Intel's 7nm Ocean Cove architecture might deliver 80 percent gains to IPC and higher clocks in 2022: enough to catch up with 5nm AMD Ryzen 6000 Raphael?

[rs]

Intel's 7nm is better than Foundry 5nm

No, it isn't. I wonder if it is at least close to 5nm+ that AMD is using for Zen 4. And even if, probably it will be available 1 or 2 years later.

[Greg]

80% increase isn't that much considering skylake will be 6/7 years old...

[Spunjji]

I enjoy a rumour as much as the next man, but 5.5Ghz is pie in the sky given Intel's issues with hitting high clock speeds on both 14nm (original, more dense variant) and 10nm (all variants, but especially the original more dense variant).

80% increase I could potentially believe - Skylake came out in 2015, so that increase over 7-8 years adds up to at least being plausible, if less remarkable. It's like comparing Zen 2 to Excavator.

[hwmook]

I don't know why you guys find it so hard to believe the 80% IPC improvement over skylake, willow cove already has 25% IPC over skylake so that is about a 20% IPC every generation to reach 180% IPC of skylake. The frequency is defintely BS though since there is no way they can be sure of the frequency at such an early stage of developement, its probably a dream rather than fact. Just for comparison sake, zen 2 is already 110% IPC of skylake, zen 3 relasing at the end of the year is 130-135% IPC of skylake. AMD just need to maintain a 15% IPC per gen improvement and they would be able to keep up. I find this target quite reasonable for intel to catch up with AMD. Its going to be interesting to see the 2 of them fighting for the crown.

[david wilcox]

why are we stopping at only a measly 80 percent? a trillion percent increase would sound even more believable, never mind, Intel is now saying the industry should move away from benchmarking anyways, makes you wonder what Intel is trying to do, whatever it is it's getting sill at this point.

[Leo Kuan]

I don't see why people are so skeptical about this...
First of all, Intel has hired Jim Keller. He's not washing the toilets there you know...
Also, they had this long span of years to work on something like this and given the process doesn't allow them to make what AMD did with chiplets, then they are forced to do this, that is to increase the IPC massively.
Jim has been heard discussing about some upcoming core (Golden Cove or Ocean Cove) that it will support ~800+ instruction window which is some 3-4X the number that Zen 2 and Skylake support.
Another thing to note is that mobile has gained back the IPC difference during these years of stagnation from Intel so now Apple A13/A14 should be very similar in IPC to what a fully features desktop core can do, that is support some 200 instruction window.
So I think this is a good time for CISC cores to move on and get another leap in what they can do. Sure, many cores are nice and AMD has capitalized on that. But commonly used software (games, browsers even productivity apps) have yet to become highly parallel so more than 8 cores won't give you much for the next 5-10 years except for some specific cases where an entry level GPU will nevertheless do a better job than even a 3990x.
So IPC still stands as a very important metric and this is what AMD and Intel need to work on and improve. This is a metric which is much harder to improve and that is why AMD has yet to catch up with the best Intel core that is Ice Lake. Zen 2 is still some 10% worse than Ice Lake and next gen (Zen 3 and Tigerlake) might be more even with what comes beyond not being clear.
Anyway, if these rumours hold true than Alder Lake will be a big improvement and will pass Zen 3 by a big margin. And I don't see AMD being capable, technically to add more than 15-20% IPC in one generation for the same uArch.
Anyway, I hope these rumours are true.

SW algorithms can be experimented, changed / implemented easier to take advantage of parallelism than hardware redesign.  Even you are just one SW guy,  there are millions  SW there,  by what qualificantion you can say that won't happen 5 to 10 years?