove the CPU performance alone is better by huge amounts always in every way is simply going to need a good reality check slap 

Sure it's not, but no benchmark is valid for all use cases. Nonetheless, this benchmark is really useful for developers, since it was well shown how the M1 is fast when it comes to compile, be it with the kernel, Go stuff or even web dev projects in JS.

It may not even be due to the CPU alone, but rather to the ram bandwidth, but it's still an indicative of overall performance for this general use case and it seems to be true for many dev cases, which is what matters for the and user anyway.

 

If I weren't so deeply entangled with my linux workflow, I'd go for a M1 Mac without a second thought, no other laptop offers such good battery life with that kind of performance (specially when you want a non-gamery-full-of-rgb model).

[leadeater]

1 hour ago, AluminiumTech said:

Sounds like we need faster RAM standards, bigger memory buses on motherboards, and more VRAM on GPUs.

More VRAM on GPUs will not resolve the performance benefits of what Apple is doing, no matter how much you put on or double the VRAM bandwidth it being isolated to an add-in board across a remote PCIe bus means there will always be significant trade-off and different approaches required compared to a unified memory archecture both logically and physically.

 

RAM standards aren't the issue here, it's electrical design constraints and the practical problems of putting RAM on DIMMs. A motherboard PCB is never going to reach the same levels of efficiency as an interposer with memory dies. So while you could grab an 8 channel AMD EPYC and use it as a desktop resulting in similar CPU memory bandwidth you'll never get that in a Mac Mini, MacBook Pro or anything smaller than SSI CEB.

[LAwLz]

54 minutes ago, leadeater said:

Well other than single core I disagree and gave you the data that proves it, you can either believe the data and facts or not 

lol, so passive aggressive.

That smiley doesn't exactly hide that you are mad.

 

 

I highly question the results from Hardware Unboxed you linked earlier.

The reason why I question it is because they say the M1 Pro uses 40 watts of "package power". Meanwhile, Anandtech measured the package power of the M1 Max to be 34 watts in Cinebench MT.

I find it odd that a scale down part with the same clock speed would end up using more power. My guess is that the M1 Pro in the Hardware Unboxed video is more along the lines of 30 watts, and that's including things such as the RAM, which I don't think Hardware Unboxed included in the Ryzen "package power" number.

I couldn't find a good explanation of how Hardware Unboxed measured the package power. I don't watch their videos so maybe they explain it somewhere. If they do, please link it to me as I am interested.

Apple's ~20-25 watt CPU (the CPU alone, not CPU+memory) beating AMD's 35 watt CPU is more along the lines of what I have seen elsewhere and what I expect.

 

 

  

36 minutes ago, leadeater said:

True, it's not like something like that is completely untrustworthy but it requires more checking to make sure. But anyone wanting to die on the hill of Linux kernel build times to prove the CPU performance alone is better by huge amounts always in every way is simply going to need a good reality check slap 

 

More so if comparing using a 1950X lol.

Linux developer: Wow, the M1 is really good at compiling this Linux kernel. It's better than my old threadripper.

You: He should get a new threadripper if he wants to compile Linux kernels. It is much faster than the M1 at compiling! Look at these benchmarks!

Someone else: You're reading those benchmarks incorrectly.

You: it's not like Linux kernel compilation times matter anyway... 

[leadeater]

17 minutes ago, LAwLz said:

I highly question the results from Hardware Unboxed you linked earlier.

Well I didn't just give HUB

 

17 minutes ago, LAwLz said:

The reason why I question it is because they say the M1 Pro uses 40 watts of "package power". Meanwhile, Anandtech measured the package power of the M1 Max to be 34 watts in Cinebench MT.

I think some of there graphs are labelled using "TDP class", not actual power. There were also issues measuring M1 package power etc, I know Anand got hit with that and corrected it at some point I believe.

 

Also yes M1(s) package power does include RAM while a Ryzen would not. So both using ~35W the M1 would be at a complete maximum theoretical more efficient of 30% as that is the power efficiency difference between TSMC 5nm and 7nm (normalizing as many other factors as possible). I doubt it's 30% better but it's better none the less. I'm not going to argue that something on TSMC 5nm is not going to be more efficient than something on 7nm however I will point to actual product comparisons

 

17 minutes ago, LAwLz said:

lol, so passive aggressive.

Yes well you'll find hypism of M1 not in the face of reality or properly defined I will treat with extreme distain. Because there is A LOT of truth stretching, selectism, and just outright BS involved with it sadly.

 

If nobody is willing to actually look at power equivalent parts that exist then there can be no credibility, simple as that. There is no nice way to put it. U, H, HS, HX Ryzen mobile CPUs all exist and they all have very good performance, the desktop parts are by no means power efficient or even trying to be compared to the mobile parts.

[leadeater]

25 minutes ago, LAwLz said:

Linux developer: Wow, the M1 is really good at compiling this Linux kernel. It's better than my old threadripper.

