Apple is an existential threat to the PC

[MfA]

Hows that? I mean ye if everyone's going apple devices and abandon AMD/NV/intel etc then we could see a future where everything will be apple-powered.

With likely very little room for Valve once Apple gets around to closing its Mac garden a little more.

That new M1Max is basically the same die size as the XSX/PS5 while offering much more cpu power and about the same GPU raw power at a fraction of the power draw.

It's effectively two nodes ahead of everyone else, TSMC is kingmaker. No 5nm AMD in sight this year.

 

[troyan]

M1 Max has 57 billion transistors. Yet going from Pro to Max brings only 256bit and 5.2 TFLOPs in exchange for additional 22.3 billion transistors. That sounds so inefficient from an architecture point. For example GA104 has 17.7 billion transistors and delivers ~19 TFLOPs within ~120W (RTX3080 mobile).

 

[snc]

so 32 gb ram, 10tf, 10 cores cpu in MacBook Pro 14 under 100w ? holly shit

 

[iroboto]

so 32 gb ram, 10tf, 10 cores cpu in MacBook Pro 14 under 100w ? holly shit

+400 GB/s memory to CPU. It's quite spacious

 

[Deleted member 13524]

The issue here is that, the world is moving further into a mobile space and ARM is increasingly becoming a better fit for mobile than CISC is. CISC will likely stick around because of performance, but they're going to be stuck in desktop workstations and server racks (even though technically IIRC all modern Intel and AMD CPUS are RISC), ARM will be the future to power those mobile devices, IOT, VR headsets, laptops, tablets, TVs, etc.

I thought all the modern x86 CPUs are essentially using RISC ALU blocks with tiny and very efficient instruction translation blocks, so whatever performance difference there is between modern ARM and x86 CPUs it's not inherent to its native instruction set.

M1 Max has 57 billion transistors. Yet going from Pro to Max brings only 256bit and 5.2 TFLOPs in exchange for additional 22.3 billion transistors. That sounds so inefficient from an architecture point. For example GA104 has 17.7 billion transistors and delivers ~19 TFLOPs within ~120W (RTX3080 mobile).

Performance per transistor has never been a very important metric.

Apple can afford to use a very low clocked and very wide GPU because they have a nearly exclusive access to TSMC's top-end fab technology and they charge at least $3000 for every laptop that carries the >400mm^2 5nm chip and 8 channels of super expensive LPDDR5.

 

[Lurkmass]

Apple silicon is not a threat to PCs at all because they strive to serve completely different models. Apple doesn't care about breaking software compatibility while PC is all about maintaining software compatibility ...

Consoles or Valve will never consider Apple as an option for the same reasons because games often like to statically link libraries or run low level kernel drivers sometimes which isn't very conducive to hardware portability ...

 

[iroboto]

I thought all the modern x86 CPUs are essentially using RISC ALU blocks with tiny and very efficient instruction translation blocks, so whatever performance difference there is between modern ARM and x86 CPUs it's not inherent to its native instruction set.

indeed, you are correct. My information is super dated. Decoding X86 to microcode is minimal increase in power, there is some, but not enough to warrant this power difference. We are likely looking at the difference in node here.

 

[pcchen]

M1 Max has 57 billion transistors. Yet going from Pro to Max brings only 256bit and 5.2 TFLOPs in exchange for additional 22.3 billion transistors. That sounds so inefficient from an architecture point. For example GA104 has 17.7 billion transistors and delivers ~19 TFLOPs within ~120W (RTX3080 mobile).

From their die photos, it looks like those surrounding the GPU cores are cache (probably SLC). If so, it's possible that M1 Max has twice the amount of SLC compared to M1 Pro, which may explain the large increase in number of transistors.

 

[Pressure]

From their die photos, it looks like those surrounding the GPU cores are cache (probably SLC). If so, it's possible that M1 Max has twice the amount of SLC compared to M1 Pro, which may explain the large increase in number of transistors.

That’s correct, twice the System Level Cache (64MB).

Also an extra media encode and decode block and larger display block.

