Question How ahead is Intel in CPU design compared to AMD?

[Adonisds]

If we could get both companies to design a CPU core now on the same process and using the same number of transistors, who would be ahead and by how much? I'm assuming Intel would be ahead because Skylake has a similar IPC to Zen 2, but Zen 2 uses a more dense node, and Skylake is from 2015 while Zen 2 is from 2019.

The both companies using the same process scenario is just to illustrate the problem and I know it will never happen, but maybe there is a smart way to compare current CPU designs and find out who is ahead and by how much. Wikichip has die sizes and die pictures of Sunny Cove and Zen 2, but I'm not sure how to interpret and compare them.

Another question: If Intel had to design future cores using the same process and the same die size as Skylake, how much further could the Skylake design be improved? Also assuming it will have to have to be x86 too and have all the same functions.

 

[Markfw]

As expected, AMD fans do not like this question.

I think if Intel CPUs were made on the same 7nm process and re engineered to take advantage of it, they would be 15-20% faster than their current 14nm.

15-20% in the server world would mean they are only 80% slower than Rome. Below is from the Phoronix test suite for Rome CPU's. A single 7742 beats 2 of Intels Flagship Platinum 8280's ?

And I don't think it has anything to do with AMD fans. Intel screwed up, and AMD is faster, way faster in the server space and HEDT. And why do you say AMD fans do not like the question ? I don't even see where that applies ?

And your whole statement that "If Intel were made on..." is just silly. I would easily say "If AMD were made on 5 nm they would be 20% faster than they are"

If I had a dollar for every stupid thing said on the internet, I would be a trillionare.

 

[tomatosummit]

As expected, AMD fans do not like this question.

I think if Intel CPUs were made on the same 7nm process and re engineered to take advantage of it, they would be 15-20% faster than their current 14nm.

It depends on what metrics you're talking.
The original leaks had 48cores on the 10nm process and assuming power draw wasn't broken and the expeced IPC gain it my guesses it would be close to Rome. But that dream died a few years ago.
The current expectation is 38cores on a problematic process that isn't going to turn any heads, guessing again at a 50% increase over their current parts coming a year late.

 

[jpiniero]

It depends on what metrics you're talking.
The original leaks had 48cores on the 10nm process and assuming power draw wasn't broken and the expeced IPC gain it my guesses it would be close to Rome. But that dream died a few years ago.

I think you are thinking Cooper. 38 was the original intent for Icelake Server. Cooper (Whitley) has two dies with up to 28 cores each. It's not entirely clear if Intel will actually release a 56 core Cooper model, or if OEMs will even bother since they hate water cooling.

 

[uzzi38]

That will be . . . weird. Rocket Lake-SP?

Nah, still Cooper. Seeing as they refreshed Cascade Lake, I wouldn't be all that surprised if Cooper also got a refresh.

As expected, AMD fans do not like this question.

I think if Intel CPUs were made on the same 7nm process and re engineered to take advantage of it, they would be 15-20% faster than their current 14nm.

Eh~

Firstly, Intel's uArchs at the moment are really tied to their own nodes, if they were to port elsewhere, it'd take quite a lot of effort. Then on top of that, you've got to worry about frequencies, and well, take a look at Zen 2. The amount of power it draws past 3.55GHz or so rises exponentially. Best example of this is Thrweadripper actually - look at 3960X, 3970X and 3990X reviews. The 3960X barely clocks higher than the 3970X in all-core workloads (like a ~200mhz difference off the top of my head), yet draws the same amount of power (because of the same 280W PPT).

I see no reason to think Intel would be able to keep - or improve - the frequencies they've been able to achieve on 14nm.

 

[maddie]

As expected, AMD fans do not like this question.

I think if Intel CPUs were made on the same 7nm process and re engineered to take advantage of it, they would be 15-20% faster than their current 14nm.

What is the question?

 

[Markfw]

What is the question?

Hes just wishing and trolling.

 

[Staples]

What is the question?

I read the OP and understood the question to be, if Intel CPUs were built on the same 7nm process as Ryzen CPUs, would they be faster than Ryzen 3xxx CPUs? I have no idea the performance differential but AMD CPUs are currently faster. I am taking a guess at the speed increase Intel would be able to squeeze out if the current COUs if they were made on a 7nm process.

and I am comparing desktop CPUs. Of course a slightly improved 10 core CPU will not beat a 24 Core CPU.

 

[Markfw]

I read the OP and understood the question to be, if Intel CPUs were built on the same 7nm process as Ryzen CPUs, would they be faster than Ryzen 3xxx CPUs? I have no idea the performance differential but AMD CPUs are currently faster. I am taking a guess at the speed increase Intel would be able to squeeze out if the current COUs if they were made on a 7nm process.

and I am comparing desktop CPUs. Of course a slightly improved 10 core CPU will not beat a 24 Core CPU.

