AMD released its Ryzen 7 5800X3D processor in April, a unique Zen 3 SKU that boasts “ultimate gaming performance” in large part thanks to an increased L3 Cache totaling 96 MB, triple that of its standard 5800X counterpart. New CPU-Z images shared by leaker OneRaichu today suggest that Intel is making a similar play for its next-generation, 13th Gen Core “Raptor Lake” processors, with figures indicating that what is presumably a Core i9-13900K CPU will boast a combination of cache totaling up to 68 MB. This is a significant upgrade over what the current Alder Lake flagships have to work with, which count just 14 MB of total L2 cache among their specifications. Previous rumors had suggested that Raptor Lake chips would feature 55% more cache for improved performance.
Now, it appears.
— Raichu (@OneRaichu) May 18, 2022
About p core L2 cache latency,non-accuracy test result shown in pic2. https://t.co/Gf8o2V04Pa pic.twitter.com/mLarcP9lhl
The Intel 12th Gen Alder Lake CPUs come with 1280 KB of L2 and 3072 KB of L3 cache per core while the Gracemont cores feature 2048 KB of L2 and 3072 KB of L3 cache per cluster (a cluster comprises four Gracemont cores). We know that Raptor Lake CPUs will come with a brand new Raptor Cove core design and will offer up to 24 cores and 32 threads. That’s 8 Raptor Cove and 16 Gracemont Enhanced cores for the top SKU for a total of 32 threads. If the existing cache counts are to be considered, then that’s 36 MB of L3 cache for all cores and 18 MB of L2 cache which gives us a total of 54 MB of Smart Cache for the top SKU. But as said before, Rapor Lake might give us an even higher per-core cache count.
Discussion (3 replies)
Join Discussion →You get some cache and you get some cache! Everyone gets more cache! For some more cash of course.
The jump from 20MB to 25MB (IIRC) of cache on the 12700K to the 12900K is what makes up most of the difference in gaming performance... as an example.
It also seems that both CPU manufacturers and GPU manufacturers are working to push performance this year with cache-embiggened parts.
Intel already did this years ago. Crystal Well. It had 128Mb of cache. Rumor was it was a custom configuration for Apple to boost the IGP, but the L4 cache was accessible to both the IGP and CPU cores. It had hit or miss benefits on CPU performance - not unlike the 5800X3D