 

[PSman1700]

What about Apple's history suggests to you they will ever bring down prices? Their M.O. is adding percieved value and charging through the nose for it.

Your correct. But who's going to be able to afford 5000 to 6000 dollar apple devices to get RTX3060-like performance though? Yes these new M1 devices are impressive for their power-draw, but the performance per dollar is quite bad. Being locked to macOS is the least the average user/gamer wants anyway. Theres 5800h/RTX3070/32gb asus laptops out there for 1500 dollars, sporting 144hz ips screens. For gaming, thats a much and much better choice, Overall performance isnt far from the highest end M1max either, if not better. yes higher powerdraw but it aint worth it to pay that much more. These apple laptops also lack ray tracing and other features consoles and pc gpus have, which are quite meaningfull going forward in the gaming world.
Like said, i think these new apple laptops are a very good fit for people in the professional market, like the macs have done so before. They just got better, doesnt mean apple sillicon will replace our pc's and consoles or even handhelds....

With likely very little room for Valve once Apple gets around to closing its Mac garden a little more.

Bound to happen. If that will lure gamers into apple's ecosystem no idea.

We are likely looking at the difference in node here.

Read that on many youtube comments and analysis touching up on that. Apple has the huge node advantage over everyone else, much of the performance comes from there for sure. Imagine if PS5/XSX had access this early to TSMC's 5nm production.

 

[Pressure]

Your correct. But who's going to be able to afford 5000 to 6000 dollar apple devices to get RTX3060-like performance though? Yes these new M1 devices are impressive for their power-draw, but the performance per dollar is quite bad. Being locked to macOS is the least the average user/gamer wants anyway. Theres 5800h/RTX3070/32gb asus laptops out there for 1500 dollars, sporting 144hz ips screens. For gaming, thats a much and much better choice, Overall performance isnt far from the highest end M1max either, if not better. yes higher powerdraw but it aint worth it to pay that much more. These apple laptops also lack ray tracing and other features consoles and pc gpus have, which are quite meaningfull going forward in the gaming world.
Like said, i think these new apple laptops are a very good fit for people in the professional market, like the macs have done so before. They just got better, doesnt mean apple sillicon will replace our pc's and consoles or even handhelds....

Other premium notebooks demand similar prices, like the Dell XPS and Microsoft Surface line. They are not gaming machines as such but capable of it.

Having ProRes 422 and ProRes RAW encode and decode in the media engine is going to be awesome as well as the 1600 nit peak brightness display. ProMotion is also really awesome.

Even Linus Tech Tips are impressed.

 

[PSman1700]

Other premium notebooks demand similar prices, like the Dell XPS and Microsoft Surface line. They are not gaming machines as such but capable of it.

The asus G15 with RTX3070 (130w max)/5800H/32gb (8 for the gpu)/144hz combo can be had for abit above 1500 dollars. You'd have to be spending what, atleast 4000 dollars to match that general performance with the M1 series? And then be limited with gaming. The upside would be powerdraw and overal build quality, but that doesnt warrant the ultra premium pricings.

Having ProRes 422 and ProRes RAW encode and decode in the media engine is going to be awesome as well as the 1600 nit peak brightness display.

Features most users will never use outside of pro-users. Besides that, GPU's these days are quite fast at those tasks aswell.

ProMotion is also really awesome.

So is a 144hz IPS asus display for most users. It might not offer the same quality and peak brightness but atleast its faster.

Even Linus Tech Tips are impressed.

Why would they not? Its a new Apple product, everyone's going to be impressed. Just as most reviewers are impressed with high performance windows notebooks.
Current Apple notebook lineup just isnt going to replace windows laptops in their current form end spec, besides running macOS. They are for the high-end creators and professional market (niche, as the price indicates).

Intel/AMD etc will probably get the smaller node advantages too in the future, diminishing apples lead probably.

 

[Pressure]

The asus G15 with RTX3070 (130w max)/5800H/32gb (8 for the gpu)/144hz combo can be had for abit above 1500 dollars. You'd have to be spending what, atleast 4000 dollars to match that general performance with the M1 series? And then be limited with gaming. The upside would be powerdraw and overal build quality, but that doesnt warrant the ultra premium pricings.

