Discussion Zen 7 speculation thread

Page 8 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

soresu

Diamond Member
Dec 19, 2014
3,932
3,362
136
Well those 4 FP Units shine figuratively and literally
In the words of the great Moss....
tenor.gif
 

Chicken76

Senior member
Jun 10, 2013
270
53
91
Verano is a village with barely 1000 inhabitants, i doubt that it was ever mentioned elsewhere than in this thread, they just made a typo out of Verona since that s a big city that is close to Venice.

It makes sense that it's Verona, right? That's where the 128 Petruchios (along with who knows how many dense Grumios) come from in order to tame the 192 Rapide-di-Diamante Katarinas. The outcome has already been written so: (act 2, scene 1)

1753965769188.png

That's another 20% server CPU market share on launch, and upon bankruptcy, half the fabs.
 

Abwx

Lifer
Apr 2, 2011
11,855
4,831
136
It makes sense that it's Verona, right? That's where the 128 Petruchios (along with who knows how many dense Grumios) come from in order to tame the 192 Rapide-di-Diamante Katarinas. The outcome has already been written so: (act 2, scene 1)

View attachment 127986

That's another 20% server CPU market share on launch, and upon bankruptcy, half the fabs.
Well, in the waiting of more marketshare in servers they say that they have no competition in HEDT and wokstations, that shouldnt be far from 100% marketshare in those segments.
 

Attachments

  • AMD TRH SLIDE.jpg
    AMD TRH SLIDE.jpg
    52.1 KB · Views: 13
  • Like
Reactions: Tlh97

LightningZ71

Platinum Member
Mar 10, 2017
2,377
3,003
136
I wonder if Zen 6 or Zen 7 era Threadrippers will get NPUs on the IODs as well? It would seem to be a missing feature for Windows OSes if they don't.
 

511

Diamond Member
Jul 12, 2024
3,296
3,214
106
While I agree that it would likely be best on a PCIe card in workstations, not having an NPU may be an issue for Windows OS going forwards.
run linux is what i would like to say but sadly it is not an option every time it has it's quirks and nothing like updating your arch install and unable to boot next time cause there was this time and you missed to add hook and waste hours on debugging a luks encrypted drive with systemd and secure boot.
 
Jul 27, 2020
26,526
18,247
146
Any PC with dGFX gets it by default.
Currently, no.

This is 10 days old:


so when Windows 11 came out,

Mhm.

Microsoft's like, you need a TPM chip.

The future is secure. Future is great,

whatever. And then not long after that,

they're like, if you really want like a

cool Windows machine, you need an NPU

and like go get an NPU

because it'll do cool things.

Yep.

However,

yep,

they've kind of boxed themselves into a

weird scenario where it's like, if you

want an AI version desktop OS, you've

got to have a co-pilot plus PC. However,

every single AI company on planet Earth,

including these new agents that have

just been announced, have just proven

that you do not need an NPU to do

anything related to AI, uh, even on your

local device at this point anymore.

And so, Microsoft's over here saying

like, "Get an NPU. It's for cool AI

stuff." And chat GBT is like, "We don't

need your NPU. Like, we'll do all this

cool stuff without it."

Brad, is there a question in here

somewhere?

I don't know. Like, I feel like

Microsoft has screwed themselves. And um

so I Yeah. Uh

how do I say this? If you think about

the evolution of MPUs on phones, which

started out as special use case chips

for like you know like um the pixels had

something called I think it was called

pixel visual core. It was sort of a

you know a way to bypass the CPU

integrated GPU to do certain camera

related things because you know they

were slow and they were doing

computational photography etc. And then

Apple and Google and whatever so now we

have MPU so they they have

There are things that are there that you

don't think about, but they speed up

certain operations, right? That's the

point. And to a

to an incredible degree, I mean,

honestly, the the problem of PCs is that

they're these big boxes that have GPUs

and whatever. And so, if you do you've

done this, I mean, if but if anyone

listening to this has ever tried to play

around with local AI, however you do it,

like AI Studio or whatever it's called,

Visual Studio Code is the one I've used.

