As we made mention in the 12th Generation review, these Gracemont Atom… err… ‘efficiency cores’ require a veritable ton of cache to work their magic. When a single core in a given four block of e-cores is active they hummed along happy as clams. When two were active, they were fine. With three… things started to get a bit iffy, and when all four were loaded down they were cache starved. As they are Gracemont cores nothing was (or easily could be) done with L1 amounts. They still have 96KB total (64KB instruction + 32KB data) L1 cache per core. L2 however has gone from 2MB per cluster of four e-cores to four Megabytes per cluster of e-cores. Put another way the i9-13900K’s sixteen e-cores have a combined 16MB of L2 cache (4 x 4MB) and the 13600K and its two 2 e-core modules / eight e-cores have a combined 8MB of L2 total (i.e. the same as last gens i9 e-core total L2 cache).
To help put those numbers in perspective, the e-cores in the 13th generation i9 now have the same amount of L2 cache as the last Intel 16 core HEDT made (Core i9 7960X… which ironically had a Turbo Boost of 4.4GHz or a mere 100Mhz more than the e-cores come clocked at in the i9-13900K), and the e-cores in the 13th generation i5 now have the same amount of L2 cache as the Intel 8 core HEDT 7820X processor (which also ironically had a Turbo Boost of 4.0GHz or a mere 100Mhz more than the e-cores come clocked at in the i5-13600K). (Almost) Needless to say, they are nowhere near as cache starved as they were last year. Especially when they too share the massive L3 cache (36MB in i9, 24MB in i5) with the P-cores.
In testing we still would like to see these e-core’s L1 increased (in size and complexity) and the L2 doubled again, but they are more consistent performers now than they were in the 12th generation. So much so we sincerely hope for a special edition i3 with only one or two P-cores and sixteen e-Cores all with a low wattage TDP. Such a model would make for an insanely powerful file and streaming home file server… and/or pfSense and/or openSense networking security appliance.
It is unfortunate that the above wish list type example will be highly unlikely to ever see the light of day as Intel is still not gone the chiple… err… “tile” route (yet) that AMD has been using for years now. This is an area of innovation that Intel is still lagging behind AMD in; and while the overall performance of Intel’s 13th generation is excellent there is still room for improvement via customization that a “tile” design would, and should have been included for this generation, brought to the table. Due in no small part to the rapid release cycle such features were pushed back to the 14th Generation. The same holds for Intel’s “5” (7nm) fab process. If either had been available for this generation Intel would have truly been back on top with no viable way beyond price for AMD to compete. They did not and as such, AMD has another generation to make up for the surprise innovation Intel unleashed upon them last year.
