The 13th generation Core is equipped with a larger-capacity L2 cache
Intel’s 13th-generation Core processor, Raptor Lake, is basically a modification of the existing Alder Lake processor. The most obvious change should be to increase the number of E-Cores and increase the cache capacity of the entire core. The L2 cache capacity of both P-Core and E-Core has been increased, among which the L2 cache of P-Core has been increased from 1.25MB to 2MB, and the L2 cache of E-Core has been increased from 2MB to 4MB. Generally speaking, the advantage of increasing the cache is to bring the data closer to the CPU core and reduce the CPU access to memory for data access operations, which can reduce latency and improve kernel efficiency. However, the increase in the cache can significantly increase the chip area, power consumption, and heat, and also increase the delay.
However, according to the tests of chipsandcheese, the L2 cache latency of Raptor Lake’s P-Core has not increased significantly. It is estimated that the huge structure of Golden Cove alleviates this problem. The L2 cache latency of Raptor Cove has slightly increased from the current 15 cycles of Golden Cove to 16 cycles, and the increase is not obvious.
The latency of the L3 cache may also increase by 1 cycle, but the L3 cache speed of the Intel processor is bound to the Ring frequency. Since it is not clear what the Ring frequency of Raptor Lake is, it is not yet easy to judge.
The above picture is a bandwidth comparison. It can be seen that the L2 and L3 cache bandwidths of Raptor Lake and Alder Lake are basically the same at the same frequency. It is now tested that the bandwidth of Raptor Lake’s P-Core L1 data cache is lower than that of the previous generation, but this may be a limitation imposed by Intel on ES processors.
The architecture of the E-Core has not changed, it is all Gracemont, so the cache latency of the E-Core of the 13th and 12th generation Core processors is the same, which is also 20 cycles.
It’s just that the increase of the L2 cache of E-Core may be more profitable than that of P-Core, because the random execution buffer of E-Core is smaller than that of P-Core, so the delay when they access the L3 cache is very high. This will cause a lot of performance loss, so increasing the L2 cache size can directly improve IPC.
The above tests are based on a Core i9-13900K ES sample with P-Core running at 4.9GHz and E-Core running at 3.72GHz. There is no mention of the Ring frequency. Since it is an ES rather than a QS sample that is closer to the official version, the test results may be different from the final retail product.
