AMD’s Zen 3 Processors May Crack 5 GHz Barrier

[Tsing]

Image: AMD



Igor Wallossek has returned with two new OPNs for AMD’s upcoming Ryzen 9 4950X/5950X processor. Yesterday’s report suggested that this Zen 3-based SKU would boost to 4.8 GHz, which is already admirable, but one of the new OPNs alludes to an even better max boost clock: 4.9 GHz. Needless to say, this has gotten Ryzen fans wondering whether AMD has finally discovered the secret sauce to attaining that magical 5 GHz figure.



“The faster of the two samples now boosts to 4.9 GHz, while the base clock is set to 3.7 GHz,” writes Wallossek. “In contrast to the previous ES, this is another...

Continue reading...

 

[Burticus]

Call me when all cores 5ghz, thx

 

[Grimlakin]

Focus on IPC. Get that better and maintain your current GHZ rating and nobody that makes decisions for larger orders will care.

 

[Dan_D]

I'll believe it when I see it. Ryzen CPU architectures have traditionally not clocked all that high. I'm not expecting that to change anytime soon.

 

[Riccochet]

Don't know why people focus solely on Ghz. It wasn't Ghz that put AMD in the lead during the Athlon64 vs P4 days.

 

[Dan_D]

Don't know why people focus solely on Ghz. It wasn't Ghz that put AMD in the lead during the Athlon64 vs P4 days.

The concern is that AMD's IPC advantage isn't enough to offset the fact that Intel has a considerable clock speed advantage. This makes Intel CPU's stronger in single-threaded or some lightly threaded applications. There are other reasons, but this is why Intel dominates on the gaming front. It's why I wouldn't hold out any hope that AMD's Ryzen 4000 series would close that gap significantly over Ryzen 3000 series CPU's.

 

[SmokeRngs]

The concern is that AMD's IPC advantage isn't enough to offset the fact that Intel has a considerable clock speed advantage. This makes Intel CPU's stronger in single-threaded or some lightly threaded applications. There are other reasons, but this is why Intel dominates on the gaming front. It's why I wouldn't hold out any hope that AMD's Ryzen 4000 series would close that gap significantly over Ryzen 3000 series CPU's.

I'm wondering how long this is going to be in effect for, though? I'm willing to bet the only reason Intel has been able to keep up clock speeds is due to still being on 14nm. I think it's quite likely that when Intel finally gets something desktop usable on 10nm that the clock speeds are going to drop. Intel would need a rather large IPC increase to make up for that, especially if clocks take a beating and can't match AMD's.

 

[Dan_D]

I'm wondering how long this is going to be in effect for, though? I'm willing to bet the only reason Intel has been able to keep up clock speeds is due to still being on 14nm. I think it's quite likely that when Intel finally gets something desktop usable on 10nm that the clock speeds are going to drop. Intel would need a rather large IPC increase to make up for that, especially if clocks take a beating and can't match AMD's.

Well, we can't say for sure how a given architecture will clock without knowing anything about it. Being built on a smaller process node doesn't guarantee the clocks will have to drop. However, it's very likely that they will if said architecture is another Skylake descendant. Unless Intel has a totally new architecture that makes a massive leap forward in IPC, clock speeds, or both while allowing for comparable (or greater) core density. If Intel's next desktop architecture ends up being a 10nm die shrink of a Skylake variant, then yes, its very likely that Intel's clock speed advantage will be minimized or even lost. Intel needs to either improve IPC dramatically, (as you said) or maintain its current clocks and IPC, while increasing core density.

I just don't think that's in the cards for a theoretical 10nm Skylake variant. At 10nm, core density should increase, but Intel will still be in a similar position as it is today without doing more. It needs chiplet based designs. That gives them far more flexibility to meet customer demands while improving yields. It needs a more efficient and smaller process node for manufacturing. It needs an architecture that actually moves forward in terms of IPC without sacrificing too much elsewhere. It's a tall order.

What Intel really needs, is another Core 2 Duo type of release. These days, that would be really hard to achieve given how competitive AMD has become in the last three years and given how far behind Intel is right now. Intel used to be the leader in the semi-conductor manufacturing business, but it's now behind TSMC, Samsung and so on. If AMD stays on track likes it seems to think it will, we may not see a competitive Intel for another 3-5 years.

