Monday, December 6th 2021
Intel Prepares Raptor Lake Designs With 24 Cores and 32 Threads, More E-Cores This Time
With the launch of Intel's Alder Lake processors, Intel has switched from a homogeneous to a heterogeneous design of processors, where smaller, high-efficiency cores are mixed with high-performance cores to create a highly efficient and high-performance processor for all kinds of workloads. And it seems like Intel is not over with adding more E-cores to its future products, as the latest leaks suggest. According to the BAPCO's Crossmark benchmark database, Intel's upcoming Raptor Lake processors will feature more E-cores than the high-performance P-cores in the SoC design. As to why this design choice is present, we are not sure and don't have a definitive answer.
E-Cores are suitable for background tasks, and adding more would potentially leave space for P-cores to do heavier workloads. In the benchmark submission, which is now offline, the samples used were a configuration with eight P-cores and sixteen E-cores. Since the big cores are hyperthreaded, it makes up for a total composition of 24 cores with 32 threads. The platform "RPL-S ADP-S DDR5 UDIMM OC CRB" was used with DDR5-4800 memory, indicating an early stage engineering sample with a probably unfinished memory controller. The Raptor Lake generation will also use LGA 1700 socket, DDR5 memory and be present in the desktop and mobile sector once it launches in Q4 of 2022. It will also use Intel's 7 semiconductor manufacturing process, similar to Alder Lake. The only difference with the next-generation design is the updated Raptor Cove core design that brings a significant IPC uplift.
Sources:
Tom's Hardware, KOMACHI_ENSAKA (Twitter), via VideoCardz
E-Cores are suitable for background tasks, and adding more would potentially leave space for P-cores to do heavier workloads. In the benchmark submission, which is now offline, the samples used were a configuration with eight P-cores and sixteen E-cores. Since the big cores are hyperthreaded, it makes up for a total composition of 24 cores with 32 threads. The platform "RPL-S ADP-S DDR5 UDIMM OC CRB" was used with DDR5-4800 memory, indicating an early stage engineering sample with a probably unfinished memory controller. The Raptor Lake generation will also use LGA 1700 socket, DDR5 memory and be present in the desktop and mobile sector once it launches in Q4 of 2022. It will also use Intel's 7 semiconductor manufacturing process, similar to Alder Lake. The only difference with the next-generation design is the updated Raptor Cove core design that brings a significant IPC uplift.
81 Comments on Intel Prepares Raptor Lake Designs With 24 Cores and 32 Threads, More E-Cores This Time
Laptops, sure.
What we should see is a core with 10 P cores, a proper replacement for the 10900/10850k platform.
But i already think having more than 4 E cores is plenty for background task in gaming. they probably have 8 for balancing reason.
it do not means that they won't give more performances to theses cores but we will see.
Because when you look at Ryzen 5xxx vs ADL, it's the P-core that give AMD trouble.
My 9700k pulled 225 watts under AVX load out of the box, and 165w once undervolted. Temps went down from 100C to 77C. Intel is feeding too much voltage out of the box. 1.35 volt out of the box, 1.25v tuned.
Alternatively, small E-core only (relatively performant) office systems would be welcome to help out on the efficiency side. I think that's what Zhaoxin was trying to do.
Adding more E cores seems like a sensible tradeoff.
No difference in applications or game FPS between 5.4 Ghz vs 5.3 Ghz since I think Im starting to bottleneck at the ram / cache side, so I just run it at 5.3 24/7.
From couple different reviews only 8 P-cores enabled at 125W limit benches quite favorably against 5800X. Computerbase.de also benched 12900K with only 8 P-cores enabled at 88W limit to match 5800X eco mode with only a couple % behind.
Intel has a choice. Or at least it is not power that limits the choices. As pointed out above - ringbus becomes inefficient at 10-12 stops. Also P-cores take up die size - not quite at 1:4 rate against E-cores but close to that.
Other than lack of HT/SMT E-cores have roughly Skylake/Zen+ performance level. Sometimes a bit better (mostly INT), sometimes worse (mostly FP).Intel had 16-core Atoms a few gens back (Goldmont, C395x) and had or has 24-core Atoms last gen (Tremont, 5962B which looks like is only available to some cell tower specific uses).
I am also sure that intel went for better general stability instead of better temps. They also gave the means for undervolting to people who need it.
I can't understand. Go Xeon or Ryzen or Threadripper if you really want that many full cores, is not like there aren't those options.
Yeah, and this comes from a guy who hates intel, but I can't deny simple business logic.
Not worth the extra heat in any scenario.