• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

RX 7900XT, 15,360 cores, MCM, Tapeout Q4

Caporegime
OP
Joined
17 Mar 2012
Posts
47,382
Location
ARC-L1, Stanton System
@Harlequin could be right by how the following is worded.

Lisa Su during the recent Investors Call

How do you read that, is it because she said "And our RDNA3 GPU's" after talking about Zen 4 being on 5nm?

Because Apple have N5 tied up till 2022, literally 90% of the entire node. Actually i think AMD will use N6 (5 layer EUV version of N7 - N7P isnt compatible with N7, but N6 is). It`ll be Q3 2022 before anyone gets anything more than early production on N5 once Apple fully moves away to N3 Intel and Apple tipped to be the first to adopt TSMC's N3 node-based chips | TechSpot

Is this a fact? Are Apple really using "literally 90% of the entire node" or is that just something you're assuming because its Apple and any extreme you can think of must be true?
 
Associate
Joined
31 Dec 2010
Posts
2,428
Location
Sussex
No real surprise since CPUs are a lot more profitable and they have the chiplet thing well sorted out already.
From yesterday's Intel Foundry article on Anandtech we have some density figures
jyFJpdr.png
https://www.anandtech.com/show/1682...nm-3nm-20a-18a-packaging-foundry-emib-foveros

Power saving would figures would be somewhere else I guess.
7^2 is 49 while 5^2 is of course 25. And 171/91 almost hits that with a x1.88 scaling.
Irony of course being that GPUs can use as many transistors as possible whereas CPUs are harder to scale.
Top Zen4 EPYC might go to 128 cores?
Luckily cloud VMs means many loads don't have to be that parallel.
 
Soldato
Joined
17 Jun 2004
Posts
7,587
Location
Eastbourne , East Sussex.
Associate
Joined
31 Dec 2010
Posts
2,428
Location
Sussex
How do you read that, is it because she said "And our RDNA3 GPU's" after talking about Zen 4 being on 5nm?
I guess it could be a weaselly way a CEO, lawyer, or politician would have to word it if RNDA3 is not 5nm.
If that reading is accurate, then the statement:
We remain on-track to launch next-generation products in 2022, including our Zen 4 processors built with industry-leading 5nm process technology and our RDNA 3 GPUs.
is borderline miss-leading though. Now if there was a comma before the and it would be far clearer.
 
Caporegime
OP
Joined
17 Mar 2012
Posts
47,382
Location
ARC-L1, Stanton System
I guess it could be a weaselly way a CEO, lawyer, or politician would have to word it if RNDA3 is not 5nm.
If that reading is accurate, then the statement:

is borderline miss-leading though. Now if there was a comma before the and it would be far clearer.

It could also just be disjointed rambling. she wanted to make an emphasis on Zen 4 and rambled RDNA3 in as an after thought, i haven't listened to the to the conference call, doing that might make clear in speech what isn't in text.
 
Soldato
Joined
17 Jun 2004
Posts
7,587
Location
Eastbourne , East Sussex.
Huawei 5nm commitment has gone to Qualcomm, AMD are at 5% of total 5nm output for H1 2021 leading to H2, when the constrained supply at the fab will start to ease as Apple moves to N3. AMD are taking as much of Apple 7nm capacity as they ramp to 5nm; however

TSMC Manufacturing Update: N6 to Match N7 Output by EOY, N5 Ramping Faster, Better Yields Than N7 (anandtech.com)

AMD want 7nm for interposers (currently at GloFo on 12nm) , which will use that new capacity, higher clocks will come from 6nm EUV for the GPU`s.
 
Caporegime
OP
Joined
17 Mar 2012
Posts
47,382
Location
ARC-L1, Stanton System
Huawei 5nm commitment has gone to Qualcomm, AMD are at 5% of total 5nm output for H1 2021 leading to H2, when the constrained supply at the fab will start to ease as Apple moves to N3. AMD are taking as much of Apple 7nm capacity as they ramp to 5nm; however

TSMC Manufacturing Update: N6 to Match N7 Output by EOY, N5 Ramping Faster, Better Yields Than N7 (anandtech.com)

AMD want 7nm for interposers (currently at GloFo on 12nm) , which will use that new capacity, higher clocks will come from 6nm EUV for the GPU`s.


Huawie here being HiSilicon, as it written in all these articles, was dropped at N7 already, that capacity was split between Qualcomm and AMD

https://www.extremetech.com/computi...ization-improves-on-orders-from-amd-hisilicon

After skimming over it, i don't see anything written in your Anand article which states HiSilicon now cancelled N5 allocation was given to Qualcomm, what it does say about Qualcomm is that they are using N6.
 
Soldato
Joined
17 Jun 2004
Posts
7,587
Location
Eastbourne , East Sussex.
your article is 2 years out of date, since then the US government has threatened sanctions over any company working with Huawei , so TSMC ended teh contractwith them, which Qualcom hoovered up along with Mediatek. Its the reason Intel along with Apple are now pushing money into N3, as n5 is booked out till 2023
 
Caporegime
OP
Joined
17 Mar 2012
Posts
47,382
Location
ARC-L1, Stanton System
How likely are thinking this thing is actually 2.5x a 6900XT? And any predictions for the price and performance of the 7800XT?

In performance terms, probably that at best, i think the prediction is 2.2 to 2.4X.

7800XT predicted to be 30 to 40% faster than the 6900XT, which with 50% more cores seems about right, +35%.

I think to keep the power down the two die that make up the 7900XT will be lower clocked, probably to 2Ghz flat, 200 Watts each to make 400 Watts, 450 Watts board power, the 7800XT would be clocked in the usual 2.4Ghz.

The 6900XT is 300 Watts board power, the RTX 3090 is 350 Watts board power.
 
Soldato
Joined
6 Feb 2019
Posts
17,467
The good news is that AMD is almost guaranteed to do away with the low bandwidth G6 memory that RDNA 2 used thanks to the US.

Parts of the U.S has introduced energy efficiency standards for PCs and now RDNA 2 based PC's can't be sold in certain areas due to maximum permitted energy use of a 256bit GPU.

Under these new regulations, the amount of power draw the GPU is allowed is determined by its memory bit rate and throughput - this now incentivizes Nvidia and AMD to give their discrete GPU's as much memory bandwidth and bit rate as possible to by pass energy regulations, lest they don't want their products to be sold in the US.
 
Associate
Joined
24 Jun 2016
Posts
842
Location
Hartlepool
It could also just be disjointed rambling. she wanted to make an emphasis on Zen 4 and rambled RDNA3 in as an after thought, i haven't listened to the to the conference call, doing that might make clear in speech what isn't in text.

RDNA3 is meant to be split across 5nm for Navi 31 (the MCM products) with the rest going with 6nm.

That would explain her statement if not all RDNA3 will be on the same process.
 
Back
Top Bottom