New Coke can

Intel Will Nix P-Cores and Fall Behind on Single-Thread Performance

Pause and referesh

After its next-generation PC processors, Intel will only employ E-cores according to a LinkedIn post by an employee. This product strategy comports with the view that general-purpose processors’ role is feeding and managing AI accelerators (NPUs) and other specialized engines. Moreover, Windows PCs have failed to make good use of heterogeneous big-little designs. An all-E-core processor rectifies that situation and should improve area efficiency—CPU throughput per mm2, aka bang for the buck.

Why Only E-Cores?

  • Reduced opex—eschewing P-cores, Intel can halve the number of CPU design teams.
  • Reduced product cost—E-cores are smaller and more area-efficient than P-cores, enabling smaller (and, therefore, less-expensive) chips.
  • Changing product requirements—E-cores are possibly fast enough for most workloads and could be sufficient if feeding and managing NPUs and other engines will be all that a CPU does.
  • Homogeneous processors are easier to use—applications and operating systems struggle to schedule work among a mixture of P- and E-cores.

Likely Outcome

  • Competitors will not follow Intel’s approach but will keep developing performance-optimized cores.
  • Customers will prefer competitors’ processors.
  • Intel will extend the final P-core processors’ product lives, but they will fall further behind new competing chips.
  • Intel will lose share.

Intel E-Core Background

Eking out more performance from any technology ultimately results in diminishing returns; for each subsequent performance gain, cost increases faster. Nowhere is this more evident than in CPU design where big, performance-optimized cores are three or four times bigger than the little, efficiency-focused ones targeting 50–80% as much throughput. In recent generations, Intel’s entry into the latter category, the E-core mont (nee Atom) series, are area-efficiency marvels, especially considering x86 architecture complexities and their wide execution paths. They’re even more area efficient than Arm Cortex-A cores. This likely comes from Intel’s CPU team employing more physical-design resources than rivals.

Several years ago, Intel’s PC processors began combining P- and E-cores, offering customers chips with up to eight dual-thread P-cores and 16 E-cores. The combination enabled the company to post high scores on benchmarks such as Geekbench, which separately report peak single-thread and multithread performance.

Practical benefits were elusive, however. Harnessing multiple CPUs to collaborate in parallel on a single task requires them to periodically synchronize. A program can’t move forward until each thread has completed its part, leaving the faster P-cores to wait for their slower neighbors to catch up. Clever software can work around this problem but is impractical because of the diversity of PC hardware. Thus, there’s a good argument for processors to employ only a single CPU type.

Bottom Line

In 1985, Coca Cola Company introduced New Coke, a beverage that tests revealed tasted better than Coke Classic and should help it regain share from Pepsi-Cola. Had the company introduced it but not withdrawn the original formulation, it could’ve avoided the subsequent backlash.

Intel’s situation has a critical difference: it’s already selling E-cores. In addition to the PC chips integrating both CPU types, Sierra Forest is an E-core-only server processor, and Twin Lake is an E-core-only PC processor. The heterogeneous PC processors have sold well, but it’s at best ambiguous that the little cores are useful. Sierra may have won some designs but isn’t displacing the big-core-only Xeons. Similarly, Twin Lake only addresses niches such as Chromebooks. In short, we know that most customers prefer P-cores.

Meanwhile, competitors are fielding stronger CPUs than ever. AMD, Apple, Arm, and Qualcomm all offer cores competitive with Intel’s fastest, a reversal unthinkable ten years ago. Customers that stuck with Intel as it faltered are likely to abandon the company because they have better alternatives than ever and because they judge a deliberate strategy change harsher than execution problems.


Posted

in

by

Tags:


error: Unable to select