Um, there are a lot of local models that

just run on the CPU and GPU and there

are some that run on the MPU for sure,

but and maybe they're better and more

efficient, but this I don't I don't mind

the notion of co-pilot plus PC, but I

feel like if you buy a gaming rig or you

have some kind of a big hulking Nvidia

whatever GPU,

right,

you should be getting this stuff. And

we're past the point where it makes

sense that that's not the case. So, I

don't I don't know if this is like

protecting their partner Qualcomm

because they're not quite there yet with

the GPU stuff or whatever, but um but

the anyway, but the primary point is

there I don't know that there is a

compelling use case on the

PC as there is on the phone because

phones are, you know, everyone takes

pictures,

you know, so you're like, "Okay, this

makes sense to me." Yep. On the PC,

there are many, many use cases for it,

but it's always like this individual

feature in some app that you may or may

not use. So, like I use Affinity Photo,

they have an object selection tool that

uses local AI. You have to download a

model, but unless I'm missing something,

I I think it runs off the CPU. I don't

think I don't think it needs

an MPU. So you could use a cap cut or

whatever video editor Da Vinci Resolve

does this I guess where one to four or

five features if you have the MPU run

better but that's not a

it's not a selling point like I I I

think when people buy like a Pixel

they're like I one of the reasons would

be I really like the quality of the

photos they're not really thinking about

the MPU or whatever. Um, when you buy a

Windows PC, you're doing it for all the

normal reasons that's always been true.

Yeah.

The apps and blah blah familiarity,

whatever it is. And the MPU is like,

it's like, okay, I mean, I think we're

at the point where people at some point

have to buy a new PC. They do it

begrudgingly and they do or do not get

whatever hardware and they do or do not

get whatever benefits. But I just did

this in an ask Paul thing on Friday. you

list out like like what's in Windows and

there are some like you if you've used

click to-do is actually you know it's

useful like there it's

could that run on any PC I don't know

could it run on any PC with a GPU yeah I

mean absolutely would it be as efficient

no probably not does anyone really care

I mean like I

yeah when you're telling somebody who

has an AMD whatever 9800X 3D

they put a 5090 in there from Nvidia and

they're like, "You know what? You can't

actually remove backgrounds in photos

because you don't have a co-pilot plus

PC." Like, that's a that's a problem.

It's artificial. That's the problem. And

they can talk. What is it like 50 40

whatever the number is for Copilot Plus

PCs? 45 tops something whatever.

Um these Nvidia GPUs have like 450 or

500 or whatever the figure is like. Plus

your PC has a combined tops rating that

is CPU, GPU, MPU. And I I always talk

about this for whatever reason, but like

this is

the the system, Windows in this case,

should orchestrate every one of those

tasks to whatever processor makes the

most sense based on what you have.

That's that type of scheduling or

whatever you want to call it,

orchestration is what an operating

system does. It it

the only reason Windows doesn't do this

right now is I I think it's artificial.

I think it's protecting the partner that

went they went to market with and it's

like they must have some we always talk

about this with Qualcomm for some reason

there's like this period of exclusivity

or whatever. I it this is beyond the

point of making sense.

I don't know.