 

[Brian_B]

Now, AMD didn't quite have the same level of resources, but how long did it take them to recover from Bulldozer? 5+ years between FX and Zen releases.

A bit beyond that, how did Intel dig itself out of the Netburst hole? It drummed up a significant marketing campaign (doo-doo-DOO and Blue Man Group, etc.), and a bit of luck that AMD was having issues with it's foundry spinoff / relationship with GF, and Intel was able to maintain their status as the "premium" brand even though they didn't have the all around price/performance balance as the competitor. Intel was able to successfully paint AMD as a the "budget knockoff" brand, and push Ghz as a marketing point.

Again, that Netburst - Core period was about 5-6 years.

So yeah, Intel's strongest advantage was always that they had one of the best, if not the best, production process to rely upon. Foundries have moved ahead of them, and their engineering isn't strong enough to close that gap any longer.

The Intel 14 nm process was introduced in CPUs with Skylake, released in 2015. We are still seeing releases of chips based on that overall architecture and process node, so we are already more than 5 years out of what has traditionally been the period between significant generations of CPU architectures, and Intel still doesn't quite have 10nm ready for prime time, and just announced that 7nm would be delayed beyond that.

A couple of things could really break open, so I don't necessarily write off Intel:

Intel could just jump over to these other Foundries, and that's exactly what we just heard them announce. That eliminates the process node advantage that AMD is enjoying right now, and brings it back to engineering (and marketing, let's be honest, if you can get developers to optimize for your platform). It's not the best thing for Intel, going third party on production will significantly cut into their margin, and they are at the mercy of someone else's production schedule, but it keeps them alive.

Intel has a long, strong relationship with vendors and customers. It will be a long time before some CTO gets fired for buying Intel, so long as they can remain competitive. I expect the Intel marketing machine to significantly pick up, like it was back in the 00's when AMD had Athlon. If AMD could have made enough chips, they may could have pulled ahead as the market leader.

And Intel has dumped a lot of resources into their foundry and fixing the issues. That could bear fruit any time and put Intel back at a significant production advantage. It could also never come around, and Intel could either continue to throw resources at it, or pull off the band aid and just sell off their foundry division.

I think if Intel's foundry woes continue to 7nm - for instance, they aren't able to make that transition by 2022, investors will have more or less lost confidence in Intel's ability to innovate and operate on the production side. I would expect a move to divest the foundry, take all of that R&D and put it back into engineering, and basically do what AMD did back in '08 - go fabless, stop bleeding cash into the production side, lean hard on solid engineering and IP, and attempt to right the ship that way.

 

[Uvilla]

Now, AMD didn't quite have the same level of resources, but how long did it take them to recover from Bulldozer? 5+ years between FX and Zen releases.

A bit beyond that, how did Intel dig itself out of the Netburst hole? It drummed up a significant marketing campaign (doo-doo-DOO and Blue Man Group, etc.), and a bit of luck that AMD was having issues with it's foundry spinoff / relationship with GF, and Intel was able to maintain their status as the "premium" brand even though they didn't have the all around price/performance balance as the competitor. Intel was able to successfully paint AMD as a the "budget knockoff" brand, and push Ghz as a marketing point.

Again, that Netburst - Core period was about 5-6 years.

So yeah, Intel's strongest advantage was always that they had one of the best, if not the best, production process to rely upon. Foundries have moved ahead of them, and their engineering isn't strong enough to close that gap any longer.

The Intel 14 nm process was introduced in CPUs with Skylake, released in 2015. We are still seeing releases of chips based on that overall architecture and process node, so we are already more than 5 years out of what has traditionally been the period between significant generations of CPU architectures, and Intel still doesn't quite have 10nm ready for prime time, and just announced that 7nm would be delayed beyond that.

A couple of things could really break open, so I don't necessarily write off Intel:

Intel could just jump over to these other Foundries, and that's exactly what we just heard them announce. That eliminates the process node advantage that AMD is enjoying right now, and brings it back to engineering (and marketing, let's be honest, if you can get developers to optimize for your platform). It's not the best thing for Intel, going third party on production will significantly cut into their margin, and they are at the mercy of someone else's production schedule, but it keeps them alive.