You: He should get a new threadripper if he wants to compile Linux kernels. It is much faster than the M1 at compiling! Look at these benchmarks!

Someone else: You're reading those benchmarks incorrectly.

You: it's not like Linux kernel compilation times matter anyway... 

I was not knocking the developer, I was knocking the usage of that tweet. He can do whatever he wants, what people do with his information is another matter

 

25 minutes ago, LAwLz said:

Someone else: You're reading those benchmarks incorrectly.

Exactly where and how? I think you'll find I went to a more trusted source and looked at relative performance between many CPUs before commenting. And cross checked with with Phoroinx benchmark from their review, 2 second difference (45s vs 47s).

 

Good data/information can, and will, be badly.

[LAwLz]

Just now, leadeater said:

Well I didn't just give HUB

Speaking of the Geekbench results you posted, did you cherry pick those?

I just did a search myself on Geekbench and it seems like you picked the highest scoring G713, one which scored way higher than the average score of around 1500/8200 (SC/MC). I was able to find several M1 Max results with higher scores than the one you posted though.

Did you deliberately pick the highest scoring 6900HS and avoided the highest scoring M1 Max or was that unintentional?

 

Here are 4 out of the 6 Strix G713RM runs on Geekbench:
1536 - 8243

1483 - 7823

1533 - 8244

1468 - 7922

 

Here are the two remaining scores:

1644 - 10609

1634 - 10641 - The one you chose 

 

 

It's the same story with the G14 you linked. The one you linked has way higher score than most other results submitted.

Most submissions for the G14 with the 6900HS hovers around 1580 SC and 9500 MC. The one you posted is at 1611 MC and 10019 MC.

[leadeater]

3 minutes ago, LAwLz said:

Speaking of the Geekbench results you posted, did you cherry pick those?

I did so for all of them, GB has a page/feature that can sort all the results and I chose the highest from those for all. Can't remember if I sorted by ST or MT.

[igormp]

14 minutes ago, leadeater said:

U, H, HS, HX Ryzen mobile CPUs all exist and they all have very good performance

And yet they're still lagging behind the M1. And when AMD gets their 5nm parts, apple will already be on 3nm, so always a generation ahead.

[leadeater]

7 minutes ago, igormp said:

And yet they're still lagging behind the M1. And when AMD gets their 5nm parts, apple will already be on 3nm, so always a generation ahead.

Yes that is true, I'm not saying the M1's aren't better. It's the by how much and when, and more so how one frames it. I choose to address bluster, people are free to ignore that if they so choose. Being better is great, acting like it's the second coming of CPUs here to liberate us yea.. not so much.

 

Edit:

FYI I remember very strongly the BS about Zen 3 being hundreds of percent more power draw than M1. Those people know who they are. Those people still continue to post in their own reality to this day.

[LAwLz]

12 minutes ago, leadeater said:

I doubt it's 30% better but it's better none the less. I'm not going to argue that something on TSMC 5nm is going to be more efficient than something on 7nm however I will point to actual product comparisons

Small correction, the 6900HS is on TSMC N6. I am not sure how big the differences between N5, N6 and N7 are though. 

 

10 minutes ago, leadeater said:

Exactly where and how? I think you'll find I went to a more trusted source and looked at relative performance between many CPUs before commenting. And cross checked with with Phoroinx benchmark from their review, 2 second difference (45s vs 47s).

A developer posted that building his kernel took 4 minutes and 22 seconds.

You replied by saying the 5950X does it in 47 seconds, even though they were not comparing the same build.

 

If I say:
"Wow my RTX 3060 Ti is really good. It runs Cyberpunk at 60 FPS"

then you can't respond with:

"You should get an AMD GPU instead. It runs games at 80 FPS" and then say you were correct because it might get 80 FPS in Minecraft or whatever.

 

Sure, you're correct that it gets 80 FPS in Minecraft, but that is not exactly "data and facts" that prove that AMD GPUs are better than Nvidia. You were comparing apples and oranges (deliberately or by mistake) said it as if they were the same.

[igormp]

Just now, LAwLz said:

Small correction, the 6900HS is on TSMC N6. I am not sure how big the differences between N5, N6 and N7 are though. 

N6 is more of an updated N7, so a bit far from N5 in the end.

 

3 minutes ago, leadeater said:

Being better is great, acting like it's the second coming of CPUs here to liberate us yea.. not so much.

Sure it's not if you only take into account the sheer performance and compare it with existing players. But it is amazing when you consider the advancements that are being done with such devices, much like what happened with AMD CPUs going from Zen through Zen 3, even though Intel wasn't that far behind.

 

Anything that does such a leapfrog advancement will get such praise, we should only hope to see more of that happening (and in more open platforms too, I really don't like MacOS).

[leadeater]

2 minutes ago, LAwLz said:

Small correction, the 6900HS is on TSMC N6. I am not sure how big the differences between N5, N6 and N7 are though.

Good point, also arch improvements on there too. 6000 series is a nice bump from the 5000 series. It is a little messy, such is life with these things.