Yeah. I I don't

I don't know. I know. Well, and the

other thing just for what it's worth and

you've done this like again local AI

versus cloud AI you can have that chat

experience. It's just like using chat

GPT or copilot whatever and have it do

things and what you'll see is it's not

as good right and the graphical way to

say this which I also mentioned and

asked Paul the other day is I always

forget which is which but I co-creator

is the feature I believe that's in paint

that is local and requires a copilot

plus PC and you can have it generate

images they look like a child's thing

that you would put on the refrigerator

and then you go to you know designer on

the web or any midjourney or all the

other things have to generate an image

and it's in disccernible from like a

photograph and that's the difference

between cloud and local AI. Now the

thing in the cloud is burning through

you know like a city's worth of energy I

guess to make that image but

um I don't think anyone cares not to be

but I mean

I believe you're correct and I'm not

sitting here trying to say like NPUs are

are a waste of time or whatever. It's I

do think locally run agents are a thing

in the future. We're not there yet. But

the problem is is that Microsoft has

intentionally fragmented

y

where they are today and like like

they're already outdated because of just

just hardware at the end of the day and

then you have more powerful things out

there that can do what those do but they

won't let them do it because it doesn't

make sense because that breaks their I

don't know

look Intel is struggling but one of the

parts of the story one of the part of

that story to use English is

apprised of co-pilot plus PC. Intel had

to rush into market this lunar lake

thing which wasn't what was going to

happen next. And they had to pull from

different parts and cobble together this

thing and use different like sheets of

you know layers and and integrate RAM

like AMD and Qualcomm already do. And it

it cost them I mean it almost sunk the

company and they're never going to do it

again. And since then they've released

desktop chips and other mobile chips,

Aerol Lakeake, whatever they're called,

that don't have those MPUs. And like,

you know, Intel, I mean, look, they're

struggling, but I mean, they they

understand making chips like that was a

big struggle for them to get there, you

know, now next year or this coming year,

I guess. Uh they'll probably correct

that. I'm sure eventually now we'll get

to that point where every mass-roduced

Intel chip other than the super low-end

stuff will have these capable MPUs, I

guess. But like again,

okay, great. So, two years in now, I

guess by the time that happens, what's

the big benefit? Is it I can hold down

the Windows key and see purple and, you

know, pink colors on my screen.

That's the whole thing.

I I don't Yeah, it's like the new Siri,

like the What's the biggest benefit of

the new Siri? It's really pretty, you

know. It's really pretty. stupid as

ever,

you know, but uh more attractive user

interface. Um

they got to fix this because Windows

should be the premier spot for all for

everything. I mean, if you're a

developer, for a developer, if you're a

user, for whatever the functions are and

it should take advantage of the features

of your computer. That's that's what

Windows does, dude. I know that's the

whole point here is like they've got

this fabricated completely fictitious

line in the sand of a co-pilot plus PC

but it's holding back

the other part. So I think uh given the

lingo of the day we could make the

argument that this is another form of

insuranceification certification as we

can say on the old show and the reason

being that Microsoft's corporate aims in

this case some strategic partnership

with Qualcomm on MPUs and blah blah blah

and forcing us down everyone's throats

for for reasons that make sense on

laptops and whatever uh has created an

artificial

bifurcation or whatever features in

Windows and you've got people have spent

more money than others on really

expensive GPU based PCs, whether they're

gaming computers, workstations, whatever

they are that Windows is like, yeah, I

don't even see that thing. And I'm

sorry, but that's freaking stupid. Like,

well, and it's even worse when Nvidia

comes along with their own app and says,

"Look, here now you can do it all. It's

not officially Copilot Plus PC, but

look, here's the software. It does

everything that you wanted." And

well, you use Microsoft's own tool, the

use of Visual Studio Code, download

local models and run them on your GPU,

and guess what? They run great. you

know, as good as those things can do. I

mean, you

correct? It's still slower. And

look, I you could drive across the

country in an efficient vehicle like

yours. You could go in a monster truck.

The monster truck is going to guzzle

more gas, but if that's what you have,

you can still drive across the country.

Like, that's your choice.

Yep.

It doesn't make sense

to

uh kneecap the most powerful computers

there are. Like, what are you doing,

Microsoft? Like, it doesn't make any

sense. Obviously, you let those guys in.

To me, I don't know.

And by the way, that doesn't threaten

Qualcomm. They don't make those kind of

computers. What's the difference? Look

at Jesus. Like,

I don't know. I Qualcomm went to market

uh with with Snapdragon X

in what I consider to be the sweet spot

of the market, right? Uh thin, light,

premium laptops. This is what most

people want. So, good, good. But for

those who are sitting there with this

hunking tower on their thing or they got

this thick whatever gaming laptop, I

don't understand not letting them be

part of this. It doesn't hurt Qualcomm,

but you it does hurt those customers,

those people. Like it there's no

rational explanation for it.

