3 ms·
This generation of Intel CPUs is essentially more E-cores, higher clocks, and perhaps an updated DRAM controller which is guaranteed to work with higher-clocked
by gautamcgoel 3y ago
This generation of Intel CPUs is essentially more E-cores, higher clocks, and perhaps an updated DRAM controller which is guaranteed to work with higher-clocked RAM. I don't find any of this particularly compelling, and would probably wait for the next generation, which is supposed to represent a much more substantial upgrade. I would also like to see more processors offered with just P-cores - I would rather buy a $200 CPU with 8 P-cores than a $250 CPU with 8 P-cores and 12 E-cores (I understand that others may have different needs).
- brucethemoose2 3y ago> I would also like to see more processors offered with just P-cores - I would rather buy a $200 CPU with 8 P-cores than a $250 CPU with 8 P-cores and 12 E-cores (I understand that others may have different needs). On the contrary, I think there's a market for, say, 2 P cores and 24+ E cores. They are extremely area efficient, and some workloads like compilation or chunked encoding just want more real cores. And this niche is (for the moment) an advantage Intel has over AMD. Also, Intel already offers P core only SKUs.
- gymbeaux 3y agoI don’t see how P+E/bigLITTLE is giving Intel an edge when Ryzen 7th Gen is still whooping Intel CPUs in the power consumption department. What are some cases where one would want 2 P cores and 24 E cores?
- brucethemoose2 3y agoCompilation or video encoding, as said above.
- gymbeaux 3y agoBut why would I want them to be E cores instead of P cores? Battery life?
- gautamcgoel 3y agoYes, I agree that there is a market for this. But I am not part of that market - I really want more beefy cores with lots of AVX units (I'm perfectly content if these beefy cores have low base clocks to conserve power).
- brucethemoose2 3y agoI am curious to know what you are doing that is AVX heavy, but doesn't benefit from more cores? The first thing that came to my mind is llama.cpp, but its actually core count bound for prompt ingestion and mostly RAM speed bound during token generation.