I think he's saying the single thread performance deficit is so great that those extra 16 e cores are meaningless.
Depends on type of workload. I specifically mentioned MT perf workloads in case you did not notice.
Because with more cores at a certain point you'll need also other resources, i.e. memory bandwidth, i/o and so on which will be limited in a desktop platform. I also understand that quite probably nT performance on desktop, regardless of the vendor, will be limited mostly by thermals and power. So adding cores over a certain point wiithin a limited power window will add diminishing returns.
Depends on type of MT workload, not all att memory bandwidth limited. Also Zen6 will likely be DDR6.
If you want > 16 cores, the next increment would likely be 32 cores.
But then, if you start comparing 32 Zen 5 cores vs. 32 E-Cores, which is not a valid comparison, as far as comparing prices and performance.
16 core Zen 5 is likely going to be closer in MT performance to 8+32 ARL than 32 core Threadripper, which is also going to have more memory channels...
Assuming it's 8C per chiplet, it could be both 24C and 32C Zen6. Also, as for Arrow Lake Refresh the rumor is it's 8P+32E (so not only 32E).
But AMD could also have the option of mixing Zen6 and Zen6C cores on DT. Too little about Zen6 at this point. Lots of configs are possible.
Will it though? Maybe in CB MT it might do better than 16c 32t Zen but in workloads that don't use all the cores I suspect Zen 5 will be quite a bit better. Take something like Puget bench and I think Zen 5 will be faster in a lot of those tests and be generally better for people who have a mix of all core / some core and single core applications in their workflow.
But I was talking about workloads where max MT perf was important, in case you did not notice.
Are you suggesting Arrow Lake Refresh 8P+32E will not be faster in max MT perf when all cores are used, compared to 16C Zen6?
Why would it?
Why do people still have this weird fantasy that multicore is faster?
Multicore has diminishing returns for almost all applications. You don't need more than 8 for 95% of DT apps, don't need more than 16 for 99% of DT apps. The rare case of needing more is in server.
Again. I'm talking about max MT perf where all cores are used. Why do you keep bouncing back to gaming or max ST perf? Max ST perf is of course interesting for some users, but not all, and not for the segment I was discussing.
I bet you were in the same crowd that shouted "Nobody will ever need more than 4C" and the rest can buy the expensive Intel HEDT platform & 6C CPU, when that was that Intel was giving us for years. 🤣
-----------
To summarize: You can think whatever you want about who needs more than 16C is needed on desktop and for what use cases. But assuming Zen6 will really be capped at 16C on desktop, AMD will effectively be handing over the max MT perf segment that Arrow Lake Refresh 8P+32E covers to Intel. Because from what adroc mentioned, AMD will not have anything matching it on MT perf in the same price range. You'll have to step up to Zen6 Threadripper to get the same MT perf, which is far more expensive.