Will Ryzen 6000 CPUs be based on the N6 fab process (Warhol chip) and release in 2021?

[g67575]

Another title could be - What are AMD's plans for 2021?

It looks like TSMC's N6 EUV process may be key to new AMD CPUs in 2021, and possibly even GPUs. Warhol / Zen 3+ processors based on this node may release this year, according to this article:
https://www.kitguru.net/components/cpu/joao-silva/amd-zen-3-zen-4-and-zen-5-details-begin-to-leak/

It's possible this will be the first AM5 / DDR5 platform, this would make sense because in 2021, AMD needs something to compete with both Rocket Lake and Alder Lake (late Q4 or Q1 2022?). Also, the first modules of DDR5 have already been produced, so I think it's definitely doable.

This is mostly relevant to me as I'm still using DDR3 1600Mhz, and some games run poorly (40-50FPS) on low frequency RAM, such as Watchdogs: Legions. So, if AMD releases a AM5 motherboard with DDR5 compatibility in 2021, this will likely be what I choose to upgrade to.

To me, the least speculative part of this information, is the plan to launch N6 fabrication process based CPUs in 2021. According to Wikichip, N6 "volume production (is) expected before the end of the year" (dated January, 2020). Link here:
https://fuse.wikichip.org/news/3227/tsmc-q4-7nm-dominates-revenue-preps-5nm-ramp-6nm-by-eoy/

For a while though, many thought an enhancement to 7nm called 7NP would be used for CPUs in 2021, with 5nm in 2022. Ultimately N6 (EUV) and 7NP (small improvement vs 7N) are both enhancements of 7nm, but I think it makes much more sense to use the more advanced N6 process, partly because there should be more production capacity available (due to less tech companies using this node).

 

[theforce]

Warhol might still be on AM4. Since they are not increasing core count, sticking with the cheaper DDR4 might make more sense. Hopefully, they will use something better than vanilla 7nm. Using a newer GF node for the IO die would also be nice.

 

[g67575]

Warhol might still be on AM4. Since they are not increasing core count, sticking with the cheaper DDR4 might make more sense. Hopefully, they will use something better than vanilla 7nm. Using a newer GF node for the IO die would also be nice.

Yeah, I really doubt they would stick with standard TSMC 7FF for 3 CPU generations in a row. With Rocket Lake already slightly ahead of the Ryzen 5000 series in single core performance, They will probably need an improved node to compete with Alder Lake's significant fab. process shrink (10nm).

 

[g67575]

There's also this too "It’s generally expected that Zen 3 will be the last AM4/DDR4 refresh for AMD and that the company will move to new DDR5 platforms in 2021". Link here:
https://www.extremetech.com/computi...-coming-in-2020-talks-challenges-in-notebooks

 

[jigger]

7nm would be more than fine to keep Intel check. Especially if AMD keep the current socket support.

 

[jigger]

Thinking about it ,AMD could probably bring out a range 14/12nm chips to compete against most of the Intel parts. It might help solve the current semiconductor supply issues.

 

[Quartz]

Paul of Paul's Hardware has reported that the next AMD CPUs are expected to be 5nm and this will relieve the pressure on 7nm fabs so there will be greater GPU availability.

I guess we'll have to wait and see.

 

[g67575]

Paul of Paul's Hardware has reported that the next AMD CPUs are expected to be 5nm and this will relieve the pressure on 7nm fabs so there will be greater GPU availability.

Wouldn't shifting CPU production over to the N6 fab. process also reduce the pressure on the in demand N7 process?

I think because of the EUV lithography, N6 is a more specialised (and probably expensive to develop) product, and this would make it less attractive to many tech companies, who don't necessarily need the extra transistor density that N6 should provide.

I've also read that the design rules for 6N are the same / similar to TSMC 7nm, meaning that AMD should be able to (more easily) port their existing designs to 6N.

 

[humbug]

RDNA3 GPU's 160 CU's Chiplet architecture, 2.5X the performance of the 6900XT. 2022.

Ryzen Warhol (Zen 3+) Q4 2021, 7nm, PCIe 4 so could be AM4 or AM5.

Zen 4, More cores (Upto 96) +30% IPC, DDR5. 2022.

RDNA4 in the works.

AMD= Unstoppable 250 Mph train.

 

[humbug]

Just thinking about that ^^^^ while Nvidia are going to be pushing DLSS hard over the next couple of years AMD look to be moving to Chiplet's and getting the Frame Rates up with just brute force, overwhelming power.

 

[g67575]

I don't believe all this nonsense about Navi 31/ Navi 41 leaks. It's made up, there's no way they could have info about RDNA 3/4 designs at this point.

Also worth nothing that doubling the Compute Unit (CU) count alone wouldn't double performance of an AMD GPU.

Is it likely 5nm GPU single GPU designs are in the works for next gen? I think so.

TSMC 6N fab process is planned for Intel's graphics cards apparently.

 

[humbug]

I don't believe all this nonsense about Navi 31/ Navi 41 leaks. It's made up, there's no way they could have info about RDNA 3/4 designs at this point.

Also worth nothing that doubling the Compute Unit (CU) count alone wouldn't double performance of an AMD GPU.

Is it likely 5nm GPU single GPU designs are in the works for next gen? I think so.

TSMC 6N fab process is planned for Intel's graphics cards apparently.

While, 'Again' this is all speculation. The same thing was said about Big Navi, "no way AMD could double the performance of the 5700XT in one generation" They did that and more.

