marcan changed the topic of #asahi to: Asahi Linux: porting Linux to Apple Silicon macs | General project discussion | GitHub: https://alx.sh/g | Wiki: https://alx.sh/w | Topics: #asahi-dev #asahi-re #asahi-gpu #asahi-stream #asahi-offtopic | Keep things on topic | Logs: https://alx.sh/l/asahi
alexstor_ has joined #asahi
alexstore06 has quit [Ping timeout: 480 seconds]
Gues__________________________ has joined #asahi
Gues__________________________ is now known as name327522
capy has joined #asahi
yuyichao has joined #asahi
capy has quit [Quit: WeeChat 3.3]
alexstor_ has quit [Remote host closed the connection]
eichin has quit [Remote host closed the connection]
mustache_ride_ has quit [Read error: Connection reset by peer]
enomem has quit [Read error: Connection reset by peer]
NekomimiScience has quit [Write error: connection closed]
jabashque has quit [Read error: Connection reset by peer]
saintdev has quit [Read error: Connection reset by peer]
guan has quit [Remote host closed the connection]
aeroraptor has quit [Remote host closed the connection]
tardyp has quit [Write error: connection closed]
eichin has joined #asahi
mustache_ride_ has joined #asahi
NekomimiScience has joined #asahi
aeroraptor has joined #asahi
guan has joined #asahi
tardyp has joined #asahi
jabashque has joined #asahi
saintdev has joined #asahi
enomem has joined #asahi
jabashque has quit [Ping timeout: 480 seconds]
aeroraptor has quit [Ping timeout: 480 seconds]
saintdev has quit [Ping timeout: 480 seconds]
guan has quit [Ping timeout: 480 seconds]
enomem has quit [Ping timeout: 480 seconds]
mustache_ride_ has quit [Ping timeout: 480 seconds]
tardyp has quit [Ping timeout: 480 seconds]
<chadmed>
citizen1[m]: apple's not really doing anything special. they have ~10 years of designing architectures that maximise perf/W as much as humanly possible, they have virtually infinite money and the desire to design a high performance heterogeneous soc. as already said too, the chip is absolutely enormous and it is highly unlikely apple makes a profit on them.
<chadmed>
theres no magic, its just the logical outcome when you pour billions and billions and billions of dollars into minmaxing perf/W in a very well designed microarchitecture implementing a very well designed ISA on the worlds most advanced process nod
<chadmed>
s/nod/node
<chadmed>
actually i should retraact my initial statement. actuaally bothering to design a high performance consumer-class ARM chip is in and of itself something special unfortunately lmao
saintdev has joined #asahi
aeroraptor has joined #asahi
guan has joined #asahi
mustache_ride_ has joined #asahi
jabashque has joined #asahi
tardyp has joined #asahi
enomem has joined #asahi
marvin24_ has joined #asahi
marvin24 has quit [Ping timeout: 480 seconds]
chadmed has quit [Quit: Konversation terminated!]
chadmed has joined #asahi
chadmed has quit [Quit: Konversation terminated!]
chadmed has joined #asahi
chadmed has quit [Remote host closed the connection]
<citizen1[m]>
<chadmed> "theres no magic, its just the..." <- if they don't make profit on then why are they making them? they put it in macs and they make money off the macs
<citizen1[m]>
if its not special let's see intel and qualcomm do the same. they even switched their software to arm over x86 just to be some to do this
<citizen1[m]>
can you reach same perf/w with x86?
