Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)

Page 941 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Geddagod

Golden Member
Dec 28, 2021
1,340
1,431
106
Conversely, I'm convinced Zen5 was as good as AMD hoped... when they started its design. The problem for them, which has tainted OUR perception was that Intel actually did have a last hurrah, with the recent Lakes, and both companies realised they would have to push their chips HARD to compete. Zen4, IMO, was pushed way harder than AMD originally intended, to compete with 13/14 series. So much so that it crossed the performance line into plannrd Zen5 territory, a generation earlier.
AMD engineers claimed they under-achieved their Zen 4 Fmax goals. I don't think, unless AMD increased nT power draw for Zen 5 while keeping the all core boost of Zen 4 arbitrarily low, Zen 5 was going to be a large upgrade in traditional nT perf.
I don't think Zen 5 is a bad uplift totally, but it's not on par with what AMD achieved with Zen 3 (core wise). Maybe because their base with Zen 4 was already so good. The Zen 5>Zen 4 PPA uplift seems more in line with Intel's Cove uplifts than Zen 3>Zen 4.
Comparing the GLC vs WLC and the Zen 5 vs Zen 4 uplift is going to be pretty interesting, if anyone can find data on a GLC vs WLC power curve (I have yet to seen anything with this data).
Kinda proof can be seen in how AMD have managed Epyc. Zen5 in that environment is a good 30-50% faster, cos it clocks easier across the board. I dunt think AMD originally wanted their first AM5 chips to instantly draw the full wattage the new socket could provide. But then they wouldn't have been competitive with Intels furnace inducing, self destructing chips. Ho hum.
Zen 5 server also increase core counts and has a new IO die. Idk if that's proof.
 

Geddagod

Golden Member
Dec 28, 2021
1,340
1,431
106
They needed something scalable for CCXes having 16 (and probably 32) cores. So this makes sense but having such CCX benchmarked would be great.
Perhaps early prep for Zen 6, which is rumored to increase CCX core count.
Someone remind me if Zen 5 dense is one or two CCXs? Wasn't Bergamo 2 CCXs?
 

Saylick

Diamond Member
Sep 10, 2012
3,855
8,910
136
AMD engineers claimed they under-achieved their Zen 4 Fmax goals. I don't think, unless AMD increased nT power draw for Zen 5 while keeping the all core boost of Zen 4 arbitrarily low, Zen 5 was going to be a large upgrade in traditional nT perf.
I don't think Zen 5 is a bad uplift totally, but it's not on par with what AMD achieved with Zen 3 (core wise). Maybe because their base with Zen 4 was already so good. The Zen 5>Zen 4 PPA uplift seems more in line with Intel's Cove uplifts than Zen 3>Zen 4.
Comparing the GLC vs WLC and the Zen 5 vs Zen 4 uplift is going to be pretty interesting, if anyone can find data on a GLC vs WLC power curve (I have yet to seen anything with this data).

Zen 5 server also increase core counts and has a new IO die. Idk if that's proof.
I agree. I think Zen 5 was likely planned on some variant of N3 but since it wasn't ready in time, Zen 5 needed to be backported to N4 and the reduced xtor budget meant some things needed to be axed. Hopefully, those things and more come back with Zen 6.
 

511

Golden Member
Jul 12, 2024
1,670
1,489
106
Zen 5 details in Japanese ..
 

Joe NYC

Platinum Member
Jun 26, 2021
2,908
4,279
106
I agree. I think Zen 5 was likely planned on some variant of N3 but since it wasn't ready in time, Zen 5 needed to be backported to N4 and the reduced xtor budget meant some things needed to be axed. Hopefully, those things and more come back with Zen 6.

I think so to. And AMD can, in theory make a "Zen 5+", put in all the things that were cut and then some, call it "Zen 6" and release it earlier than expected. Such as Mid 2026.

AMD already has a template for the IODs / SoCs in Strix Halo...

TSMC is already scheduled to be processing 50,000 to 80,000 N2 wafers by the end of 2025. So N2 capacity is coming online quickly, and Apple is unlikely to use is for new iPhone in 2026.
 
Reactions: lightmanek

511

Golden Member
Jul 12, 2024
1,670
1,489
106
I think so to. And AMD can, in theory make a "Zen 5+", put in all the things that were cut and then some, call it "Zen 6" and release it earlier than expected. Such as Mid 2026.

AMD already has a template for the IODs / SoCs in Strix Halo...

