Current gen AMD cooler forward compatibility looks likely

[Xlucine]

Even a 24 core CPU will pale in comparison to 64 core TR. We'll continue seeing little socket CPUs getting very powerful, but I think AMD's current strategy of copying big socket servers for HEDT is going far too well for them to drop it. Intel's HEDT line has been completely killed - AMD must be doing something right.

The heatspreader contact area is the same as AM4, so there won't be much more room for more chiplets. Good news is that coolers that get good coverage on AM4 will also cover the whole AM5 heatspreader.

[CAT-THE-FIFTH]

Even a 24 core CPU will pale in comparison to 64 core TR. We'll continue seeing little socket CPUs getting very powerful, but I think AMD's current strategy of copying big socket servers for HEDT is going far too well for them to drop it. Intel's HEDT line has been completely killed - AMD must be doing something right.

The heatspreader contact area is the same as AM4, so there won't be much more room for more chiplets. Good news is that coolers that get good coverage on AM4 will also cover the whole AM5 heatspreader.

They could just make the single socket Epyc systems,the "new" TR socket. The Ryzen 9 5950X is around £750,so unless Intel can really match/beat a 16 core Zen4 CPU, a 24 core Zen4 CPU is probably going to be priced even higher. That will wipe out entry level TR systems:
https://www.scan.co.uk/shop/computer...gen-processors

Cheapest 24 core Zen2 based TR is over £1000.

We are already seeing core count stagnation and this is because Intel can't even have a proper 16 core competitor to a Ryzen 9 since 2019. Its why an 8 core new generation AMD CPU is over £300(and Intel is barely cheaper).

[g8ina]

watching !

[Iota]

We are already seeing core count stagnation and this is because Intel can't even have a proper 16 core competitor to a Ryzen 9 since 2019. Its why an 8 core new generation AMD CPU is over £300(and Intel is barely cheaper).

Intel went backwards in the core counts didn't they? They went from 10/20 back to 8/16. Even with the BIG.little core architecture I wouldn't even start to compare them to Ryzen as the little cores don't really count in my own opinion, that's just there to lower their TDP.

It'll be interesting to see cooler compatibility between AM4 and AM5 sockets, and which manufacturers support that or provide updated brackets to support customers. Assuming it happens.

[DanceswithUnix]

We are already seeing core count stagnation and this is because Intel can't even have a proper 16 core competitor to a Ryzen 9 since 2019. Its why an 8 core new generation AMD CPU is over £300(and Intel is barely cheaper).

But that Genoa mockup shown nice rows of three chiplets for the cores. Divisible by three gives an obvious product line split: up to 96 cores for server, 64 cores for Threadripper, 32 cores for AM5.

If they did that, then 32 cores at 170W feels about right (if you can afford it). It would also make 8 cores well into i3 territory, which would be interesting for Intel to respond to. Even if the make it 24 cores, that would hardly be core count stagnation.

[watercooled]

Regarding the chiplet count, AMD have only just gone to 8 core CCX (so one CCX per chiplet) with Zen3; immediately increasing the core count by 50% could pose some challenges in terms of latency etc. Without examining rumours etc, I was still expecting 8 core CCX for Zen4.

Also, piling more cores onto a desktop CPU can start pushing memory bandwidth, although DDR5 may alleviate that to some degree.

[CAT-THE-FIFTH]

Intel went backwards in the core counts didn't they? They went from 10/20 back to 8/16. Even with the BIG.little core architecture I wouldn't even start to compare them to Ryzen as the little cores don't really count in my own opinion, that's just there to lower their TDP.

It'll be interesting to see cooler compatibility between AM4 and AM5 sockets, and which manufacturers support that or provide updated brackets to support customers. Assuming it happens.

Yep,and ATM because of that there is really no reason for AMD to increase core counts.

But that Genoa mockup shown nice rows of three chiplets for the cores. Divisible by three gives an obvious product line split: up to 96 cores for server, 64 cores for Threadripper, 32 cores for AM5.

If they did that, then 32 cores at 170W feels about right (if you can afford it). It would also make 8 cores well into i3 territory, which would be interesting for Intel to respond to. Even if the make it 24 cores, that would hardly be core count stagnation.

Most rumours say 24 cores for AM5. But the reason I am saying there is core count stagnation,is because there has been no big count change under £400 since 2017 now. Remember all the rumours than the Zen2 Ryzen 5 would be 8C?? They never happened because AMD realised Intel had nothing on the 12C/16C models they had.

AMD is adding more cores,but at the top end with a price increase,which is what happened when AMD stagnated with BD/Phenom II,and Intel were quite happy to add more cores at the top. So for average consumers,cores stagnated to 4C under £300. You can see that with the new generation CPU releases from AMD and Intel and how the 8C is still a premium priced product(£300~£400).

We have older generation 8C CPUs below that,but once the old stocks run out,I can see no 8C "large core" CPUs under £300 at launch for a while now,unless one of them really wants to push things. Intel is the one who needs to push AMD ATM,and they are really nowhere when it comes to core count. AMD is better in everyway now,and Intel are holding back progression now.

So unless AMD were to fall behind Intel significantly in per core performance,I can see AM5/whatever Intel has being wedded to 6C(plus little cores) below £300 for the next few years.

