Canalys: AMD Will Surpass Intel on CPU Performance Next Year

I see them pushing performance, but also pushing the horizons of our current computing paradigm. Using machine learning and derivatives to write better code; potentially even reviving VLIW in the process. Perhaps abandoning the current processor model altogether.

Also bringing in stuff like persistent RAM, and even embedding the two. Other wackiness.

But in the immediate future, I expect Intel to take the crown back as soon as they have their successor arch running in volume on 10nm, assuming AMD manages to push ahead at all with Zen2.
 
IdiotInCharge said:
I see them pushing performance, but also pushing the horizons of our current computing paradigm. Using machine learning and derivatives to write better code; potentially even reviving VLIW in the process. Perhaps abandoning the current processor model altogether.

Also bringing in stuff like persistent RAM, and even embedding the two. Other wackiness.

But in the immediate future, I expect Intel to take the crown back as soon as they have their successor arch running in volume on 10nm, assuming AMD manages to push ahead at all with Zen2.
Click to expand...

What I have read is that the 10nm that we will see is a gimped 10nm in order to get something, anything, to market, and will wind up performing similarly to what one would have expected a 12nm Intel process would.

Internally I believe Intel is betting more on the next gen process after 10nm to be where they catch up.

If we are to believe Charlie Demerjian's analysis Intel's own internal plans say they are unlikely to catch back up until after Sapphire Rapids, their 2022 server part. So 2023? 2024?

That's in server's though, where many cores matter much more than they do in client workloads. Because of this and their already excellent standing as the per core performance champ, they will likely have an easier time with client loads.
 
Last edited:
Zarathustra[H] said:
What I have read is that the 10nm that we will see is a gimped 10nm in order to get something, anything, to market, and wilol wind up performing similarly to what one would ahve expected a 12nm Intel process would.
Click to expand...

This already exists with Lenovo shipping limited Cannon Lake parts.

I assume that they will get 10nm ramped up, and then move beyond Cannon Lake (which is 10nm Skylake). That's the part that will be positioned in the market with Zen2.

[I can't take the server stuff as much of anything; totally different cadence with different use cases, so mostly it would mean that stuff will work by then, but doesn't tell us how long before]
 
Zarathustra[H] said:
Yeah, but Intel's trend setting process was tick-tock, which has publicly been dead since early 2016 (but internally probably longer)

I can't help but wonder how much longer it's going to make sense to fight for the x86 CPU performance crown. As Intel's difficulties with 10nm have shown, this stuff is hard, even for industry veterans. Hard things are expensive, and x86 CPU's aren't exactly the biggest growth market anymore.

There will always be a performance leader, but I suspect in the future the gaps between the market leader and the 2nd place will be much smaller than we are used to, because there just won't be as much margin in the technology anymore to allow for the types of differences we have been used to.

AMD may take the lead, or they may not, but if they do, I'd imagine Intel will fight to take the performance crown again, but I don't see a future where the gaps are as big as they were in the decade from ~2007 to 2017. Maintaining that kind of lead is becoming way more expensive than it probably is worth.
Click to expand...

Well thought out assessment Zarathustra, I am in agreement with you. Intel's Fabs/Node issues are really starting to eat it up. Having to outsource to TSMC has to be a bit of an embarrassment as well. There is more than enough room for two in not only the x86 CPU space but the GPU space as well. Hopefully AMD can finally turn the corner and turn it's unique position of being in both of those areas into a technological advantage. I've been waiting for a APU success from them for years but it just hasn't materialized yet. ZEN appears to just be turning the corner there wit the announcement of new mobile APUs.

AMDs success in the chiplet area with Infinity Fabric, so successful Intel is copying and playing catch up there. AMD should have a pricing advantage on the lower end of servers for years to come. Abet, Epic may be in a bit of a niche area, in certain workloads - and conditions such as virtualization with vm security/isolation for example.
 
IdiotInCharge said:
This already exists with Lenovo shipping limited Cannon Lake parts.

I assume that they will get 10nm ramped up, and then move beyond Cannon Lake (which is 10nm Skylake). That's the part that will be positioned in the market with Zen2.

[I can't take the server stuff as much of anything; totally different cadence with different use cases, so mostly it would mean that stuff will work by then, but doesn't tell us how long before]
Click to expand...

That is the failed junk they pawned off to China, thats the only thing I have heard coming from the 10nm Intel fabs. 10nm has been a huge failure for Intel, I dont expect the first chips they roll off the line for the desktop or server to be impressive, they will be lucky to match their 14nm process performance. Now it's possible Intel can fix all these problems and get it competitive with AMD but not in the first or likely second round of 10nm products. This is also why analyst have been bullish on AMD and bearish on Intel. funny how many thought it was a huge mistake for AMD to spin off their fabs, but it's looking like that was their best idea. Either way I am looking forward to seeing how it turns out and I prefer them to be close in performance to each other, nothing good happens when either one gets too far ahead of the other.
 
Gideon said:
That is the failed junk they pawned off to China, thats the only thing I have heard coming from the 10nm Intel fabs. 10nm has been a huge failure for Intel, I dont expect the first chips they roll off the line for the desktop or server to be impressive, they will be lucky to match their 14nm process performance. Now it's possible Intel can fix all these problems and get it competitive with AMD but not in the first or likely second round of 10nm products. This is also why analyst have been bullish on AMD and bearish on Intel. funny how many thought it was a huge mistake for AMD to spin off their fabs, but it's looking like that was their best idea. Either way I am looking forward to seeing how it turns out and I prefer them to be close in performance to each other, nothing good happens when either one gets too far ahead of the other.
Click to expand...


I'd imagine it is a mixed blessing.

On the one hand, if you have your own fabs, you can optimize the process to meet the needs of your own products, which has a benefit. You also control the production, and thus make sure you get all the production volume you need. You also don't have to share your profit with a 3rd party fab which expects a cut.

On the flip side, when the process goes wrong, you are stuck with it, like Intel is finding they are now. Being fabless thus is both a risk, and a benefit.

The thing is, CPU design is very complex and difficult work. In the 10nm era, so is silicon fab design. In business there is something to be said about the ability to focus on what you do best, and not being distracted.
 
Estimating 5 years of CPU "Competition" until Jim Keller's Intel team comes up with "Son of Zen". Knowing Keller, "Son of Zen" should be something new and exciting finally from Intel in 2024?
 
Mega6 said:
Estimating 5 years of CPU "Competition" until Jim Keller's Intel team comes up with "Son of Zen". Knowing Keller, "Son of Zen" should be something new and exciting finally from Intel in 2024?
Click to expand...


He has to do this, it's job security?

Why else would AMD hire him back in 2025? :p
 
You must log in or register to reply here.
Back
Top