The problem with the question, how its worded, and many replies, is that the node and the technology used on that node do not compare or in any way equate to be able to compare the CPU's. The final product is a combination of technology, design and quality of the implementation, so the question is irrelevant.

Its also possible that the reason that Intel has failed (so far) with its 10 nm node, is that they did not change the rules with how their design was, to fit the new technology in that node. I may not be explaining myself well, so feel free to reword it if you understand what I am trying to get at.

You can't (for example) take AMDs design that uses 7 nm on TSMC, and just pop it into a 10 nm Intel, even though they are supposed to be similar,. You can't compare apples and Oranges.

 

[Zucker2k]

The problem with the question, how its worded, and many replies, is that the node and the technology used on that node do not compare or in any way equate to be able to compare the CPU's. The final product is a combination of technology, design and quality of the implementation, so the question is irrelevant.

So you actually don't like the question.

The question is hypothetical and needs to be treated as such. I also believe he dealt with node disparity by taking that out of the equation, effectively rendering this a matter of brains and resources at each company's disposal.

 

[Markfw]

So you actually don't like the question.

The question is hypothetical and needs to be treated as such. I also believe he dealt with node disparity by taking that out of the equation, effectively rendering this a matter of brains and resources at each company's disposal.

Its not that I don't like the question, its irrelevant. hypothetical would be ok, if it wasn't fantasy. Brains and resources have nothing to do with the end result. Obviously Intel has way more resouces as an example. Why is 10 nm still broken ? And brains ? You would think with all their money they could have bought good brains and fixed it. No so.

 

[LikeLinus]

Its not that I don't like the question, its irrelevant. hypothetical would be ok, if it wasn't fantasy. Brains and resources have nothing to do with the end result. Obviously Intel has way more resouces as an example. Why is 10 nm still broken ? And brains ? You would think with all their money they could have bought good brains and fixed it. No so.

"Why is 10 nm still broken " Probably because they didn't expect AMD to farm out their actual production. It's easy to blame a company who does everything top to bottom, but to congratulate a company who only designs something and a specialized company to build it. No one seems to talk about how AMD doesn't actually have to manufacture the product. Congrats to them, it was a smart move.

 

[Markfw]

"Why is 10 nm still broken " Probably because they didn't expect AMD to farm out their actual production. It's easy to blame a company who does everything top to bottom, but to congratulate a company who only designs something and a specialized company to build it. No one seems to talk about how AMD doesn't actually have to manufacture the product. Congrats to them, it was a smart mood.

They have farmed out their production for years. I can't remember the date they sold off their fabs, but it was way more than 3 years ago. And Intel failing at 10nm is all their fault. They were supposed to be the world leader in fabs, before 10nm.

 

[Ajay]

"Why is 10 nm still broken " Probably because they didn't expect AMD to farm out their actual production. It's easy to blame a company who does everything top to bottom, but to congratulate a company who only designs something and a specialized company to build it. No one seems to talk about how AMD doesn't actually have to manufacture the product. Congrats to them, it was a smart move.

Uh, Intel's 10nm node is failing to deliver reasonable yields. There are problems with the cobalt interconnects and quad patterning (doesn't really *self align* so well). This results in chips that just outright fail (defects) and fail to perform at expected power/frequency. So, yeah, the node is busted.

 

[eek2121]

Why, at 10nm, we'd have the mythical 10GHz Pentium4. Obviously.

Which would be an interesting part, to say the least. While I said that jokingly, The last time Intel pulled that kind of stunt, we got the Core architecture. (Core was just basically a tweaked Pentium M).

I’m confident Intel will still be selling allot of 14nm server CPUS. They only have 3 Fabs on 10nm. They would have to have high yields to have any hope of making SR a mainstream product.

The trouble for Intel is that the longer they wait to shrink, the harder it'll get to sell those CPUs. Sure, it's a tight race now, but what about when AMD (presumably) moves to 5nm next year? What about 3nm? If AMD EPYC chips in 2 years have 50%=100% more per-core perf/watt than equivalent Intel chips, would anyone realistically continue to spec out, validate, and purchase new Intel chips? I'm sure there is a small minority of folks who swear by Intel, but most companies I have worked for don't just take a look at marketing fluff. In addition, Intel is starting to lose OEMs...

Or, YASKLV (yet another skylake variant). Intel has built up a ton of 14nm capacity - they *have* to use it till that start converting 14nm plants over to 7nm.

They are kind of stuck. If they take away from 14nm capacity, they hurt supply. They can't charge a premium price because of lower demand. Therefore, their margins are going to drop short term. Actually, the smarter thing for Intel to do would be to scale up 10nm production and don't touch 14nm. Build new fabs.