Features most users will never use outside of pro-users. Besides that, GPU's these days are quite fast at those tasks aswell.

So is a 144hz IPS asus display for most users. It might not offer the same quality and peak brightness but atleast its faster.



Why would they not? Its a new Apple product, everyone's going to be impressed. Just as most reviewers are impressed with high performance windows notebooks.
Current Apple notebook lineup just isnt going to replace windows laptops in their current form end spec, besides running macOS. They are for the high-end creators and professional market (niche, as the price indicates).

Intel/AMD etc will probably get the smaller node advantages too in the future, diminishing apples lead probably.

I'm not sure what you are trying to argue. These notebooks are specifically for content creators and no other solution support hardware encode and decode of ProRes. It's faster than the 28-core Mac Pro with the $2,000 AfterBurner card.

The graphic card alone uses much more power than the entire M1 Max and would probably not be competitive on battery power anyway. No other mobile solution offers that much frame buffer (up to 64GB) or 400GB/s bandwidth.

It isn't just the node advantage but Apple have been making low power devices for a decade now and are able to leverage that know-how. Arm is just more power efficient than x86.

The price is absolutely warranted but I get your point, you don't need it and can already find alternatives that better suit you needs, great.

 

[MfA]

Intel/AMD etc will probably get the smaller node advantages too in the future, diminishing apples lead probably.

Apple will likely have an understanding with TSMC to outbid for any 3nm capacity coming online for consumer PC products for at least 2 years (Qualcomm&co and Intel server products can get some space to make the monopoly dealings a little less obvious, they don't need the node advantage to push iphones).

AMD hopefully getting 5nm in 2024 and reducing the node disadvantage to a single node will be an improvement on the current situation, but the market is busted.

 

[iroboto]

AMD hopefully getting 5nm in 2024 and reducing the node disadvantage to a single node will be an improvement on the current situation, but the market is busted.

I thought Zen 4 was going to be on 5nm for 2022?

 

[MfA]

I sincerely doubt it, they might have a Zen 3 update with N6 but I get the impression N6 is more a PR exercise and not even as good as N7+.

 

[Deleted member 13524]

AMD hopefully getting 5nm in 2024 and reducing the node disadvantage to a single node will be an improvement on the current situation, but the market is busted.

Navi 31's compute dies are supposed to be N5 already, an those are coming in 2022.
Rembrandt is N6 in 2022, so I don't think it'll take another 2 years for the APUs to reach N5.

 

[Deleted member 11852]

I'm pretty sure, there will be a TB3/USB-C (DP alternate mode) to HDMI 2.1 adapter. With Apple logo for 70 bucks or so.

I wonder what HDMI 2.1 features the MBP need to support as an output device? As you say, there are three TB4 ports and non-Apple adaptors/dongles are inexpensive. It feels like DP and TB have eclipsed HDMI outside of the consumer space.

For what it's worth, we (work) have ordered two of the 14" 32 GPU core M1Max configurations. 32Gb RAM, 1Tb SSDs. As somebody who has not used an M1 Mac at all, I'm really interested to see how fast these things are.

 

[Jawed]

That’s correct, twice the System Level Cache (64MB).

Also an extra media encode and decode block and larger display block.

Do you have a link for the best, most-detailed, description of these new chips?

There's about 23B transistors extra in M1 Max (57B) versus M1 Pro (33.7B) and it seems there are two key differences:

1. GPU cores: 32 versus 16
2. Off-die memory system: 400GB/s versus 200GB/s

So how does Apple spend so many transistors? 23B transistors is equivalent to pretty much all the transistors in Navi 21 (26.8B). All of 6700XT is only 17.2B transistors.

 

[pcchen]

AnandTech has a brief article discussing some technical details:

https://www.anandtech.com/show/1701...m1-max-giant-new-socs-with-allout-performance

The SLC size is just a guess, but I think it's quite reasonable.