TSMC is already scheduled to be processing 50,000 to 80,000 N2 wafers by the end of 2025. So N2 capacity is coming online quickly, and Apple is unlikely to use is for new iPhone in 2026.
Bruh Apple uses a ton of capacity the initial capacity is usually booked by apple and N2 is yet to ramp
 

Joe NYC

Platinum Member
Jun 26, 2021
2,908
4,279
106
Bruh Apple uses a ton of capacity the initial capacity is usually booked by apple and N2 is yet to ramp

Did you read what I posted? TSMC is expected to be processing 50k to 80k wafers per month by the end of 2025 (meaning, it will be ramping capacity substantially for 2026)

Unlike other times, when Apple buys most of the new node, there were rumors that Apple will not use N2 for 2026 mode iPhone. There is nothing else that Apple makes that can use a ton of leading node capacity. Laptop Mx use a fraction of iPhone capacity.

 

Win2012R2

Senior member
Dec 5, 2024
741
744
96
there were rumors that Apple will not use N2 for 2026 mode iPhone
Apple pre-pays advanced node expansion, maybe this time Nvidia outbid them but it seems to me that for Apple to miss such an important node would be unthinkable - they did N3B and N2 can't be any worse
 
Last edited:

Joe NYC

Platinum Member
Jun 26, 2021
2,908
4,279
106
Apple pre-pays advanced node expansion, maybe this time Nvidia outbid them but it seems to me that for Apple to miss such an important node would be unthinkable - they did N3B and N2 can't be any worse

It is still only a speculation, that only iPhone "Pro" models will use N2. and higher volume lower end models will not.
 

linkgoron

Platinum Member
Mar 9, 2005
2,533
1,210
136
It is still only a speculation, that only iPhone "Pro" models will use N2. and higher volume lower end models will not.
This wouldn't be a first that such a large difference would exist between the pro and non-pro line. The iPhone 15 had A16 Bionic and the iPhone 15 Pros had A17 Pro, and there was a similar difference with the iPhone 14, which had A15 while the iPhone 14 Pro had the A16 Bionic.
 

Abwx

Lifer
Apr 2, 2011
11,712
4,630
136
Did you read what I posted? TSMC is expected to be processing 50k to 80k wafers per month by the end of 2025 (meaning, it will be ramping capacity substantially for 2026)

They are actually targeting 100k/month for december according to a TSMC employee, but would be content with 50-80k as a worst case, 100k/month would be enough for 40M 150mm2 chips/month, almost 1bn/year, so unless Apple manage to take 100% of the smartphones marketshare they dont need that much waffers, at most 20% would be more than enough, and even at 50k/month they wouldnt need more than 40% of the output.
 

511

Golden Member
Jul 12, 2024
1,670
1,489
106
They are actually targeting 100k/month for december according to a TSMC employee, but would be content with 50-80k as a worst case, 100k/month would be enough for 40M 150mm2 chips/month, almost 1bn/year, so unless Apple manage to take 100% of the smartphones marketshare they dont need that much waffers, at most 20% would be more than enough, and even at 50k/month they wouldnt need more than 40% of the output.
only if the yield is good the yield is initially not that high it increases during process ramp up and i don't think all the designs are ~115mm2 the only chips around that are Zen6 CCD(expecting around 90-100 mm2) and A20 SOC.

a reference N5 was the best yielding node in TSMC History and they are not claiming N5 like D0 for N2 i would take the curve around N7 than for predicition.

 

Abwx

Lifer
Apr 2, 2011
11,712
4,630
136
only if the yield is good the yield is initially not that high it increases during process ramp up and i don't think all the designs are ~115mm2 the only chips around that are Zen6 CCD(expecting around 90-100 mm2) and A20 SOC.

I wrote 150mm2, not 115.

That s enough area for quite complexe chips given the augmented density vs previous nodes, they said density is 313M/mm2, that s 2x the density of the N4 based 9070XT GPU, beside at 50k/month they cant afford to have 50% yields, they would reduce the flow the necessary time to correct the hardware rather than keeping manufacturing excess scraps, so they seems confident in their capabilities.
 
Last edited:

511

Golden Member
Jul 12, 2024
1,670
1,489
106
I wrote 150mm2, not 115.
My bad lol I mixed up the numbers.
That s enough area for quite complexe chips given the augmented density vs previous nodes, they said density is 313M/mm2, that s 2x the density of the N4 based 9070XT GPU, beside at 50k/month they cant afford to have 50% yields, they would reduce the flow the necessy time to correct the hardware rather than manufacturing excess scraps, so they seems confident on their capabilities.
That quoted density is wrong I would assume something in line 250mxtor/mm2 for the correct N2 density considering it's 215mxtor/mm2 for N3E and N2 is 1.15 times the density N4 density is 143mxtor/mm2 and N5 is 135 mxtor/mm2.

It can't be 2X cause as even according to TSMC that's not true N3 was 1.3X Chip density and N2 was 1.15X Chip density on top of it using finflex the density was 1.6X for logic only (2-1 fin vs 2-2 fin for N5) and N2 is additions 1.15X chip density( logic may be higher but is yet to be confirmed also I am taking N3E as reference cause tsmc compared N2 2-1 vs N3E 2-1.)

