Discussion Anyone else bored out of their mind due to mainstream CPU market stagnation?

[gdansk]

About how many Zen 5 compute dies per wafer to they get assuming 100% yield?

>850

The neat part is they still get >400 even with a defect density that no one would consider viable for modern high volume manufacturing.

 

[OneEng2]

>850

The neat part is they still get >400 even with a defect density that no one would consider viable for modern high volume manufacturing.

Which is exactly why monolithic design is a dead end for the future. I don't see how a monolithic design can hope to compete with a chiplet design. Seems like AMD figured out a way to get the latency down between tiles, and/or found ways to work around it. Intel needs (really badly) to do the same .... and quick.

 

[DaaQ]

AMD was preparing for the turnaround since 2012. When did Intel even start to prepare? Then add 5 years to that date and check if Intel managed or is in progress to manage that, assuming Intel can match AMD's nimbleness.

AMD was on maybe second iteration of Construction core in 2012, I think they didn't even bring in Keller until after that. They ran the Construction cores all the way into AM4 even.

Point is, Zen wasn't concepted in 2012, AMD had 2 huge debts that were coming due, and everyone thought they were going bankrupt. Many were positive of it. Global Foundries wouldn't be able to deliver, Zen would not be faster than Broadwell. People even thought there was no way they could compete with Broadwell HEDT.

You are trying to strawman me, based off what you remember from AMD in 2012.

Let me help, Bulldozer 2011, Piledriver 2012, then was Steamroller, which never released on desktop, then Carrizo, also a laptop only part.

 

[gdansk]

Point is, Zen wasn't concepted in 2012,

There is an interview with Keller in which he says was brought in by the CEO Rory Read to help build Zen. That was in 2012.

That Zen started in 2012 is also confirmed in Clark's interview in 2021.

Zen: The Beginning​

Ian Cutress: This quarter for AMD is all about the 5 Years of Zen and Ryzen, ever since those press events and first microarchitecture disclosures at Hot Chips in August 2016. Realistically, when did the Zen journey start for you - who were the big names, and were you the lead architect off the bat?

Mike Clark: Well, it started in 2012 for me. We realized we needed to do something different from the Bulldozer line. Jim came in and helped re-organize the team, and I was the Lead Architect. So it's been almost 10 years for me.

For personnel, since we started in 2012, there are so many people, and the team is awesome. I am so thankful that I get to represent the work of so many awesome engineers. Suzanne Plummer was the lead of the Zen team, managing the team, and was just keeping the team together, she was just awesome. Then there’s also Mike Tuuk, Tim Wilkens, Jay Fleischman, Leslie Barnes - all kinds of people that were contributing from all parts of the company to make Zen a success.

So it's kind of funny to say I’ve been working on it since 2012 - if I go back, I still have our HLD (High-Level Design) deck that we did for Zen. You wouldn't believe how different, after taking five years to get something to production, it looks. I mean the bones are still there, you see it, but so many things changed along the way. That's one of the keys of this business - being able to be dynamic and have things change because it's such a long time. But also still be able to deliver a competitive design, it’s pretty amazing. Once in a while, when we were starting up, when the teams were worried or feeling weird about their HLD, I'm the one who turned around and said that ‘this is what Zen was, everything's not going to be perfect coming out HLD, stuff is going to change, and it's going to get better’. So that's the art of this job.

 

[igor_kavinski]

They ran the Construction cores all the way into AM4 even.

WOW. I did NOT know that! https://www.anandtech.com/show/11669/amd-releases-bristol-ridge-to-retail-am4-gets-apus

So if they really wanted, they could release cheap 5900X3D2 with two V-cache dies on AM5! Only trouble is, they are not desperate for money anymore

 

[DrMrLordX]

I'm thinking at least an Intel 3 based performance part that instills the public's confidence in their engineering capabilities.

Backporting a 20a/TSMC N3 design to Intel 3 would be a mistake. Porting Raptor Cove to Intel 3 would be . . . okay-ish? Intel 4 isn't going to see anymore use so there's no point in mentioning it. Intel 3 is just flat-out better.

 

[igor_kavinski]

Backporting a 20a/TSMC N3 design to Intel 3 would be a mistake.

They just need to replace the Skymonts with two Lion Cove cores in this: https://www.intel.com/content/www/u...-12m-cache-up-to-5-30-ghz/specifications.html

and stitch three of them together for a quasi-monolithic 12-core gaming CPU based on high clocking leaky dies.

