Intel Raptor Lake And AMD Zen 4 CPU Cores Compared In IPC Benchmark Showdown
It seems like regular leaker Raichu (@OneRaichu on Twitter) has gotten his little Pokémon paws on a pair (at least) of pre-release processors from both Intel and AMD. With the hardware in hand, he did what any proper enthusiast
would do, and benchmarked them against each other.
However, these benchmarks aren't just the usual Geekbench
or CPU-Z, but instead the server-oriented SPECCPU2017 suite, and he ran them using fixed 3.6 GHz clock rates. This gives us an interesting window into the real per-clock performance of Intel's Raptor Lake processors as well as AMD's Zen 4 cores.
Raichu's tweet with images omitted; reproduced below.
There are a few caveats to keep in mind here; for one, neither of these chips are intended to operate at such a low clock rate—Yes, 3.6 GHz is low these days
! For another thing, SPEC is a benchmark suite really targeted at HPC users, not client desktop applications. Because of that, the performance here may not be representative of actual application performance when these CPUs finally launch.
It's still quite fascinating to peer over the data, though. Raichu tested both chips with DDR5-4800 RAM, which is notably below the stock speed for both chips. Then, he tested them using DDR5 memory running at 6000 MT/s—which AMD previously confirmed to be the "sweet spot" for Zen 4 processors.
DDR5-4800: RPC = Raptor Cove, GLC = Golden Cove, RPH = Raphael
Regardless of the configuration, Intel's Raptor Cove and AMD's Zen 4 (represented here by "RPH" for Raphael, the codename of the first Zen 4 desktop CPUs) are neck-and-neck. Intel's parts generally have a small advantage except for in SPECint with the slower 4800 MT/s speed, where the difference is so small it may as well be margin of error.
Contrary to what you might expect, Zen 4 actually gains less performance from the faster memory than Raptor Lake does. Raichu comments that Zen 4 has a "great" cache configuration, which makes it "not very dependent on DRAM."
DDR5-6000: GCM = Gracemont
While testing at the higher memory speed, Raichu also did the math for the Intel chips' Gracemont cores. Unsurprisingly, they come out far behind the big cores in floating-point performance, although they're impressively close in integer math. Raptor Lake offers a tiny IPC uplift for those cores too, likely due to the doubled L2 cache
available to those cores.
Earlier leaks have promised
that AMD and Intel's next-generation chips will be neck-and-neck for performance, and this testing bears out those promises. If the IPC is this close, it's likely that the performance crown will come down to whoever can clock the highest, and it looks like Intel's gunning for that title by any means necessary