I wouldn't be so sure, in just 3 years AMD have gone from no where to making CPU's that are more than twice as fast as Intel's best with half the power consumption.


With Chiplets, i think AMD can do it, Intel have no answer to AMD and this is why. Fantastic technology innovation. What Intel calls "Glue" is turning everything on its head. You're next Nvidia.

 

[jigger]

While, 'Again' this is all speculation. The same thing was said about Big Navi, "no way AMD could double the performance of the 5700XT in one generation" They did that and more.

I wouldn't be so sure, in just 3 years AMD have gone from no where to making CPU's that are more than twice as fast as Intel's best with half the power consumption.


With Chiplets, i think AMD can do it, Intel have no answer to AMD and this is why. Fantastic technology innovation. What Intel calls "Glue" is turning everything on its head. You're next Nvidia.

While AMD are Glueing* everything together, Nvidia and Intel just seem to be sniffing Glue.

AMD have a habit of over delivering.

 

[Quartz]

Wouldn't shifting CPU production over to the N6 fab. process also reduce the pressure on the in demand N7 process?

Presumably so, but I expect they're working up to launch. We shall just have to wait and see.

 

[humbug]

While AMD are Glueing* everything together, Nvidia and Intel just seem to be sniffing Glue.

AMD have a habit of over delivering.

They do:

Zen 1 vs Excavator claimed +40% IPC: reality +52%
Zen 2 vs Zen 1 claimed +15% IPC: Reality +15%, higher in games.
Zen 3 vs Zen 2 claimed +15% IPC: Reality +20% and as high as +35% in some games.

RDNA1 vs GCN claimed +50% performance per watt: Achieved.
RDNA2 vs RDNA1 claimed +50% performance per watt, double 5700XT performance: Reality achieved and +65% performance per watt.

With RDNA2 AMD introduced a new cache structure that enables a large GPU to have better performance with a 256Bit Bus than it would have if it had a 384Bit Bus and no cache, saving die space and power.

RDNA3 will probably be much like Zen 2 / 3 with a central IO Die containing the IMC, IO and this new cache, maybe larger, from 128MB to 256MB, two 80 CU core dies, with the larger cache, a small clock bump and a small IPC bump taking each 80 CU die to 1.25 of RNDA2 performance and you have your +250%.

 

[g67575]

Tbh, I can see AMD delaying Warhol until Intel's Alder Lake CPUs (10nm desktop) are available to buy with good availability. So Q1 at the latest. This doesn't answer how they will beat Rocket Lake, though, but maybe the Ryzen 5000 series chips will sell well enough regardless.

Another possibility is, there is no 'Warhol' chip, AMD will attempt to rush to 5nm (EUV) CPUs out to compete with Alder Lake, likely in 2022. The lack of concrete information regarding Warhol, is a bit strange, if it is due to release this year.

Kinda tired of seeing "INSANE... Performance" in every video title from Paul (RGT). Clearly, this guy doesn't do understatement.

 

[humbug]

Tbh, I can see AMD delaying Warhol until Intel's Alder Lake (10nm desktop) is available to buy with good availability. So Q1 at the latest.

Another possibility is, there is no 'Warhol' chip, AMD will attempt to rush to 5nm (EUV) CPUs out to compete with Alder Lake, likely in 2022.

Warhol is not the main thing here, i think its a filler reaction to Rocket Lake, i think it will be much the same as Zen+ (Ryzen 2000) A few architectural optimisations they found since launch and a small clock bump, It may even be a test platform for the new IO die, +10% overall performance over Zen 3, on the same AM4 board and 7nm, its low hanging fruit picked to spoil Intel's Rocket Lake party.

Zen 4, new node, new socked, DDR5 6 months later. < That is the Alder Lake competitor.

 

[jigger]

Tbh, I can see AMD delaying Warhol until Intel's Alder Lake CPUs (10nm desktop) are available to buy with good availability. So Q1 at the latest.

Another possibility is, there is no 'Warhol' chip, AMD will attempt to rush to 5nm (EUV) CPUs out to compete with Alder Lake, likely in 2022.

TBH, I can see Intel struggling to grapple with AMD’s APU lines. AMD will just keep pulling miles ahead in every other sector. People should probably keep in mind that although AMD are crushing it with everything they bring to market, they are tiny in business terms compared to Intel. If anything AMD will want to keep pushing.

 

[g67575]

I think Warhol will be significant if utilizes the 6N Fab process. If not, then I agree that it will probably only be a small performance improvement over the Ryzen 5000 series. If DDR5 is ready, can't see a reason for AMD not to release it along with Warhol.

EDIT - It's possible AMD could introduce an AM4+ socket, with support for both DDR5 and DDR4. Apparently, DDR5 has the same 288 pin count as DDR4. That would be a pretty safe option in my opinion (and would give customers lots of choice), then design AM5 just for Zen 4 / future CPUs.

This would be similar to AM3, which had support on some motherboards for both DDR3 and DDR2.

In terms of overall performance improvement, I think it's likely to be another 10% or more vs the 5000 series, based on previous Zen generations - Even the 2700X managed about 10% single core performance vs the 1600X (a bit more vs the Ryzen 1600).

 

[humbug]

Yea could be AM4+ with a Zen 3 DDR5 version. all that would require is a change in the IO die, like i said it could be a test bed for a new IO die, maybe that's it.