Intel 13th Gen Raptor Lake Optimizes 12th Gen Alder Lake Latencies Across A Wider Array of Cores
The upcoming Intel 13th Gen Raptor Lake CPUs will increase the L2 cache sizes compared to the Alder Lake series. The current generation of Alder Lake-S processors is powered with the 1.25 MB Golden Cove P-cores, while the new Raptor Cove P-Cores, totaling eight each, have 2 MBs of L2 cache dedicated to each core.
The amount equals a sixty percent increase in the upcoming generation compared to the current Alder Lake cores. Additionally, the Gracemont E-core clusters equal to four E-cores are two times the size of the L2 cache shared throughout the four cores in the cluster, which is now increased to 4 MB. L3 cache between the P and E-core also see an increase to 36 MB from the previous 30 MB.
Chips and Cheese followed and calculated the latencies in L2 cache latencies of previous Intel microarchitectures and discovered an increase in latency levels. Increased caches drive performance since more data is open and near the processor's cores, limiting the more considerable fetch/store procedure that travels to the RAM. Luckily, power, heat, die-area, transistor count, and latency are relatively quicker than a system's DRAM.
Furthermore, OneRaichi has pointed out that the changes within the cache structure of the Intel Raptor Lake 13th Gen CPUs have allowed for a big optimization in latencies overall. The previously encountered wall between the P-Core and E-Core has now more or less disappeared which was due to the inefficient latencies for the E-cores.
This is why I just say the RPL is just not only a refresh.
you can see the ringbus structure has some changes and it makes the boundary of the P and E core is almost disappeared.
I guess the network topology of ringbus is different. pic.twitter.com/6pZSLe6JFx
— Raichu (@OneRaichu) August 23, 2022
Chips and Cheese report that the four-way associative 256 KB reserved L2 cache with the Skylake processor cores has an L2 cache latency equal to twelve cycles. Sunny Cove and Cypress Cove cores have a latency of 13 cycles due to the 512 KB size increase. Willow Cove utilizes an associative cache (20-way), while Golden Cove uses only ten. Furthermore, the latency increases to fourteen cycles. With Raptor Cove P-core on the horizon, the new core offers 2 MB of 16-way L2 cache and reduced latency of fifteen cycles.
This new report shows that intel has done a significant amount of designing and testing to assist with cache and power management to make the new cache more power-friendly. Readers should also note that the new 13th Gen Core Raptor Lake processors are constructed utilizing the 10 nm Enhanced SuperFin node currently used in the current generation of Intel processors.