name327522 has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
Gues__________________________ has joined #asahi
chadmed has joined #asahi
maor26 has joined #asahi
Synth[m] has joined #asahi
alexstore06 has joined #asahi
alexsto__ has quit [Ping timeout: 480 seconds]
alexstor_ has joined #asahi
alexstore06 has quit [Remote host closed the connection]
Gues__________________________ has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
Gues__________________________ has joined #asahi
frytaped has quit [Quit: went bye bye]
alexstor_ has quit [Remote host closed the connection]
dbr^ has joined #asahi
maor26 has quit [Remote host closed the connection]
maor26 has joined #asahi
bgb_ has joined #asahi
bgb has quit [Ping timeout: 480 seconds]
aleasto has joined #asahi
<nsklaus_>
chadmed: "it is highly unlikely apple makes a profit on them (m1 soc)" i think they make various kind of profits with them. to give 3 examples: 1) they sell more macs, since they succeded in creating a desirable product. 2) they are free of waiting for intel available stocks and free from paying royalties and contract with them. 3) this ties to point 1 they gained more liberty to
<nsklaus_>
pursue they own designs (and we now see what they've done with it). so, selling more, paying less to intel, and freedom in design
<chadmed>
yeah i just meant on the actual chip itself. the dies are quite large
<nsklaus_>
i don't think apple is doing the same think micro$oft did with xboxes (making them at a loss just to be present on the market)
<nsklaus_>
beside, the dies are quite large indeed, but the price is quite large too, especialy if you 'boost' the options when buying one (i mean larger ssd, more memory etc ..) it can quickly reach 4000$ for a laptop for example
<nsklaus_>
i'm not defending apple's choices here, i'm just pointing that i bet they make money rather than loosing some
Gues__________________________ has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
<mort_>
does it even make sense to talk about whether Apple makes or loses money for each chip sold when they're not selling chips? I'm sure every product the chips go into earns them money, and you're not exactly paying separately for each component
<Bastian[m]>
I can't imagine that apple doesn't save money on the chips now that they don't need to buy them from intel
<mort_>
yea Intel's chips are probably marked up like crazy, or at least were in the many-year period where they had literally 0 competition
<mort_>
making big dies isn't cheap, but Apple can sell chips even if a bunch of the CPU and/or GPU cores turn out to be broken, an M1 Max with over half its GPU dead and two of its CPU cores dead can presumably be sold as the lowest end M1 Pro
<chadmed>
mort_: you have a point in that theyre not selling chips, i believe the original context was me explaining to citizen1 why these chips arent exactly "magic" -- apple doesnt care about minmaxing perf/mm^2 and as such can make huge chips that other companies would be eating a fat loss on
<mort_>
aha
<jannau>
I doubt partially defective m1 max dies are sold as m1 pro unless jade c chop is literally chopped
<mort_>
being a node or two ahead of the rest of the industry also helps, and that's just down to being Apple with a giant pile of money and customers
<chadmed>
well its basically the nvidia gt200 strategy except this time it worked because this chips arent crap
<chadmed>
and yeah apple has infinite money
<mort_>
I would be a bit surprised if they don't have a way to salvage Jade C-Die with even a single dead CPU core, unless the Max is so low volume that it doesn't matter
<chadmed>
theyll probably have a way. even if the current m1 max is low volume once they start chucking it in more devices theyll need a way to minimise the amount of wasted wafer
<mort_>
all the die shots make the C-Chop look like literally a C-Die with the extra GPU and memory interface chopped off
<mort_>
I'd imagine they manufacture the C-Chop separately as well though, otherwise there'd be a lot of wasted silicon
<chadmed>
surprised they didnt go MCM from the get go for these things, its not like the concept is experimental or anything
<mort_>
chadmed: I suppose one way to characterise that you said is: if Apple were selling M1 chips separately, they wouldn't be able to price them competitively from a performance/price perspective
<mort_>
from that leak which gave us the Jade C-Die/Chop names I would guess that the 2C-Die and 4C-Die will be MCM
<mort_>
they're still just at 8 big cores to be fair
<mort_>
doesn't even AMD have their 8-core chiplet as their building block, from which they build their 16- and 32-cores
<chadmed>
yeah but the soc itself is like 60bn xsistors
<mort_>
yeah, that's probably mainly GPU
<chadmed>
unless amd has a patent on the very concept of offloading IO to a separate die to optimise yields then frankly i have no idea why you wouldnt do that for every design going forward. of course its likely that these chips have been in at development since before amd proved the concept.
<mort_>
they seem much more aggressive at salvaging chips with dead GPU cores than chips with dead CPU cores, maybe they just don't see defects in that area as a big problem?