135*1.6 = 216
216*1.15 = 248

And N4 is 143 Mxtor/mm2

Source
 

Joe NYC

Platinum Member
Jun 26, 2021
2,908
4,279
106
I wrote 150mm2, not 115.

That s enough area for quite complexe chips given the augmented density vs previous nodes, they said density is 313M/mm2, that s 2x the density of the N4 based 9070XT GPU, beside at 50k/month they cant afford to have 50% yields, they would reduce the flow the necessary time to correct the hardware rather than keeping manufacturing excess scraps, so they seems confident in their capabilities.

And how many would AMD need if Zen 6 were to ramp instantly?

Total market of optimistically 25 million per month, and optimistically AMD has 40% market share in 2026, and of that again optimistically, 50% are premium chips.

That would be 5 million Zen 6 CCDs per month which are estimated to be 75 mm2. How many wafers would it take to supply the client? (Server would be separate).
 

511

Golden Member
Jul 12, 2024
1,670
1,489
106
And how many would AMD need if Zen 6 were to ramp instantly?

Total market of optimistically 25 million per month, and optimistically AMD has 40% market share in 2026, and of that again optimistically, 50% are premium chips.

That would be 5 million Zen 6 chips per month which are estimated to be 75 mm2. How many wafers would it take to supply the client? (Server would be separate).
Depends on the CCD and we are assuming AMD will keep Growing it will only take 1 good Intel gen to sway everything back.
 

Joe NYC

Platinum Member
Jun 26, 2021
2,908
4,279
106
Depends on the CCD and we are assuming AMD will keep Growing it will only take 1 good Intel gen to sway everything back.

The Zen 6 CCD on N2 is rumored to be around 75 mm2. Using default defect rate in the calculator, I get 747 good die per wafer.

So in order for AMD to get 5 million of these CCDs per month (per post above, which is optimistic scenario for AMD), AMD would need ~6,700 wafers (out of the 100,000 wpm planned for December. So it would less than 10% of the wafers as of Dec 2025.

This is just a theoretical exercise, but it looks like it would be quite feasible for AMD to bring a Zen 5+ (called Zen 6) to market, just by putting back in all the things that were meant to be in Zen 5, and then some minor optimizations.
 
Reactions: lightmanek

511

Golden Member
Jul 12, 2024
1,670
1,489
106
AMD has been strong and dominent for years, and especially in servers. And you think one Intel generation will take it all away ? Yea right...

Just stay in your Intel threads.
I meant the Desktop.Servers are different game where AMD will keep taking share
 
Last edited:
sale-70-410-exam    | Exam-200-125-pdf    | we-sale-70-410-exam    | hot-sale-70-410-exam    | Latest-exam-700-603-Dumps    | Dumps-98-363-exams-date    | Certs-200-125-date    | Dumps-300-075-exams-date    | hot-sale-book-C8010-726-book    | Hot-Sale-200-310-Exam    | Exam-Description-200-310-dumps?    | hot-sale-book-200-125-book    | Latest-Updated-300-209-Exam    | Dumps-210-260-exams-date    | Download-200-125-Exam-PDF    | Exam-Description-300-101-dumps    | Certs-300-101-date    | Hot-Sale-300-075-Exam    | Latest-exam-200-125-Dumps    | Exam-Description-200-125-dumps    | Latest-Updated-300-075-Exam    | hot-sale-book-210-260-book    | Dumps-200-901-exams-date    | Certs-200-901-date    | Latest-exam-1Z0-062-Dumps    | Hot-Sale-1Z0-062-Exam    | Certs-CSSLP-date    | 100%-Pass-70-383-Exams    | Latest-JN0-360-real-exam-questions    | 100%-Pass-4A0-100-Real-Exam-Questions    | Dumps-300-135-exams-date    | Passed-200-105-Tech-Exams    | Latest-Updated-200-310-Exam    | Download-300-070-Exam-PDF    | Hot-Sale-JN0-360-Exam    | 100%-Pass-JN0-360-Exams    | 100%-Pass-JN0-360-Real-Exam-Questions    | Dumps-JN0-360-exams-date    | Exam-Description-1Z0-876-dumps    | Latest-exam-1Z0-876-Dumps    | Dumps-HPE0-Y53-exams-date    | 2017-Latest-HPE0-Y53-Exam    | 100%-Pass-HPE0-Y53-Real-Exam-Questions    | Pass-4A0-100-Exam    | Latest-4A0-100-Questions    | Dumps-98-365-exams-date    | 2017-Latest-98-365-Exam    | 100%-Pass-VCS-254-Exams    | 2017-Latest-VCS-273-Exam    | Dumps-200-355-exams-date    | 2017-Latest-300-320-Exam    | Pass-300-101-Exam    | 100%-Pass-300-115-Exams    |
http://www.portvapes.co.uk/    | http://www.portvapes.co.uk/    |