 

[DrMrLordX]

They just need to replace the Skymonts with two Lion Cove cores in this: https://www.intel.com/content/www/u...-12m-cache-up-to-5-30-ghz/specifications.html

and stitch three of them together for a quasi-monolithic 12-core gaming CPU based on high clocking leaky dies.

Nah that would make less sense than trying to do the same thing on TSMC N3B.

 

[moinmoin]

AMD was on maybe second iteration of Construction core in 2012, I think they didn't even bring in Keller until after that. They ran the Construction cores all the way into AM4 even.

You know that is kind of the point of a turnaround? Trying to live off of what's already there while getting and taking the time needed to work on something that can turn the ship around?

Point is, Zen wasn't concepted in 2012

Rory Read and the board of directors saw in 2011 that AMD's then current roadmap was leading to doom and brought in Mark Papermaster as CTO. He and other IBM veterans on the board then brought on other IBM veterans in different leadership positions, Lisa Su was one of them. Keller came on later as the result of those earlier changes.

You are trying to strawman me, based off what you remember from AMD in 2012.

Let me help, Bulldozer 2011, Piledriver 2012, then was Steamroller, which never released on desktop, then Carrizo, also a laptop only part.

You are taking it wrong. I'm not trying to strawman you, I'm using your post as a peg to turn the spotlight on what needs to happen at Intel for a turnaround and how to easily check whether Intel may be in progress of one or not. Products released are the results of years of setting up structures within the company and planing of roadmaps, then working on specific products to launch.

But your example should still help enlightening others since that pattern should sound very familiar, after all Intel had plenty of mobile only cores. Also the repeat of Construction cores well into AM4 should also sound familiar, after all Intel did the same first with Skylake and then with Alder Lake. Unfortunately I'm still not really seeing Intel's Zen turnaround moment, neither in the leadership structure nor in the products.

 

[Thunder 57]

WOW. I did NOT know that! https://www.anandtech.com/show/11669/amd-releases-bristol-ridge-to-retail-am4-gets-apus

So if they really wanted, they could release cheap 5900X3D2 with two V-cache dies on AM5! Only trouble is, they are not desperate for money anymore

Bristol Ridge was surprisingly good for a construction core. As for a 5900X3D2 on AM5, I don't know where you get your ideas. Some of them are just bonkers. I don't know if you are joking or not with all of them, but I think you are serious more often than not.

 

[DrMrLordX]

Steamroller, which never released on desktop

Just noticed this remark. I had a 7700k and a 7870k, both Steamroller APUs, in a desktop motherboard with desktop form-factor (FM2+). Not sure if that's really the focus of this topic per se, but there you have it.

 

[igor_kavinski]

but I think you are serious more often than not.

Let me re-iterate: In a world that has seen Pentium 4 and Bulldozer, I see no reason why companies shouldn't take risks. 5900X3D2 (prototype exists!) on AM5 wouldn't be any less crazy than consumer Zen 5 with massive FPU with no way to properly feed it (AVX-512 needs insane RAM bandwidth) or Arrow Lake that should never have seen the light of day with its myriad of unbalanced performance related issues. Better that AMD create weird products while they are financially healthy rather than introduce something crazy when they are down on their luck.

 

[GTracing]

Let me re-iterate: In a world that has seen Pentium 4 and Bulldozer, I see no reason why companies shouldn't take risks. 5900X3D2 (prototype exists!) on AM5 wouldn't be any less crazy than consumer Zen 5 with massive FPU with no way to properly feed it (AVX-512 needs insane RAM bandwidth) or Arrow Lake that should never have seen the light of day with its myriad of unbalanced performance related issues. Better that AMD create weird products while they are financially healthy rather than introduce something crazy when they are down on their luck.

But what's the point of such a product? A 7900X3D is faster in practically every workload, and it's cheaper to manufacture than your hypothetical CPU would be.

 

[igor_kavinski]

But what's the point of such a product?

Assuming it is financially viable (not more expensive than the 7900X3D), we let the market decide whether it wants it or not. Release it as a limited edition like the 5600X3D.

 

[Thunder 57]

Let me re-iterate: In a world that has seen Pentium 4 and Bulldozer, I see no reason why companies shouldn't take risks. 5900X3D2 (prototype exists!) on AM5 wouldn't be any less crazy than consumer Zen 5 with massive FPU with no way to properly feed it (AVX-512 needs insane RAM bandwidth) or Arrow Lake that should never have seen the light of day with its myriad of unbalanced performance related issues. Better that AMD create weird products while they are financially healthy rather than introduce something crazy when they are down on their luck.

It was the first X3D demonstrated, but on AM4. Why spend time and resources redoing it for AM5?

 

[igor_kavinski]

Why spend time and resources redoing it for AM5?

To give some new engineers experience? Maybe it's as simple as pairing the Zen 5 IOD with the older compute dies (Zen3+ exists)? Maybe we are underestimating AMD engineers and they can build like Legos since that's what their focus has been, unlike Intel who prefer the moronic and expensive way of doing things? Maybe they have surplus Zen 3 dies and first gen V-cache?

 

[Thunder 57]

To give some new engineers experience? Maybe it's as simple as pairing the Zen 5 IOD with the older compute dies (Zen3+ exists)? Maybe we are underestimating AMD engineers and they can build like Legos since that's what their focus has been, unlike Intel who prefer the moronic and expensive way of doing things? Maybe they have surplus Zen 3 dies and first gen V-cache?

AM4 still sells well enough so need need to put Zen 3 on AM5. Besides Zen 3 is over four years and two generations old. Just seems like a waste of resources.

 

[igor_kavinski]

Just seems like a waste of resources.

Tell that to AMD. Someone I know got a 7520U laptop manufactured in July 2024. I was like, there should be laws against this deception. Clearly advertise laptop as, "Disclaimer: CPU will run 99% of x86 software but it's based on 2019 technology".

 

[OneEng2]

Every new product incurs a silly amount of NRE and cost to the company in time, money, and resources that could be used elsewhere.

For these reasons, and some technical ones which I have outlined already, things like going back to an old node and designing a new processor revision on it, or moving back to a monolithic approach, are all examples of really bad strategic decisions.

These kinds of bad decisions can result in companies failing completely if this thinking continues too long.

You know that is kind of the point of a turnaround? Trying to live off of what's already there while getting and taking the time needed to work on something that can turn the ship around?


Rory Read and the board of directors saw in 2011 that AMD's then current roadmap was leading to doom and brought in Mark Papermaster as CTO. He and other IBM veterans on the board then brought on other IBM veterans in different leadership positions, Lisa Su was one of them. Keller came on later as the result of those earlier changes.


You are taking it wrong. I'm not trying to strawman you, I'm using your post as a peg to turn the spotlight on what needs to happen at Intel for a turnaround and how to easily check whether Intel may be in progress of one or not. Products released are the results of years of setting up structures within the company and planing of roadmaps, then working on specific products to launch.

But your example should still help enlightening others since that pattern should sound very familiar, after all Intel had plenty of mobile only cores. Also the repeat of Construction cores well into AM4 should also sound familiar, after all Intel did the same first with Skylake and then with Alder Lake. Unfortunately I'm still not really seeing Intel's Zen turnaround moment, neither in the leadership structure nor in the products.

This is exactly correct. These patterns at Intel show a company that has failed to identify key strategic moves, has switched gears painfully only to switch back, and then looks to switch gears again.

Eventually, your number is up. You can only fool around with the financial reports so long before it catches up to you and the bad decision making comes home to roost.

I personally believe Intel has done SOME really correct strategic thinking ..... albeit mixed in with more bad thinking that may well derail any good the "correct" thinking might have done.

Aiming for lower power and more efficient cores .... good idea.

Moving away from monolithic design ..... good idea.

Developing EUV capabilities (finally)..... good idea.

Developing 18A with GAA .... good idea.

Now .... some questionable decisions (recent).

Running parallel development paths for an external fab and an internal fab strategy at the same time ..... very expensive and likely to result in doing neither one as good as you needed to..... bad idea.

Sticking with vertically integrated strategy without regard to the business case associated with exponentially increasing costs of new nodes...... bad idea.

18A with no variant for GAA only (must also include BSPDN) increasing risks and limiting the attractiveness for outside organizations to their fab process.... foolish IMO.

Releasing Arrow Lake with hideous L3 and ring bus latency .... seriously? It is sad to imagine what a crazy good chip Arrow lake could have been. It does alot of things really well!

 

[DaaQ]

There is an interview with Keller in which he says was brought in by the CEO Rory Read to help build Zen. That was in 2012.

That Zen started in 2012 is also confirmed in Clark's interview in 2021.

Ok you got me, Keller didn't come on until August of 2012.

Ian Cutress interview

Zen: The Beginning
IC: This quarter for AMD is all about the 5 Years of Zen and Ryzen, ever since those press events and first microarchitecture disclosures at Hot Chips in August 2016. Realistically, when did the Zen journey start for you - who were the big names, and were you the lead architect off the bat?

MC: Well, it started in 2012 for me. We realized we needed to do something different from the Bulldozer line. Jim came in and helped re-organize the team, and I was the Lead Architect. So it's been almost 10 years for me.

You know that is kind of the point of a turnaround? Trying to live off of what's already there while getting and taking the time needed to work on something that can turn the ship around?


Rory Read and the board of directors saw in 2011 that AMD's then current roadmap was leading to doom and brought in Mark Papermaster as CTO. He and other IBM veterans on the board then brought on other IBM veterans in different leadership positions, Lisa Su was one of them. Keller came on later as the result of those earlier changes.


You are taking it wrong. I'm not trying to strawman you, I'm using your post as a peg to turn the spotlight on what needs to happen at Intel for a turnaround and how to easily check whether Intel may be in progress of one or not. Products released are the results of years of setting up structures within the company and planing of roadmaps, then working on specific products to launch.

But your example should still help enlightening others since that pattern should sound very familiar, after all Intel had plenty of mobile only cores. Also the repeat of Construction cores well into AM4 should also sound familiar, after all Intel did the same first with Skylake and then with Alder Lake. Unfortunately I'm still not really seeing Intel's Zen turnaround moment, neither in the leadership structure nor in the products.

I agree with second half of your statement. Also that you may not have tried strawing me, but 2012 is 4 quarters,
Keller came in August 2012.

EDIT: @moinmoin I am thinking its more semantics now. Thank you for the Cutress article, long but worthy read.

Also, wasn't Skylake the second 14nm? I honestly can't remember.

From the Papermaster interview.

MC: It definitely takes a big investment because of the long timeframe. With that long lead time, it's tough for the business - the market wants a product every year, and you keep trying to refresh, waiting for the new big thing to come. So it was definitely necessary so that we could do what we needed to do to get the job done.

It was a tough time. I mean, one of the hardest problems we had was holding the team together. A lot of people did leave, and it was a very aggressive programme. From where we were, we spent a lot of time both trying to convince people that we would succeed.

Just noticed this remark. I had a 7700k and a 7870k, both Steamroller APUs, in a desktop motherboard with desktop form-factor (FM2+). Not sure if that's really the focus of this topic per se, but there you have it.

These were all APUs A10 - 7700k and and A10- 7850k. Some were released in FM2+ sockets, but all were Kaveri parts which IIRC would be going up against Haswell or the Ivy Bridge tick.

But they never released as high performance parts, AM3+ FX parts. They je bated it with the FX 9590 and I think 9370 FX parts.

 

[Thunder 57]

Ok you got me, Keller didn't come on until August of 2012.

Ian Cutress interview





I agree with second half of your statement. Also that you may not have tried strawing me, but 2012 is 4 quarters,
Keller came in August 2012.

EDIT: @moinmoin I am thinking its more semantics now. Thank you for the Cutress article, long but worthy read.

Also, wasn't Skylake the second 14nm? I honestly can't remember.

From the Papermaster interview.




These were all APUs A10 - 7700k and and A10- 7850k. Some were released in FM2+ sockets, but all were Kaveri parts which IIRC would be going up against Haswell or the Ivy Bridge tick.

But they never released as high performance parts, AM3+ FX parts. They je bated it with the FX 9590 and I think 9370 FX parts.

Broadwell was the first 14nm but it was near non existent on desktop. I don't remember Stemroller or Excavator being relevant on FM2+. If only they had something like Bristol Ridge for AM3+ back then.

 

[igor_kavinski]

https://videocardz.com/newz/intel-nova-lake-preliminary-desktop-specs-list-52-cores-16p32e4lp-configuration

This monster is the way forward.

Intel needs to make this happen so AMD can also respond in kind.

 

[MS_AT]

https://videocardz.com/newz/intel-nova-lake-preliminary-desktop-specs-list-52-cores-16p32e4lp-configuration

This monster is the way forward.

It's too early to get exited they may axe it still And without sufficient bandwidth all those cores will go to waste anyway. It will be reverse of the situation with Sapphire Rapids HBM, which has much more bandwidth than the cores can use, here it will be much more cores than BW to sustain them. I would prefer they catch up to Apple in ST than double the cores to shine in Cinebench.

 

[igor_kavinski]

And without sufficient bandwidth all those cores will go to waste anyway.

DDR6-10000 CUDIMMs

And maybe high speed 512MB eDRAM cache in the I/O tile to feed the beast continuously without incurring the higher latency penalty of DDR6 (timings gonna be 80-95-95-150).

 

[Thunder 57]

https://videocardz.com/newz/intel-nova-lake-preliminary-desktop-specs-list-52-cores-16p32e4lp-configuration

This monster is the way forward.

Intel needs to make this happen so AMD can also respond in kind.

That monster would surely have a monsterous TDP and MSRP.