As expected, AMD fans do not like this question.

I think if Intel CPUs were made on the same 7nm process and re engineered to take advantage of it, they would be 15-20% faster than their current 14nm.

Many of the people on this forum (including Mark and myself) aren't necessarily fanboys. I'll speak for myself rather than mark, but if Intel actually pushed out a superior chip tomorrow, I'd buy it. The problem is they aren't even in the same league as AMD currently. You cannot buy a 128 core/256 thread 1U-3U server from Intel AT ALL as far as I'm aware, much less for under 20 grand. On the desktop, you can't get a 16 core/32 thread CPU AT ALL from Intel (unless you go with a Xeon, but at that point, you are better off just buying a Threadripper 39990X).

If Intel launched a competitive 16 core/32 thread part that beat AMD in perf/watt and came in at a competitive price, I'd buy it in a second. However, they cannot do that while being hamstrung by 14nm.

Uh, Intel's 10nm node is failing to deliver reasonable yields. There are problems with the cobalt interconnects and quad patterning (doesn't really *self align* so well). This results in chips that just outright fail (defects) and fail to perform at expected power/frequency. So, yeah, the node is busted.

That news is 2 years old (at least). 10nm has been reworked. Their current issues have nothing to do with yields. They have an architecture problem and a clock speed problem. They could put out a desktop 10nm part tomorrow (mobile outsells desktop/DIY, and they have 10nm mobile parts), but they would have a clock ceiling of around 4.0-4.4 GHz (including overclocking). The drop in clock speed does not compensate for any efficiency gains in 10nm, and they can't easily squeeze more performance out of their current design.

 

[jpiniero]

That news is 2 years old (at least). 10nm has been reworked. Their current issues have nothing to do with yields. They have an architecture problem and a clock speed problem.

Err... no. If 10 nm yields were even just bad, there would be no 14 nm shortage.

 

[JasonLD]

Err... no. If 10 nm yields were even just bad, there would be no 14 nm shortage.

Moving to 10nm should be a part of the reason why there are 14 nm shortages.

 

[geegee83]

Will we be seeing Intel and AMD designs go with significant IPC increases (>20% a year) or have we hit diminishing returns?

Is single core performance going to just rely on frequency increases while the way forward will be to add more and more cores?

 

[coercitiv]

Their current issues have nothing to do with yields.

Moving to 10nm should be a part of the reason why there are 14 nm shortages.

I'm a having a hard time digesting this. I can see how the move became a problem in the context of higher than expected market demand, but that was last year.Today, with presumably good yields in their 10 foundries, with a nearly two-fold improvement in manufacturing velocity (according to Intel), they're still unable to meet demand. I'll just go with Occam's razor here and say that the simplest explanation is still yield related, not some awkward inability to properly transition nodes from the biggest CPU manufacturer in the world.

Last but not least, if they had fixed 10nm for good, they would shout it from the roof tops, keynote slides with fancy graphs would leak all over the media. Instead we get this:

We will have sufficient supply in the first quarter, sufficient supply throughout the year. I think our challenge is week-on-week supply as our customers are hoping for. And second, particular SKUs or mix, making sure that we have the right product mix.

They're still trying to put a positive spin on what HP and Dell are telling us impacts their sales forecasts, as if their problem was week-on-week supply.

 

[Insert_Nickname]

Which would be an interesting part, to say the least. While I said that jokingly, The last time Intel pulled that kind of stunt, we got the Core architecture. (Core was just basically a tweaked Pentium M).

Oh, I was quite in on the joke...

But I will say that Intel's Core moment wasn't completely unexpected if you were paying close attention back then. Pentium M, particularly the 2MB L2 Dothan variety, was a real beast. Yonah was "only" an optimized dual core version of it, and Conroe was "just" a desktop optimized version of Yonah, with a few more tweaks up it's sleeve.

But unlike then, Intel didn't have a "backup" architecture to fall back on when the 10nm issues started. They're paying for not having a plan B, or even plan C, now.

Will they come back? Of course. Meanwhile, AMD being competitive benefits everyone.

 

[moinmoin]

Will we be seeing Intel and AMD designs go with significant IPC increases (>20% a year) or have we hit diminishing returns?

As the introduction of Zen has shown there's room for sporadic IPC increases of that size. Per gen the IPC increases in the Zen line were also bigger than we are used from Intel. But >20% a year is too much to happen as regularly as every year.

Is single core performance going to just rely on frequency increases while the way forward will be to add more and more cores?

Assuming same frequency (which is where Intel's 10nm currently fails) single core performance increases whenever IPC increases, so this is kind of a non-question.

 

[Nereus77]

When Intel fans can't face reality we get threads like this.

Look. When Intel was top-dog, they made the best CPUs. If you wanted value and best performance, they were difficult to beat. Now, AMD isn't yet in the same position but they should be there some time in the foreseeable future. Intel still wins for certain software/tasks, but only by a whisker, whereas AMD dominates in most other uses. This makes an AMD CPU the better choice at this point in time.

You can neither entirely put Intel down nor AMD. Its not black & white, there is an entire scale of grey in between. However, I maintain the view that I am a fan of the best CPU regardless of the house it comes from.

I'm neither team red nor team blue, but team gray. Show me a great CPU and I'll buy it regardless of brand.

Both AMD and Intel have amazing engineers who just want to make the best damned CPU you've ever seen.

 

[zinfamous]

So you actually don't like the question.

The question is hypothetical and needs to be treated as such. I also believe he dealt with node disparity by taking that out of the equation, effectively rendering this a matter of brains and resources at each company's disposal.

um no. it's a stupid question is the problem--just watch the Jim Keller interview.

The architecture is designed for the node. You don't just plop the same design down on a node that it was never designed for. You don't take a 10nm (or what: 14nm+++++) design and just toss it into the 7nm TSMC microwave and expect the same chip, but way faster, to just pop out. It simply doesn't work that way. Intel hasn't designed anything for 7nm afaik; certainly not anything that has been consumer-facing for years, now.

Has nothing to do with "not liking the question," but pointing out that it doesn't even rise to the level of "interesting hypothetical," because the question simply doesn't accept (or doesn't understand, or lazily ignores...) the reality of how this works. You can't take node disparity out of the equation, because it is fundamental to the design. It absolutely makes no sense--except that one would do this because they don't want to accept that the question really isn't worth asking.

 

[Adonisds]

As the OP, I find it funny that this became the "Official Intel vs AMD thread". I didn't mean to or predict that it would happen, I just wanted to tackle a very specific thought experiment that almost no one here tried answering. Because of that I then tried answering my own question and no one cared too. 😂

I'm not complaining, I am fine with how it turned out, you guys can talk about whatever you want obviously. I guess I should have predicted what would happen.

 

[Adonisds]

When Intel fans can't face reality we get threads like this.

Look. When Intel was top-dog, they made the best CPUs. If you wanted value and best performance, they were difficult to beat. Now, AMD isn't yet in the same position but they should be there some time in the foreseeable future. Intel still wins for certain software/tasks, but only by a whisker, whereas AMD dominates in most other uses. This makes an AMD CPU the better choice at this point in time.

You can neither entirely put Intel down nor AMD. Its not black & white, there is an entire scale of grey in between. However, I maintain the view that I am a fan of the best CPU regardless of the house it comes from.

I'm neither team red nor team blue, but team gray. Show me a great CPU and I'll buy it regardless of brand.

Both AMD and Intel have amazing engineers who just want to make the best damned CPU you've ever seen.

When people have poor capacity of abstraction we get jabs like these that totally misunderstand OP's text and still get lots of likes.

I need to try to remember that more, I did add "The both companies using the same process scenario is just to illustrate the problem and I know it will never happen" but I should have also removed anything concrete from the text and used instead "company X, company Y, Z nm process".

I am also team grey, I don't care who is ahead. This thread was not meant to be fanboy wars, it was not meant to discuss who is in a better position between Intel and AMD. It was not even meant to discuss who has better IPC.

The replies that try to be technical get much less response than stuff like this reply I am quoting.

This thread probably has lots of responses because I unintentionally used a clickbait title. Many people probably didn't even read the OP. It's my fault, I will try to avoid that in the future.

 

[scannall]

As the OP, I find it funny that this became the "Official Intel vs AMD thread". I didn't mean to or predict that it would happen, I just wanted to tackle a very specific thought experiment that almost no one here tried answering. Because of that I then tried answering my own question and no one cared too. 😂

I'm not complaining, I am fine with how it turned out, you guys can talk about whatever you want obviously. I guess I should have predicted what would happen.

In all fairness that is a loaded question, with an assumption. At this point I don't think Intel is ahead with products in the market. Things you can actually buy. Marketing slides and roadmaps don't count. Currently, with shipping products AMD is better at SMT, much better at security, pretty much even in IPC and much better at connectivity PCIe 4 vs PCIe 3. Intel is ahead in memory controllers/latency. And semi-niche things like AVX512.

Depending on your use case, AMD is more likely to be a better pick than Intel today. What will next year bring in actually shipping parts? I have no idea, my crystal ball is broken. And while speculation and rooting for your favorite team is all well and good, the only parts that matter are the ones you can buy today.