- Mar 3, 2017
- 1,626
- 5,910
- 136
In my humble opinion, it's gonna be 8C or 12C 4nm Zen 5 CCD, 8C or 12C 4nm Zen 5 CCD + Vcache, 16C 3nm Zen 5c.
But Zen 5 is so far out still, this is just complete speculation
Plus AMD has gotten so good at preventing leaks (cough zen 4 cough)
Stacking the CCD's on the IOD will probably come with time, but I'd be surprised to see it for Zen5.Could also have cores on the IOD.
There is already a separate monolithic APU in the Zen4 generation for laptops and low end desktops.The 8c CCX is optimal since it can be used in laptops, desktop, and servers. Increasing that to 10-16c CCX would be unfit for laptop APUs since the highend is being catered by rebranded desktops already. Similar thing applies to the "completely 3D L3" which would make the entry level desktop offerings pricey.
But that would be an insanely complex design of stacking logic on logic, idk. Seems very ambitious.Idle power consumption.
Finally, someone got the core count numbers right.Well... this ought to be interesting
In my humble opinion, it's gonna be 8C or 12C 4nm Zen 5 CCD, 8C or 12C 4nm Zen 5 CCD + Vcache, 16C 3nm Zen 5c.
But Zen 5 is so far out still, this is just complete speculation
Plus AMD has gotten so good at preventing leaks (cough zen 4 cough)
Yup. It's a shame imo. I mean prob not for the company itself, but us speculation enjoyers hahaAMD has gotten boring due to the lack of leaks IMO.
I suspect Zen 5 desktop will still have a 16 core max, but I hope I am wrong. We should see a decent performance uplift, however.
Cores on top of the IO die wouldn't make much all that much sense. It's not like there would be much value in selling a product without them, after all. But I could definitely see AMD adding a few cores within the IO die for low power usage and to further free up the main compute die. Would be particularly helpful for the Dragon Range successor.Could also have cores on the IOD.
Intel did that with Lakefield. Probably not too bad. And with hybrid bonding, maybe they could move all the caches to directly below the stacked cores, avoiding the thermal issues of stacked logic. Or maybe cores + L2 on top and a system cache right beneath? Would be interesting, if nothing else.But that would be an insanely complex design of stacking logic on logic, idk. Seems very ambitious.
I'm not convinced that those necessarily have to converge. Right now, AMD has two basic architecture templates. Monolithic for mobile chips, and a common compute die + IO die chiplet arch for desktop through server. Intel also has two such templates. A currently monolithic, and future chiplet arch for client, and a separate, large tile based arch for server. So both have two basic constructions; it's just that AMD draws the line between mobile and desktop, while Intel does between desktop and server. I don't think AMD necessarily has to change that breakpoint, but I do think adding more logic (e.g. GPU, AIE in the future?) to the "IO" die will substantially blur the lines with mobile.At some point in the future Mobile and Desktop will converge for AMD as well - just as Intel is trying with MTL.
Stacking the CCD's on the IOD will probably come with time, but I'd be surprised to see it for Zen5.
Cores on top of the IO die wouldn't make much all that much sense.
Not going to give any weight to AdoredTV's ramblings, but those core count numbers seem far more realistic for a 4nm Zen 5 with significant architectural additions.Finally, someone got the core count numbers right.
Yes, it's 128 cores for Turin and 192 cores for Bergamo+1
And yes, it's still 8 core CCDs for full-fat Zen 5.
Ah, was just going with the flow there. Then yeah, I think adding a few cores to the IO die would be a very reasonable future addition. Not sure if this has been a problem for AMD, but more logic within the IO die would also help give it more beachfront/shoreline for IO breakout.Surprised you both interpreted it that way. Never meant to suggest actual stacking.
AMD has gotten boring due to the lack of leaks IMO.
I suspect Zen 5 desktop will still have a 16 core max, but I hope I am wrong. We should see a decent performance uplift, however.
I'm very certain these core counts are correct, I heard them a few months ago now. I'm also dead certain of who Jim's second "source" is as well - the one that said it was 128c and 192c - and trust me if I were to say who it almost certainly is you'd believe those core counts too.Not going to give any weight to AdoredTV's ramblings, but those core count numbers seem far more realistic for a 4nm Zen 5 with significant architectural additions.
Stop being such a tease. 😁Anyway at this point I'm also pretty certain about some other info as well, which is making me VERY excited for Zen 5 now.
Intel isn't adding any more E-cores. Arrow Lake is still the same core count as RPL.They have to be quite exciting indeed, if AMD still plans to compete on desktop with 16 cores when Intel will add tons of E-cores for multithread.