TSMC 3nm coming soon

Cmaier

Site Master
Staff Member
Site Donor
Posts
5,331
Reaction score
8,524
I wonder if Apple will be able to use this for the M2 Pro/Max/Ultra?

That has been one set of rumors. I always doubted those rumors, but as I reported on here some time ago, I later got reason to possibly believe them.
 

Cmaier

Site Master
Staff Member
Site Donor
Posts
5,331
Reaction score
8,524
Is this the process that is described using pairs of numbers, like 3-2 is a different power-draw-switching profile than, say, 2-1 so the actual "metal" configuration will at least partially determine P vs E designs?
If i remember correctly that will be the 2nm node, where they go to the wrap-around gates.
 

Cmaier

Site Master
Staff Member
Site Donor
Posts
5,331
Reaction score
8,524
Looks like the press agrees with me that M2 Pro/Max are going to be on 3nm.


I wasn’t the first one to suggest that, of course. But good to know that when I hear things they are sometimes true. :)
 

Cmaier

Site Master
Staff Member
Site Donor
Posts
5,331
Reaction score
8,524
Further to the above, i am predicting a clock speed increase. Around 4GHz on the performance cores.
 

Cmaier

Site Master
Staff Member
Site Donor
Posts
5,331
Reaction score
8,524
It looks like "FinFlex" is a 3N methodology whereas 2N will be using a "nanosheet" arrangement.

Thanks for tracking that down.

This isn’t really all that excitiong - before fins we used to customize the shape of every single transistor in order to provide the right balance between power and speed. This got harder with fins because, in a practical sense, you can’t muck around arbitrarily with the size/shape of the gated active area for these sorts of FINFETs, and instead muck around with the number of fins; you end up sort of quantizing things instead of allowing arbitrary configurations.

I don’t expect a customer like Apple to use these in the way some in the press assume they will. They are likely to mix and match different configurations even in the same ”circuit.“ At the very least they will have a library of slow, medium and fast standard cells that the designers can choose between on a gate by gate basis along each path.

At Exponential we had the same issue - the technology we were using had pre-designed transistors that you couldn’t arbitrarily muck with on-the-fly, so we picked different cells out of a library. Even had a tool called “Hoover” that would automatically pick the cells based on timing constraints (it “sucked the power” out of the circuit, hence Hoover).

Of course, the use of Hoover caused a very bad problem. They did it before I got there and got the chip back after I got there, and the problem took us a few days to figure out… but that’s a story for another day.
 

Yoused

up
Posts
5,624
Reaction score
8,943
Location
knee deep in the road apples of the 4 horsemen

Not sure about what N3 can offer, but Apple has been extremely conservative with clock speeds on their SoCs. I remember looking at the Mac Pro XeonW, where the model options where clock went down as core count went up, and most Anthill CPUs slow down as more cores are recruited. I suspect Apple wants to pick a baseline speed and hold to it howsovever many cores are pumping, up to nearing TDP limits. "Turbo-boost" is such BS, that Apple wants to avoid.
 

Cmaier

Site Master
Staff Member
Site Donor
Posts
5,331
Reaction score
8,524
Not sure about what N3 can offer, but Apple has been extremely conservative with clock speeds on their SoCs. I remember looking at the Mac Pro XeonW, where the model options where clock went down as core count went up, and most Anthill CPUs slow down as more cores are recruited. I suspect Apple wants to pick a baseline speed and hold to it howsovever many cores are pumping, up to nearing TDP limits. "Turbo-boost" is such BS, that Apple wants to avoid.
Yes, apple’s design will work fine with all cores at the top speed; that’s one of the advantages of being power efficient. You don’t have to worry about IR drop on the power rails, or on thermal hot spotting that bleeds over into neighboring cores.
 

DT

I am so Smart! S-M-R-T!
Posts
6,405
Reaction score
10,455
Location
Moe's
Main Camera
iPhone
Further to the above, i am predicting a clock speed increase. Around 4GHz on the performance cores.

M2 powered Studio machine should be pretty tasty :D I guess we'll see an M2 Max and M2 Ultra?
 

Colstan

Site Champ
Posts
822
Reaction score
1,124
Yep. Maybe even a 2xUltra for Mac Pro.
The people over at the other place keep insisting that it's going to be another Xeon Mac Pro, Apple wants to "keep pros happy", and because they paid more for their machines than the lesser Mac plebeians, they deserve at least 10 years of support, and continued parallel x86/Arm development.