Thank god they don't use this kind of

reasoning when they lay people off

because just

Yeah. I don't know. I I just can't get

this like out of my head that Microsoft

has backed itself into a corner. And

part of me and and I think I understand

one of the reasons I bet that they

charge more money for that co-pilot plus

PC branding. So if you want to have a

device designated that they probably

there's upcharge. So if they just say

like, "Oh, everything that's whatever

5090 is, well, Microsoft's like, well,

we didn't make money from it, so we're

not going to give it to them." Like,

which is

okay because there's whoever Lenovo has

Legion computers. Dell has what they

Alienware, HP, whatever they have.

He's telling me that these people, these

companies are selling a two $3,000

gaming whatever,

and they're cheapening out on the

Windows license. Like,

I'm not even worried about the laptops.

I mean the desk.

No, but I'm saying like Windows 11 Pro

is probably the cheapest, right? Windows

11 Pro and then you you're what you're

saying and I'm probably right, is

Copilot Plus is essentially a third

skew, although it could be Home or Pro,

right?

But essentially a sort of an added cost

license like a like it's not just

meeting the spec. You actually have to I

that's an that's a good question. I I

wonder if I could probably find that

out. I'm gonna I'm gonna ask around and

see if that's

um because you would think look I'm sure

am or no not am well AMD and Intel but

also the PC makers that are like look

most of the computers we sell are Intel

like I mean uh what are you doing you

know like what do we what do we what's

this weird requirement but then again I

don't know dealing with Microsoft all

these years they're probably used to

stupidity but um yeah this is not the

like the margins are so low on PCs This

is not the type of thing you like nickel

and dime companies over. It's like just

give them this. Like what are you doing?

If you want to advance the platform,

I don't know what's happening. I wish

you hadn't mentioned this. Now my brain

is all fried.

Happy to ruin your week.

Thank you. Happy Monday, you bastard.

Interestingly, Thurrott saying that Lunar Lake exists because of M$'s NPU push.
 

eek2121

Diamond Member
Aug 2, 2005
3,387
5,014
136
They can pry the GPU compute cycles from my cold, dead fingers.
They already do use it for virus scanning and stuff.
FINALLY some interesting leaks to talk about:


It looks like AM6 has over 2100 pins (at least 22% increase from AM5). Though this probably means we won't have any radical changes on the desktop, such as 3 memory channels or something.
Not likely. More likely to be an increase in PCIe lanes.
 

Gideon

Platinum Member
Nov 27, 2007
2,022
5,014
136
This is the original source article:


It also has pictures:

lQuij2N.png


The pin density is really tight:

1754393156985.png
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
27,114
16,026
136
This is the original source article:


It also has pictures:

lQuij2N.png


The pin density is really tight:

View attachment 128296
thats for server, not desktop. eek2121 was tasking about desdktop pints on that is 6096+
 
  • Like
Reactions: Tlh97 and Gideon

CouncilorIrissa

Senior member
Jul 28, 2023
668
2,592
106

StefanR5R

Elite Member
Dec 10, 2016
6,591
10,380
136
Whoa, these images look like terrible generative AI slop.
But the patent seems to exist indeed:
https://www.freepatentsonline.com/y2025/0149428.html
Abstract:
A method for dimensioning a land grid array pad can include forming an initial landing area of a land grid array pad, wherein the initial landing area is dimensioned to cause at least a majority of a landing surface of one or more socket pins to land off of the initial landing area prior to actuation of one or more sockets including the one or more socket pins. The method can also include forming a final landing area of the land grid array pad, wherein the final landing area is dimensioned to maintain electrical contact with the landing surface of the one or more socket pins after actuation of the one or more sockets including the one or more socket pins. Various other methods and systems are also disclosed.
Edit: Translation = How to be stingy with LGA contact pad size. AFAIU.

(And yes, whoever claimed that this looks like AM5 probably does all his research by asking a random AI bot and call it a day.)
 
Last edited: