Speculation: Ryzen 4000 series/Zen 3

[Richie Rich]

Why are we talking about Apple on a Ryzen 4000 thread ?

Simply because AMD's Zen technology leap comes from Jim Keller. Interesting thing about Keller is that he created in AMD an available technology table (IMHO something like what performance can bring at what cost/devtime). The same he did in Intel. I'm wonder if Apple have same tech map. Probably yes. And because first 6xALU CPU (Apple A11 Monsoon core) development was started around 2012 (entered market in 2017) and Keller moved to AMD at 2012 this means Keller saw Apple's technology map and CPU road map too. In other words he knew about 6xALU CPU will be developed in Apple soon and what tech it consists of. So I suppose his AMD K12 project, two CPUs (x86 and ARM) based on common back-end, contained also 6xALUs and similar advanced tech inside. And because Zen 3 is probably the x86 version of K12 (must be because AMD didn't had a resources for development 3 different uarchs at the same time: Zen2, Zen3 and K12. There were just Zen2 and K12 which ARM branch was killed and K12 x86 become Zen 3) this means Zen 3 will have 6xALU core too. Apple is known for being the infinite game player (I like famous Simon Sinek explanation here) and Keller tried AMD to be transformed into infinite game player too.

The answer has nothing to do with hardware. And everything to do with software installed base. If you came up with a magic CPU that was %500 percent faster clock for clock it would still take 15 or 20 years to make x86 irrelevant. Maybe longer. So, for now and the foreseeable future x86 is where the money is.

How much time did it take for TESLA to change automotive market? 14 years including building two gigafactories on two continents? IT is much faster, especially when you do not need any fabs, just to make an order at TSMC. 500% faster CPU would take major server market share in two years. No company will move his hundreds of servers when Zen 2 is a bit faster than old Intel but still slower than Ice Lake. For 5% IPC you don't want to change supplier, service and continuity (it's not economical at short term, IPC advantage is uncertain in long term). Same parformance for EPYC's half price? That's strong argument definitely but can be eliminated by Intel's discount. But Nuvia with such a huge +80% IPC advantage will make all x86 servers technologically obsolete in just one day (and every other ISAs too). Economically not even worth to be plugged into electricity.

You really going to Geekbench* the number eh?

In the real world it doesn't scale 100%, otherwise there wouldn't be any focus on improving the memory subsystem or the I/O.

Some like HPC, scale even less than that. SpecFP is representative, and the scaling factor is often 60-70%.

(Geekbench scales pretty much 100% with clocks, so do many less realistic benchmarks and also synthetics)

I never talked about Geekbench. I'm using SPEC2006 benchmark results provided by AnandTech. It consists of several sub tests based on some typical algorithms so you can analyze where the performance come from. Definitely the +80% IPC advantage doesn't come from L1 and L2 cache. You can see that machine is fast just everywhere:

Stop with the thread derailing. This is a Ryzen 4000 speculation thread, not an Apple thread, keep your posts about Ryzen or start your own thread


esquared
Anandtech Forum Director

 

[DrMrLordX]

Simply because AMD's Zen technology leap comes from Jim Keller.

That Mike Clark guy doesn't seem to count for much, eh?

 

[IntelUser2000]

I never talked about Geekbench. I'm using SPEC2006 benchmark results provided by AnandTech.:

I see you didn't get it.

"Geekbench" as in you are treating real world applications as if it were Geekbench.

SpecCPU don't scale 100%. I replied to you before with the reasons why the 82% number is wrong.

 

[Panino Manino]

That Mike Clark guy doesn't seem to count for much, eh?

I see this being discussed more than one time, that people are giving Keller much more credit than he deserves, that's he is more responsible for the IF.
I imagine that it was an advantage for AMD that he came from Apple and knew things, but in the end the one that makes the decisions about what AMD will actually do is Lisa, right? No matter what he wanted to do, Lisa's vision for the future of AMD takes priority.

 

[itsmydamnation]

Its just dumb rockstar fallacy stuff . People like Jim Keller are managers of managers of managers. There job is to set the focus, drive the people/program processes, manage the high level targets. Even people like Mike Clark have teams of high skilled architects doing the work, coming up with the idea's. This one person stuff is so juvenile you can tell which people have never worked on big technology programs/projects.

Upper layers of management (people/technical/etc) have a massive effect on the effectiveness of teams to deliver products to specifications but they don't do the work or come up with the idea's , they have far to much to manage to focus on the millions and millions of specifics. People at that level involving themselves in specifics (outside of the need for specific issue resolution/decision) are bad managers, I think we can see from Jim illustrious record over many high level, high importance roles that he is not a bad manager! Keeping everything moving in the same direction on the same time scale, make the choices to add or cull, triaging issues and resources, that stuff over millions of man hours of effort is the really hard stuff.

The other dumb thing @Richie Rich statements is the only really new thing in the Zen Core at a functional block level is the Uop cache which Apple didn't do, everything else has a very clear AMD legacy just look at the FPU and how FMA4 works, how the SMU is a straight evolution of excavator, the ALU latencies, AGU work like they always did, seperate FPU etc . What changed was the target and focus.

 

[Thunder 57]

This is very interesting part from AMD's CEO interview:

AMD at CES 2020: Q&A with Dr. Lisa Su

www.anandtech.com

..."Yeah we did big mistake to shut down K12 project and let Mr. Keller leave."

If AMD put all their resources into K12 they would probably be bankrupt right now.

Hopefully they cancelled K12 ARM branch only (6xALU core+SMT) and its x86 branch will be Zen 3 (both based on same back-end core so Zen 3 is 6xALU core too).

I don't know where you get this crap. There was never any serious K12 leaks AFAIK, because it never got far enough along.

...slower more than 40% (Bulldozer era) means you are going to bankruptcy. Wouldn't you as CEO be a little nervous when somebody has CPU with +80% IPC advantage? More than double what you consider as bankruptcy level...

That's why Zen has to be great. AMD was in deep trouble. If they put resources into K12 and delayed/killed Zen, who would be buying K12? If they even got it out? I can't imagine many investors would be willing to invest into that strategy.

If you think about what we did in desktop, 1st Gen Ryzen was good, 2nd Gen Ryzen was better, and now you look at 3rd Gen Ryzen today, if you go to Amazon, AMD has 12 out of the top 12 best-selling desktop processors

I can quote interviews too. AMD is finally making money, all thanks to Zen. How many K12 CPU's would be best sellers? None.

Lisa Su could be calm only under condition of Zen 3 is Keller's K12 (x86 branch) with 6xALUs. If this is true a lot of people will be surprised by Zen 3 performance. Something like those leaks about 40-50% IPC uplift is true, but it's uplift everywhere (int + FPU) and not just FPU.

Ha! Zen was a unicorn in that it did 52%, largely because BD was so abysmal. You really think AMD is going to pull another rabbit out of its hat just that easily? Even if they could, all previous Zen inventory would be obsolete and have to be sold for peanuts or written off. Solid business strategy right there.

Simply because AMD's Zen technology leap comes from Jim Keller. Interesting thing about Keller is that he created in AMD an available technology table (blah blah blah) Apple is known for being the infinite game player (I like famous Simon Sinek explanation here) and Keller tried AMD to be transformed into infinite game player too

Jim Keller is human. He is not Midas. Everything he touches does not turn to gold. The guy is very talented but he has been sensationalized to the point of being annoying. Besides, since he is now at Intel, shouldn't they be worrying about them instead? Well, they should be, since they are the only real competitor in x86. Why should AMD push ARM and give ARM more credibility? it is not in their best interests at this point, if ever.

 

[A///]

Why are we talking about Apple on a Ryzen 4000 thread ?

At least we're past SMT4. I hope.

 

[soresu]

At least we're past SMT4. I hope.

Aaarrgghh speak not the word! It's like Candyman - say it 3 times and someone appears to praise it....

 

[DrMrLordX]

I see this being discussed more than one time, that people are giving Keller much more credit than he deserves, that's he is more responsible for the IF.

That's basically correct. Keller did a good job of pulling together a team at a company that was struggling and keeping it focused long enough to produce a competitive product on a "shoestring" budget. Note the quotation marks.

I imagine that it was an advantage for AMD that he came from Apple and knew things, but in the end the one that makes the decisions about what AMD will actually do is Lisa, right? No matter what he wanted to do, Lisa's vision for the future of AMD takes priority.

Ultimately, yes, though it's also true that the team members themselves likely contributed more to the Zen design than we'll ever know. It's telling that Keller was mostly gone from AMD by the time serious work on Matisse/Rome began. Those who stayed behind at AMD have kept the momentum going. Clearly they have some talent left.

Aaarrgghh speak not the word! It's like Candyman - say it 3 times and someone appears to praise it....

Funny, that's usually how I react when people bring up Internet Strongman Juan-

whoa

almost said it.

 

[moinmoin]

Mark Papermaster is the true MVP. If not even Nick Donofrio.

 

[kapulek]

https://twitter.com/x/status/1217021504560320514

Kapulek, you must post some form of commentary on a link.

Daveybrat
AT Moderator

 

[lobz]

Why are we talking about Apple on a Ryzen 4000 thread ?

Edit: nevermind, I was being redundant.

 

[Ajay]

Mark Papermaster is the true MVP. If not even Nick Donofrio.

There's no i in Team. It took an outstanding team, with leadership, vision, quality management, excellent execution, etc.,etc. AMD already had the raw talent -it took managers from Clark and Keller up to Papermaster and Su to get that team on track an produce the outcome we have today. IMO.

 

[Markfw]

There's no i in Team. It took an outstanding team, with leadership, vision, quality management, excellent execution, etc.,etc. AMD already had the raw talent -it took managers from Clark and Keller up to Papermaster and Su to get that team on track an produce the outcome we have today. IMO.

I definitly have to agree there. Before I retired, I could have done 5 times the work I was doing (IT) if I had had management support for policies that I knew should have been enacted. Up to the senior VP level even, maybe all the way at the top. I was constantly help back by stupid decisions from managers who did not know what they were doing. My biggest problem was the senior VP.

 

[Atari2600]

I definitly have to agree there. Before I retired, I could have done 5 times the work I was doing (IT) if I had had management support for policies that I knew should have been enacted. Up to the senior VP level even, maybe all the way at the top. I was constantly help back by stupid decisions from managers who did not know what they were doing. My biggest problem was the senior VP.

Thats why they are called managers...

We "manage" to get work done despite them.

 

[moinmoin]

There's no i in Team. It took an outstanding team, with leadership, vision, quality management, excellent execution, etc.,etc. AMD already had the raw talent -it took managers from Clark and Keller up to Papermaster and Su to get that team on track an produce the outcome we have today. IMO.

That's exactly why I suggested Papermaster. He joined AMD in 2011 and essentially assembled all the significant teams (like the one that led to Zen) and did all the significant recruitment and hires since (including Keller and Koduri, likely other IBM alumni like Su), changing corporate culture on the way. And Donofrio, an IBM alumnus then on AMD's board, recruited Papermaster.

 

[CHADBOGA]

That's exactly why I suggested Papermaster. He joined AMD in 2011 and essentially assembled all the significant teams (like the one that led to Zen) and did all the significant recruitment and hires since (including Keller and Koduri, likely other IBM alumni like Su), changing corporate culture on the way. And Donofrio, an IBM alumnus then on AMD's board, recruited Papermaster.

What will be interesting is if and when AMD are in a position to strike out and take on some of the special projects that Intel does and keeps failing at.

If AMD were able to succeed in an area outside of their core competency where Intel has failed or is failing in.

 

[Ajay]

What will be interesting is if and when AMD are in a position to strike out and take on some of the special projects that Intel does and keeps failing at.

If AMD were able to succeed in an area outside of their core competency where Intel has failed or is failing in.

I think that would be a terrible thing for AMD to do. They should stick to what they are currently doing and get even better at doing that before entertaining even adjunct projects - IMHO.

 

[CHADBOGA]

I think that would be a terrible thing for AMD to do. They should stick to what they are currently doing and get even better at doing that before entertaining even adjunct projects - IMHO.

I'm not saying they should be rushing into doing stuff outside of core competency, but if things continue to go well for them, I think it is not unrealistic for them to slightly branch out in 3 or so years time.

 

[lobz]

That's exactly why I suggested Papermaster. He joined AMD in 2011 and essentially assembled all the significant teams (like the one that led to Zen) and did all the significant recruitment and hires since (including Keller and Koduri, likely other IBM alumni like Su), changing corporate culture on the way. And Donofrio, an IBM alumnus then on AMD's board, recruited Papermaster.

But... what if he had a team that helped him do all that?

 

[Atari2600]

I'm not saying they should be rushing into doing stuff outside of core competency, but if things continue to go well for them, I think it is not unrealistic for them to slightly branch out in 3 or so years time.

Yep - it definitely is something they should be looking at.

A good place to start would be FPGA co-processor that can hook into infinity fabric.

So a 2P EYPC motherboard can become a single EPYC CPU and a co-processor (like the old Torrenza initiative - that was a brilliant idea which never gained traction - too early for the market maybe)

 

[maddie]

I'm not saying they should be rushing into doing stuff outside of core competency, but if things continue to go well for them, I think it is not unrealistic for them to slightly branch out in 3 or so years time.

Yep - it definitely is something they should be looking at.

A good place to start would be FPGA co-processor that can hook into infinity fabric.

So a 2P EYPC motherboard can become a single EPYC CPU and a co-processor (like the old Torrenza initiative - that was a brilliant idea which never gained traction - too early for the market maybe)

The chiplet layout seems to be a low risk way to try these things. ML especially is definitely going to be huge. They can even subcontract a chiplet as an extreme example.

 

[Atari2600]

I was thinking more a solution that can go onto a separate socket, rather than be within the one package.

That'd allow more flexibility and allow AMD to bring in 3rd parties to the ecosystem easier.

For example, the Xilinx Alveo U280 is supposed to support CCIX on AMD EYPC already, so thats a 3rd party on PCIe3...

Looking down the line to the Zen4 timeframe, a PCIe5x16 card that could access main memory via CCIX could allow use of system memory - a massive boon for co-processing.

At the same time, a future SP4 socket version of the Xilinx may reduce latency by a significant degree and allow much easier integration of accelerators to the SP4 platform.

Instead of 3rd parties doing FPGA, what if AMD decided to make an FPU heavy solution as an alternative to AVX512? Zen4 CPU in one socket and the FPU-accelerator in the other...

 

[Ajay]

I definitly have to agree there. Before I retired, I could have done 5 times the work I was doing (IT) if I had had management support for policies that I knew should have been enacted. Up to the senior VP level even, maybe all the way at the top. I was constantly help back by stupid decisions from managers who did not know what they were doing. My biggest problem was the senior VP.

Been there done that. "We want you to innovate", then they find out innovation costs money and/or manpower. Plan rejected - dumb asses.

 

[Markfw]

Been there done that. "We want you to innovate", then they find out innovation costs money and/or manpower. Plan rejected - dumb asses.

Oh, worse than that. My suggestions would have saved money. Circa 2003, Opterons used less power than P4 servers, and cost less, and less for AC. But the response was "we only use real CPUs, Only Intel"