Move Over 3D V-Cache, Intel Raptor Lake Could Pack A Huge Cache Upgrade For Gaming
Even after all of our refinements to the technologies; even despite innumerable advancements, the single biggest bottleneck for superior CPU performance is still simply getting data into and out of the CPU. That fact was brought into sharp focus with the release of AMD's Zen 2 CPUs, which offered drastically-improved gaming performance over their Zen+ predecessors largely due to their massive L3 cache.
AMD's been leaning hard on the cache play recently; not only did the company ship graphics processors that are heavily reliant on large near-GPU caches for high performance, but the red team's response to Intel's Alder Lake snatching away the gaming performance crown was to slap an extra 64 MB of L3 cache on its fastest single-threaded CPU. It remains to be seen who's the real victor in that fight, but the approach stands in stark contrast to Intel's idea of simply ramping up clock speeds to the sky.
Well, it looks like Intel sees at least some merit in AMD's approach—at least if the latest leaks on Twitter are to be believed. Largely-anonymous leaker OneRaichu tweeted vaguely at first, saying that "68M cache is interesting.😄" OneRaichu clarified in a later tweet that he was talking about Intel's 13th-generation Raptor Lake's total L2 and L3 cache.
The image above is a die shot of Intel's Alder Lake-S silicon, used in the Core i9-12900K, Core i7-12700K, and Core i5-12600K, among other CPUs. You can clearly see illustrated that Alder Lake includes 1.25 MB of L2 cache for each Golden Cove P-core, and a shared 2MB of L2 cache for each cluster of four Gracemont E-cores. Each section (P-core and E-core cluster) also gets 3MB of L3 cache, totaling 14 MB of L2 and 30 MB of L3.
A reply in the thread from Twitter user "ns4e921" gives an idea of the possible layout of this 68 MB of cache on Raptor Lake: 2*8+4*4+3*12. Applying that formula to the expected layout of Raptor Lake gives us eight Raptor Cove cores with their own 2 MB of L2 cache, four E-core clusters with 4 MB each, and twelve sets of 3MB L3 (one for each core or core-cluster.)
If you're confused, check out this diagram by Twitter user Olrak where he helpfully laid things out. It also makes a nice comparison against the Alder Lake shot above. Assuming this information is accurate, we're looking at a 60% increase in L2 cache for the P-cores and double the L2 cache for the E-core clusters. The L3 cache per object seems to remain the same, although the two extra clusters of E-cores raise the total amount of L3 cache to 36 MB.
This latest information lines ups neatly with earlier rumors that Intel's Raptor Lake would feature increased L2 cache and E-core allotments over its 12th-generation Alder Lake CPUs, so we find little reason to doubt its veracity. Still, given that this is a leak and not an official release, take it with however many grains of salt that you wish.
Intel hasn't made any official announcement regarding Raptor Lake or its launch window, although leaked roadmaps pegged it for 2022, and earlier leaks suggested we'd see it in Q3 of this year. That might be pushed back a bit depending on how the market goes, though; after all, Intel just launched its mainstream Alder Lake desktop parts two weeks ago, and AMD has its Zen 4 "Raphael" waiting in the wings, too. Naturally, we'll let you know the score as soon as we do.