Intel has a long, strong relationship with vendors and customers. It will be a long time before some CTO gets fired for buying Intel, so long as they can remain competitive. I expect the Intel marketing machine to significantly pick up, like it was back in the 00's when AMD had Athlon. If AMD could have made enough chips, they may could have pulled ahead as the market leader.

And Intel has dumped a lot of resources into their foundry and fixing the issues. That could bear fruit any time and put Intel back at a significant production advantage. It could also never come around, and Intel could either continue to throw resources at it, or pull off the band aid and just sell off their foundry division.

I think if Intel's foundry woes continue to 7nm - for instance, they aren't able to make that transition by 2022, investors will have more or less lost confidence in Intel's ability to innovate and operate on the production side. I would expect a move to divest the foundry, take all of that R&D and put it back into engineering, and basically do what AMD did back in '08 - go fabless, stop bleeding cash into the production side, lean hard on solid engineering and IP, and attempt to right the ship that way.

But what is really going to happen is their sales will show stagnation or a miniscule drop, and they will announce mass layoff 1or 2 quarters after... Without really knowing how things will get affected.. morale , so on.

 

[SmokeRngs]

Well, we can't say for sure how a given architecture will clock without knowing anything about it. Being built on a smaller process node doesn't guarantee the clocks will have to drop. However, it's very likely that they will if said architecture is another Skylake descendant. Unless Intel has a totally new architecture that makes a massive leap forward in IPC, clock speeds, or both while allowing for comparable (or greater) core density. If Intel's next desktop architecture ends up being a 10nm die shrink of a Skylake variant, then yes, its very likely that Intel's clock speed advantage will be minimized or even lost. Intel needs to either improve IPC dramatically, (as you said) or maintain its current clocks and IPC, while increasing core density.

I just don't think that's in the cards for a theoretical 10nm Skylake variant. At 10nm, core density should increase, but Intel will still be in a similar position as it is today without doing more. It needs chiplet based designs. That gives them far more flexibility to meet customer demands while improving yields. It needs a more efficient and smaller process node for manufacturing. It needs an architecture that actually moves forward in terms of IPC without sacrificing too much elsewhere. It's a tall order.

What Intel really needs, is another Core 2 Duo type of release. These days, that would be really hard to achieve given how competitive AMD has become in the last three years and given how far behind Intel is right now. Intel used to be the leader in the semi-conductor manufacturing business, but it's now behind TSMC, Samsung and so on. If AMD stays on track likes it seems to think it will, we may not see a competitive Intel for another 3-5 years.

The reason I brought it up in the first place is because of Intel's 10nm failure. They've been trying to push out designs on 10nm and it has failed rather badly. Even when they do get a working product out of it the yields are bad and they're low clocked and small core count parts. Fixing a broken fab node is a lot more difficult than tweaking or changing an architecture. With Intel's history of trouble with 10nm I have no expectation or faith that there will be anything of real use out of that node no matter what architecture changes they make. At least not on the desktop side of the equation.

Intel does have 10nm desktop releases on the roadmap but I highly doubt those designs will be able to come close to the current clock speeds outside of rare boosting instances many people bash AMD for. The fact that 10nm yields have been bad on low core count and low clock speed parts from what we've seen lends evidence to this being as bad or worse for larger and higher speed parts. This is the reason I mentioned the requirement of having greatly increased IPC. We have no evidence that Intel can provide high frequencies on 10nm and a lot of evidence showing it's not possible on parts the size of CPU dies.

I don't normally speculate on the future of CPUs no matter who is making them simply because I don't know enough about the mechanics or industry. Intel's current predicament is a different story in that all the evidence we have currently shows Intel will not be able to keep clock speeds high once production is moved to 10nm from 14nm.

 

[Uvilla]

So basically Intel will just be starting to deal with clock speed regression, when AMD will have been dealing with it for several years.
If Intel is suffering brain drain that they can't reverse effectively enough they might be and stay behind... A la IBM.