<mort_>
the separate IO die has been an issue for AMD in terms of memory latency, hasn't it? IIRC people have been talking about the 5000 series as the first generation which doesn't suffer in gaming from having a separate IO die
<chadmed>
i dont think it was the IO die as much as it was their shithouse IMC. zen/zen+ had utterly atrocious memory latency too
<mort_>
memory latency and bandwidth seems like a huge focus for Apple
<mort_>
I see
<chadmed>
i do believe the IO die introduced some, but it was miniscule compared to the penalty incurred by the imc design itself\
<mort_>
I'm guessing imc stands for integrated memory controller
<chadmed>
yessir
<mort_>
do you have any details on what makes the imc so bad?
<mort_>
not that this is #amd
<chadmed>
no design details im afraid, but well documented experimentation from press outlets (aand my own) shows that zen, zen+ and zen 2 cores are all extremely starved for data and that tuning tertiary memory timings provided a significantly higher performance boost than did a core frequency overcloc
<chadmed>
benchmarked memory access times were also absolutely horrible, and the memory controller in those microarchitectures could not train and hold anything above 3200Mbps on the memory bus, even with loosened timings
<mort_>
right. I did read that zen loves right RAM timings and high memory frequencies, I suppose I either assumed or heard that's due to the MCM stuff with a separate IO die
<chadmed>
some golden samples with very very good b-die DIMMs managed to hold 3400CL16 but this was rare until zen 3
<mort_>
what do you work with by the way? Anything related to CPU design? You seem suspiciously deep in the weeds of that industry
<chadmed>
ahahahah nope im a medical researcher right now by trade :P
<mort_>
aha :p Just a casual interest then
<chadmed>
yeah though im having more and more trouble keeping it casual. if australian unis had better computer science degrees i 100% would have done one already
X-Scale` has quit [Ping timeout: 480 seconds]
<mort_>
there's someone I know who designs risc-v chips for a living who also said almost verbatim that the M1 chips aren't magic
<chadmed>
yeah theyre not. theyre well designed chips and a textbook implementation of a highly integrated heterogeneous soc, but magic they are not
<chadmed>
just what having infinite money and a perpetual arm architectural license can do
<mort_>
seems pretty clear though that they're a significant departure from what the industry has been doing for ages
<chadmed>
things seems to have been heading this way for a while. the consumer desktop market is really the last one in which arm hasnt made any significant inroads and the biggest problem for it has always been the chicken-and-egg paradox. apple can resolve this by being both the chicken and the egg in thaat they control the software and hardware
<mort_>
fair
<mort_>
I know Microsoft has tried to move to arm for a while but without something like Rosetta 2, and without being able to just force everyone to recompile everything, Windows on arm is completely worthless
<chadmed>
windows is completely worthless* ftfy
<mini>
mort_: from what I understand M1 is very much quite conventional, but they've just gone *huge* on everything - it's super wide for instruction decode, it has a *massive* reorder buffer, it has *giant* caches
<mort_>
well, Windows on x86 has some value as a compatibility system for certain applications
<chadmed>
and therein lies why the consumer desktop segment has been so hard to crack for arm (or any x86 competitor from the release of NT onwards)
<chadmed>
its a segment almost entirely under the yoke of microsoft and intel, who set virtually all the standards that the devices in this segment are built around
<TheLink>
ms makes most of its money with companies licensing their software
<TheLink>
this business is very conservative and slow moving
<TheLink>
compatibility is all and everything there
<mort_>
mini: any particular reason Intel hasn't just done that and 4x'd their performance per watt in laptops?
<mini>
my workplace actually bought an ARM Surface Pro - it was *slow*, even running native ARM apps
<chadmed>
mort_: its almost entirely impossible to do that with an amd64 chip
<mini>
and almost nothing is native, and the performance hit running x86 stuff (and at the time it was only 32-bit apps you could emulate) was substantial
<mort_>
with an x86 chip* I assume
<chadmed>
what _is_ magic in chip design is how intel has managed to get 6-way decode on their upcoming alder lake chips
<mort_>
oh
<mort_>
you wrote amd64, I read arm64
<chadmed>
no problem lol
<mort_>
those names are too similar
<mini>
mort_: afaik, x86 / x86_64 is very hard to do big parallel instruction decode? variable length instructions cause a big issue apparnetly?
