Doesn't the fact that Intel is matching/beating the 9950X with fewer threads prove the fact that having smaller, carefully designed single-threaded cores provides better results than bigger cores with SMT in fully multi-threaded workloads? Which is exactly my point? Adding little cores that take up less space and power is the way to go for multi-core workloads, and everyone agrees. Yes, AMD still has the same core architecture and HT in their little cores for now, but Intel's showing obvious advantages in changing that by having superior IPC.
And for Intel having to run Windows, you completely ignored my point that Zen had THESE EXACT SAME issues. Radically new CPU architecture provided outstanding multi-thread performance by boosting core count with the dual-CCD design, but faced lots of issues with Windows and apps and games not knowing how to handle thread assignment. Exact same issues Intel is facing now. And those issues haven't even been fully solved yet. As I pointed out, even the Zen 5 launch was compromised by Windows not handling the new architecture properly because 24H2 is trash.
But in hindsight, AMD absolutely made the right choice with Zen and the chiplet design. And in 8 years, I'm sure we're going to look back at 12th gen and Core Ultra and say the same thing about e-cores and removing SMT.
And once again you're missing the point. Sigh.
I'm struggling to see the point you're trying to make? That having more (24) cores perform roughly the same as 33% less (16) is meant to be impressive somehow?
You never remotely addressed my flipping of the statement to talking about cores, and again chose to focus on threads.
For starters, you're comparing E-cores to SMT, which these are not mutually exclusive concepts - they can coexist. Secondly, SMT provides good MT value for die space area, something even Intel discussed when announcing Arrow Lake despite it not having it. SMT largely reuses existing core hardware to implement a second thread, the additional footprint of SMT isn't massive - plus you're more likely to maximise core resource utilisation under SMT. Whereas, adding an E-core is literally just adding an entire extra core.
An SMT thread may not be as strong as an E-core thread, but it also correspondingly takes up much less area than an E-core.
The only sensible justification for removal of SMT/HT from the P-core is if removing it enables optimisation of the architecture which results in better ST, which to be fair I do think we see in ARL.
I don't think anyone is disputing that having high performance and low power/area cores is the way to go, just that Intel has jumped the gun with a solution that Windows isn't designed to take full advantage of.
Plus, Intel's P-cores are so big they literally need E-cores to compete with AMD at all. AMDs standard cores are already reasonably sized, the c cores more so.
The fact that we may end up with a solution in "8 years" that aligns to Intel's approach doesn't change the pitfalls of Intel's approach now.
