And on top of that, it is fully ISA compatible with Zen 5 down to having AVX512 still.
Unlike gracemont or whatever mont they're following up with
Intel added AVX and AVX2 to Gracemont, kind of a huge change there. It would be funny to see Intel add hyperthreading and/or AVX512 to E-cores in the future, but somehow keep the die area small.
Inb4 AVX1024.
Amd milking way... one launch per year, same product but with 5% increase performance. New bugs... "Pay more for the bugs of tomorrow".
Zen 2, Zen 3, Zen 4, probably Zen 5, all around ~25% performance increases from increased IPC and clock speeds.
Please go through part 1 of our Zen 4 coverage, if you haven’t done so already. This article picks up where the previous one left off. To summarize, Zen 4 has made several moves in the fronte…
chipsandcheese.com
Which makes one question, why do they even bother at all?!
If they only have less cache then the only thing that they could possibly do is be slower than the normal cores while still using pretty much the same amount of power...
Cache uses a lot of die area, and taking away half of it doesn't necessarily kill performance. See Cezanne Zen 3 with 16 MB L3 vs. Vermeer Zen 3 with 32 MB L3. A 5700G has most of the performance of a 5800X, and has increased cache and cache per core from the Renoir chips that came before it.
I think they are also using some unspecified tricks to pack the Zen 4C cores closer together, maybe necessitating lower clock speeds. Regardless of why the clock speeds are lower, power consumption can plummet because of it.
The end result is something with more multi-core performance in a given area, like Intel's E-cores. Time will tell how good it is, and if it's any good for consumers (see rumored Zen 4 + Zen 4C variant of Phoenix).