[Tabbykatze]

So I see Jim @ Adored is back and he is very certain that Genoa is a 16 die 8 core configuration (maximum) for 128 cores which means that they're still 8 core chiplet dies rather than 96 core max at 12 cores dies maxing out at 8 dies.

Interesting, seems to be a lot of diametric opposition on the topic...

[watercooled]

Where are the rumours about 12C dies from? Not that I've been digging, but almost everything I've seen has pointed to 8C. Even a quick search for 12C dies doesn't turn up much besides uncited forum posts etc.

[CAT-THE-FIFTH]

AdoredTV Jim,leaked an Intel roadmap. It seems Intel is going to be at 8C for the next few generations,but adding lots of Atom cores,to bump up thread count.

It also seems the mainstream is going to be stuck at 6C(big cores) for a few years at least,and might not even get many of the Atom cores either!

[watercooled]

The Alder Lake route of 8 big cores plus smaller cores might not be such a bad thing. The large cores are there for peak performance and lightly-threaded applications, but if an application is well-threaded enough to use more than 16 threads in the first place, then the small cores* could be a valuable addition within the limitations of die size and power budget. Indeed, they *might* even be a more efficient use of power and die space than big cores in that scenario.

*When talking about 'small' cores, perspective is important. They are claimed to be in the area of Skylake performance at considerably less power. That's not to be sniffed at, if they pull it off in the final product.

[CAT-THE-FIFTH]

The Alder Lake route of 8 big cores plus smaller cores might not be such a bad thing. The large cores are there for peak performance and lightly-threaded applications, but if an application is well-threaded enough to use more than 16 threads in the first place, then the small cores* could be a valuable addition within the limitations of die size and power budget. Indeed, they *might* even be a more efficient use of power and die space than big cores in that scenario.

*When talking about 'small' cores, perspective is important. They are claimed to be in the area of Skylake performance at considerably less power. That's not to be sniffed at, if they pull it off in the final product.

I get the impression the initial ones will be more akin to the ones in current Atom class SOCs. If they were Skylake level performance they would be competitive with even current Zen2 cores!

[watercooled]

Take a look: https://www.anandtech.com/show/16881...rchitectures/4

It's Intel themselves making the comparison between Gracemont and Skylake. Tremont already performed on-par with Haswell outside of AVX2 workloads, at iso-frequency.

[Tabbykatze]

Where are the rumours about 12C dies from? Not that I've been digging, but almost everything I've seen has pointed to 8C. Even a quick search for 12C dies doesn't turn up much besides uncited forum posts etc.

I think the rumours are generally coming around from Genoa potentially being an 8 CCD system at 96 cores which means it'll be 12 cores per die at 8 dies but if it is topping out at 128 cores, it makes no sense because 128 is not divisible by 12 but it is divisible by 8. So I guess a small form rumour has moved about and then Overclock 3D had an article pointing to Genoa being 12c per die but then 8c per die for desktop which makes no sense (somewhat) becuase AMDs bread and butter has been being able to use the CCDs for their entire product stack and only changing out the IO die. So unless AMD has decided to start segmenting their stack and isolating desktop and enterprise and thereby limiting the overall binning capability of their dies, it doesn't make much sense.

AdoredTV Jim,leaked an Intel roadmap. It seems Intel is going to be at 8C for the next few generations,but adding lots of Atom cores,to bump up thread count.

It also seems the mainstream is going to be stuck at 6C(big cores) for a few years at least,and might not even get many of the Atom cores either!

What's nuts is the 8c/32c (8 big, 32 little) one destined for Arrow Lake. I can understand having a set of little cores for non intensive thread offloading but 32 little cores makes the entire processor feel anemic...Xeon Phi anyone?

[CAT-THE-FIFTH]

What's nuts is the 8c/32c (8 big, 32 little) one destined for Arrow Lake. I can understand having a set of little cores for non intensive thread offloading but 32 little cores makes the entire processor feel anemic...Xeon Phi anyone?

Well if they are Skylake level,then it might not be as crap as I thought they are. But AMD might have more Chad level cores.

[watercooled]

I think the rumours are generally coming around from Genoa potentially being an 8 CCD system at 96 cores which means it'll be 12 cores per die at 8 dies but if it is topping out at 128 cores, it makes no sense because 128 is not divisible by 12 but it is divisible by 8. So I guess a small form rumour has moved about and then Overclock 3D had an article pointing to Genoa being 12c per die but then 8c per die for desktop which makes no sense (somewhat) becuase AMDs bread and butter has been being able to use the CCDs for their entire product stack and only changing out the IO die. So unless AMD has decided to start segmenting their stack and isolating desktop and enterprise and thereby limiting the overall binning capability of their dies, it doesn't make much sense.

That sounds a likely explanation TBH. It doesn't take much for speculation to take off online these days. I also can't see AMD having separate designs for desktop and server; not just because of having to produce two separate chiplet designs, but also because it would mean two entirely different core complex implementations. I find the latter especially unlikely.

What's nuts is the 8c/32c (8 big, 32 little) one destined for Arrow Lake. I can understand having a set of little cores for non intensive thread offloading but 32 little cores makes the entire processor feel anemic...Xeon Phi anyone?

You might be surprised by the projected performance of the 'small' cores, see link in my previous post.