 

[Daniel_Doty]

I guess we will see what happens over the next couple of years. Heck, they are saying the possibility of a 128 core Zen chip in the next couple of years!!! Talk about a World Community Grid cruncher.

None the less, I have to think that Intel is coming with an answer to AMD's chips. Don't know if they will catch up any time soon, but the next couple of years should be interesting to see how it all plays out.

 

[Uvilla]

There is no guarantee whatsoever Intel will catch up.. matter of fact if history is prelude, Intel could very well become the next dying tech dinosaur. Doesn't mean they will just be gone, doesn't mean they won't do anything... They will just not be dominant nor innovation leaders . Fact is the innovative part has been fairly flat for them for quite a while.
Yes it doesn't mean they won't do anything whatsoever.. but I think Intel is in very serious trouble, not a little trouble, as far as ' dominance' and leadership is concerned.
Remember when ryzen started some would say, blah, Intel is so big they just have to get something out of the closet... Well all they had in the closet was accountants and poor leadership. Unless they inject a soul in the company, proper leadership and better employee treatment.. yeah Intel will be around, will be big and will be there, with termites, sometimes brilliant, sometimes the leader here and there,.but nothing what is was... Just there fumbling most things, with some highlights.

 

[Brian_B]

There is no guarantee whatsoever Intel will catch up.. matter of fact if history is prelude, Intel could very well become the next dying tech dinosaur. Doesn't mean they will just be gone, doesn't mean they won't do anything... They will just not be dominant nor innovation leaders . Fact is the innovative part has been fairly flat for them for quite a while.
Yes it doesn't mean they won't do anything whatsoever.. but I think Intel is in very serious trouble, not a little trouble, as far as ' dominance' and leadership is concerned.
Remember when ryzen started some would say, blah, Intel is so big they just have to get something out of the closet... Well all they had in the closet was accountants and poor leadership. Unless they inject a soul in the company, proper leadership and better employee treatment.. yeah Intel will be around, will be big and will be there, with termites, sometimes brilliant, sometimes the leader here and there,.but nothing what is was... Just there fumbling most things, with some highlights.

True.

IBM, Xerox, Sun, Silicon Graphics, Palm, Nokia, etc.

The physical tech companies that didn't get eaten in a buyout or die the Chapter 11 death sort of live on by virtue of their patent income, but are just shadows of the giants they once were.

 

[Uvilla]

I often imagine what goes on at Intel... Its all invention in my mind, but I do because I feel, from reports I read about them for decades now really feel like they created the terrible corporate culture I've experienced in some ways.
I am not in anything tech, but I've been a peon on a very large American company.. a company were everyone was sufficiently competent, but were the company made sure, in so many ways , you are not safe, ever. Your job is not safe, your benefits adjust quite a lot too, take or leave it. Not sure the idea behind this way of running the company was, I suspect is based on a belief that a scared employee was a productive employee and maybe as a bonus employees willing to tear each other too, somehow this being beneficial. By the time I started, this has obviously been going on for a while, it was the culture of the company... It was terrible for the company... Catastrophic is the only term.. long term employees where profoundly demoralized and within a year I was in the process of the same. Do the minimum, do the orders, make a fight about what was the order if performance is being questioned and call it day. Screw the company at every possible turn.. always in little ways, little vindictive actions, you know as a co worker, but management won't. I mean.. during a fire drill you were supposed to like help evacuate or some such.. I had 2 co-workers (which will do the drill and repeat rules like robots ) tell me, yeah bs, if theres a fire here is the nearest exit, I take this door, and disappear, the company doesn't do squat for you, no one cares, I won't , was basically the motto. These employees as experienced as they were for any company should have been essentially work horses, reliable, trusted employees, instead they were demoralized vindictive passive aggressive employees, and believe me it was all directed towards the company as an abstract, and management as target. Something like this Is what I imagine Intel is like inside, based on their basically failures and their seemingly cyclical mass layoffs. Ive post comments along the same lines, never get a dispute nor confirmation, I imagine ndas and such, but there should be plenty of ex intel employees where these things don't matter they are not trade secrets of any sort.