<mini>
but the ARM surface pro really has phone cores designed to be phone cores propping up a tablet, and they don't really keep up
<chadmed>
microsoft's arm efforts have been so half arsed
<mini>
it was unpleasant for web browsing, even.
<mini>
decent battery life though?
<mort_>
qualcomm chips have been so bad, and the entire non-Apple world have been stuck with them
<chadmed>
chicken and egg. theres been no incentive for better chips. or rather, these incomprehensibly large and fiscally conservative companies have not been willing to take the gamble on trying to displace x86
<mort_>
but even compared to the ultra-expensive Android phone market, Apple has been many years ahead in just their phones
<mini>
I'm curious as to what google do next
<mini>
they've done their first own SoC, I am genuinely curious if that translates into them doing their own cores next
<Bastian[m]>
msft also started hiring for a surface SoC
<chadmed>
amd should have just released K12. theyd still have keller, theyd have a huuuuge head start on intel in terms of decoupling from dependence on x86 for revenue
<mini>
but this process has a multi-year lead time
<mini>
look at how long apple have taken to get to this point
alexstore06 has joined #asahi
<mini>
apple are of course helped by the fact that they can just throw money at things like ensuring they have TSMC capacity though
phire has quit [Ping timeout: 480 seconds]
al3xtjames1 has joined #asahi
al3xtjames has quit [Read error: Connection reset by peer]
al3xtjames1 is now known as al3xtjames
alexstor_ has joined #asahi
alexstore06 has quit [Ping timeout: 480 seconds]
lanodan has quit [Quit: WeeChat 3.1]
lanodan has joined #asahi
X-Scale has joined #asahi
LiamGibson[m] has joined #asahi
X-Scale has quit []
arahael1 has quit [Ping timeout: 480 seconds]
arahael1 has joined #asahi
dbr^ has quit [Ping timeout: 480 seconds]
barkbork has joined #asahi
arahael1 has quit [Ping timeout: 480 seconds]
arahael1 has joined #asahi
bgb_ has quit [Read error: Connection reset by peer]
bgb_ has joined #asahi
barkbork has quit [Quit: barkbork]
X-Scale has joined #asahi
alexstor_ has quit [Remote host closed the connection]
alexstore06 has joined #asahi
alexstor_ has joined #asahi
alexstore06 has quit [Ping timeout: 480 seconds]
alexstore06 has joined #asahi
alexstor_ has quit [Ping timeout: 480 seconds]
bps has quit [Ping timeout: 480 seconds]
rohin2 has joined #asahi
aleasto has quit [Quit: Konversation terminated!]
___nick___ has joined #asahi
___nick___ has quit []
___nick___ has joined #asahi
zopieux has quit [Ping timeout: 480 seconds]
rohin2 has quit [Ping timeout: 480 seconds]
zopieux has joined #asahi
alexstor_ has joined #asahi
alexstor_ has quit [Remote host closed the connection]
alexstor_ has joined #asahi
alexstore06 has quit [Ping timeout: 480 seconds]
leo60228- has joined #asahi
alexstor_ has quit [Ping timeout: 480 seconds]
Gues__________________________ has joined #asahi
malvo has quit [Ping timeout: 480 seconds]
alexstore06 has joined #asahi
gladiac has quit [Remote host closed the connection]
gladiac has joined #asahi
alexstore06 has quit [Ping timeout: 480 seconds]
alexstore06 has joined #asahi
gladiac has quit [Read error: Connection reset by peer]
gladiac has joined #asahi
gladiac has quit [Remote host closed the connection]
gladiac has joined #asahi
Gues__________________________ has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
alexstore06 has quit [Ping timeout: 480 seconds]
gladiac7 has joined #asahi
gladiac is now known as Guest3875
gladiac7 is now known as gladiac
Guest3875 has quit [Read error: Connection reset by peer]