I did some babysitting over at the Mac Pro forum recently. While demanding all of this from Apple, they also compare the company's management to infectious diseases, say that the NIH has to clean them out, and anyone who says otherwise is sharply denounced. In the same breath, they claimed that "this forum (MR) was instrumental in getting Apple to change its plans for pros", and they cite statistics that people from Cupertino visit the site as proof.

I don't understand how comparing Apple to the bubonic plague is going to convince anyone to release another Xeon Mac Pro, but that appears to be the common attitude. There's also this supreme sense of entitlement, comparing the MacBook Air to a "diaper for children", and the belief that because they spent more on the Mac Pro, they therefore deserve special treatment. I then quoted the various things you expect from the next Mac Pro, provided direct links to this site, and mentioned your credentials. They were not happy. I also invited them to come over here and ask you about your thoughts on the Mac Pro and Apple Silicon. Curious, but for some reason, nobody I chatted with has bothered to ask you, thus far.

This all started when the guy who originally leaked the Mac Studio also leaked that the upcoming Mac Pro used the same Apple Silicon that every other Mac uses, just scaled up, and the test board only had one PCIe slot. I'm not sure what stage of grief they are at, but right now it seems to be anger. I guess bargaining comes next. Acceptance either comes when they purchase the Apple Silicon Mac Pro, or punish Apple by switching to Linux or Windows workstations. At least they'll get their Xeons. I honestly can't wait until the next Mac Pro is announced. I'm going to lurk over there just to watch the implosion.
 

Colstan

Site Champ
Posts
822
Reaction score
1,124
In regards to the competition, according to wccftech, Intel is planning to launch "Extreme Performance" mode with Raptor Lake. It's designed for the i9-13900K on high-end Z790 motherboards with an unlimited power setting. Instead of the standard 125w PL1 boosting to 250w, this will allow the i9 to go all the way up to 350w. That's going to go along nicely next to the 600w RTX 4090 that has been rumored. At least you wouldn't need central heating during the winter months.

Keep in mind that the entire Mac Studio featuring an M1 Ultra has an absolute maximum power consumption of 215w for the entire computer.
 

Cmaier

Site Master
Staff Member
Site Donor
Posts
5,331
Reaction score
8,524
The people over at the other place keep insisting that it's going to be another Xeon Mac Pro, Apple wants to "keep pros happy", and because they paid more for their machines than the lesser Mac plebeians, they deserve at least 10 years of support, and continued parallel x86/Arm development.

I did some babysitting over at the Mac Pro forum recently. While demanding all of this from Apple, they also compare the company's management to infectious diseases, say that the NIH has to clean them out, and anyone who says otherwise is sharply denounced. In the same breath, they claimed that "this forum (MR) was instrumental in getting Apple to change its plans for pros", and they cite statistics that people from Cupertino visit the site as proof.

I don't understand how comparing Apple to the bubonic plague is going to convince anyone to release another Xeon Mac Pro, but that appears to be the common attitude. There's also this supreme sense of entitlement, comparing the MacBook Air to a "diaper for children", and the belief that because they spent more on the Mac Pro, they therefore deserve special treatment. I then quoted the various things you expect from the next Mac Pro, provided direct links to this site, and mentioned your credentials. They were not happy. I also invited them to come over here and ask you about your thoughts on the Mac Pro and Apple Silicon. Curious, but for some reason, nobody I chatted with has bothered to ask you, thus far.

This all started when the guy who originally leaked the Mac Studio also leaked that the upcoming Mac Pro used the same Apple Silicon that every other Mac uses, just scaled up, and the test board only had one PCIe slot. I'm not sure what stage of grief they are at, but right now it seems to be anger. I guess bargaining comes next. Acceptance either comes when they purchase the Apple Silicon Mac Pro, or punish Apple by switching to Linux or Windows workstations. At least they'll get their Xeons. I honestly can't wait until the next Mac Pro is announced. I'm going to lurk over there just to watch the implosion.

Apple cares more about the folks who buy a new $1500 computer, $1000 phone, and $500 watch every few years than someone who pays $10k for a Mac Pro and wants to use it for a decade.
 

Yoused

up
Posts
5,624
Reaction score
8,943
Location
knee deep in the road apples of the 4 horsemen
At N3, if stuff is really small, maybe they will not have to go "Extreme" (4xMax) but make Max so heavy with the (workhorse) GPU cores that just Ultra is enough. Or perhaps they have a new layout so slick that they can trim off the GPU block and fuse it onto another SoC, so that the entire thing does not have to be duplicated. Because, for heavy work, GPU is the tool, and for getting basic work done 3E+13P would put everything else away almost before you just got it out.
 
Top Bottom
1 2