Memory subsystem performance
Let's kick off our testing with a quick look at main memory performance. Both the Ryzen Threadripper CPUs and every Core i9 in this test are running with the same DDR4-3600 kit, so we can easily make apples-to-apples comparisons about their performance using AIDA64's built-in memory benchmarks.
Intel's memory controller on the high-core-count Skylake-X die seems to favor read performance over writes. Given the potential hunger for data of these chips' AVX-512 units, that's probably the right balance to strike. Copy bandwidth is slightly higher than on the low-core-count chips, but only slightly. The Ryzen Threadripper duo beats out the Skylake-X CPUs in memory writes and copies, but read bandwidth lags behind Skylake-X chips of equal or higher core counts.
Even though they have many more cores and threads—and thus a broader mesh to traverse—than low-core-count Skylake-X chips, the i9-7960X and i9-7980XE deliver memory access latencies in line with their less resource-endowed cousins. That's a testament to the scalable nature of the Skylake Server mesh architecture.
Some quick synthetic math tests
To get a quick sense of how these chips stack up, we turn to AIDA64's synthetic benchmarks. Photoworxx stresses the integer SIMD units of these chips with AVX. FPU Julia tests single-precision floating-point throughput and uses AVX instructions (though not AVX-512), while FPU Mandel puts those same instructions to work in the service of double-precision throughput.
Hm. It seems Photoworxx may not be fully optimized for CPUs with this many cores. Perhaps the benchmark will be optimized for these chips in the future, at which point we'll have to retest.
As we'd expect, throwing more cores at the AIDA64 Hash benchmark produces basically linear increases in bandwidth for our Skylake-X chips.
Ryzen Threadripper chips do outpace the Intel competition in this benchmark, but that's because the Zen architecture has what seems to be little-publicized support for Intel's SHA Extensions. These extensions permit hardware acceleration of some of the SHA family of algorithms, and CPU Hash uses SHA-1 as its algorithm of choice. SHA-1 isn't particularly useful in practice any longer, but SHA-256 is, and the folks at SiSoft report similar speedups for that algorithm. AVX implementations of other SHA versions might help Intel processors close the gap, though.
Thanks to their large complements of wider AVX units compared to Threadripper CPUs, the i9-7960X and i9-7980XE come close to doubling the throughput of the 1950X in the Julia test, and they still maintain a healthy lead in the Mandel test. That's excellent performance. Now, let's see how these chips handle games.