 

[Uvilla]

Oh jeez this was about AMD and 5 ghz... Sorry to help derail the conversation. To that I say cool if it happens, okay if it doesn't. Performance is all that matters, how its achieved is secondary. Also seems they are readying their own big little architecture, I suspect this is for mobile and tablets.. a crack at it once again. If ARM.goes Nvidia and starts closing itself down opportunity grows for x86 in phones and tablets ( arguably is always there)

 

[Dan_D]

The reason I brought it up in the first place is because of Intel's 10nm failure. They've been trying to push out designs on 10nm and it has failed rather badly. Even when they do get a working product out of it the yields are bad and they're low clocked and small core count parts. Fixing a broken fab node is a lot more difficult than tweaking or changing an architecture. With Intel's history of trouble with 10nm I have no expectation or faith that there will be anything of real use out of that node no matter what architecture changes they make. At least not on the desktop side of the equation.

Intel does have 10nm desktop releases on the roadmap but I highly doubt those designs will be able to come close to the current clock speeds outside of rare boosting instances many people bash AMD for. The fact that 10nm yields have been bad on low core count and low clock speed parts from what we've seen lends evidence to this being as bad or worse for larger and higher speed parts. This is the reason I mentioned the requirement of having greatly increased IPC. We have no evidence that Intel can provide high frequencies on 10nm and a lot of evidence showing it's not possible on parts the size of CPU dies.

I don't normally speculate on the future of CPUs no matter who is making them simply because I don't know enough about the mechanics or industry. Intel's current predicament is a different story in that all the evidence we have currently shows Intel will not be able to keep clock speeds high once production is moved to 10nm from 14nm.

Well, the above is highly speculative. There are few reasons to conclude much of the above. It is very likely that Intel will lose some clock speed with a transition to 10nm, but that's not guaranteed, nor is a guarantee that Intel can't bring its IPC up to par or surpass AMD. Variations on existing Skylake descendants would seem to suffer from such things, but Intel could literally be developing concurrent architectures along side it that we don't know about. Core 2 practically came out of no where and was a huge leap forward.

So far, the designs we've seen from Intel on 10nm have been underwhelming, but that doesn't mean that this is going to be the future for all things 10nm from Intel.

So basically Intel will just be starting to deal with clock speed regression, when AMD will have been dealing with it for several years.
If Intel is suffering brain drain that they can't reverse effectively enough they might be and stay behind... A la IBM.

That's speculative, and while it may be the case for some architectures and even for a time, we really can't say how long that will be the case or if that will be the case. Remember, a serious desktop part from Intel is still a long ways off. I wouldn't expect to see anything until early 2022 and that's being overly optimistic on my part.

There is no guarantee whatsoever Intel will catch up.. matter of fact if history is prelude, Intel could very well become the next dying tech dinosaur. Doesn't mean they will just be gone, doesn't mean they won't do anything... They will just not be dominant nor innovation leaders . Fact is the innovative part has been fairly flat for them for quite a while.
Yes it doesn't mean they won't do anything whatsoever.. but I think Intel is in very serious trouble, not a little trouble, as far as ' dominance' and leadership is concerned.

No, there isn't. However, Intel still has quite a few resources and outsourcing chip production could get them back on track. Intel still has some of the largest R&D facilities and budget in the industry, so you can never totally discount them. People made all the same doom and gloom commentary back in the days of the Pentium 4, despite this, Intel came through financially strong. Core 2 came out and surprised everyone. Granted, Intel didn't have the chip shortages it has today, which is all the more reason why I think they'll outsource chip production to other fabs if necessary.

 

[Grimlakin]

With everyone outsourcing chip fabrication it makes me want to get a hundred billion or so together to open a fab.. naaa just kidding I'll Let Bezos open a few then buy your next CPU from Amazon.

 

[Dan_D]

With everyone outsourcing chip fabrication it makes me want to get a hundred billion or so together to open a fab.. naaa just kidding I'll Let Bezos open a few then buy your next CPU from Amazon.

Unfortunately, the cost of updating them is rather extreme. Back in the late 1990's and early 2000's, Intel invested four billion dollars a year in fab upgrades and another 4 billion in R&D each year. Facilities like the Chandler Arizona fab have been in operation for decades and upgraded through each process node the company has used. I believe its a 10nm fab today.