AMD 3rd Gen EPYC Milan ‘EPYC 7543’ CPU With 32 Cores & 3.7 GHz Boost Clocks Benchmarked, Faster Than Dual Intel Cascade Lake Xeon CPUs With 56 Cores

Another AMD 3rd Gen EPYC Milan CPU has leaked out and this time we get to see benchmarks of the EPYC 7543 with 32 Zen 3 cores. The 3rd Gen EPYC Milan server processor lineup is expected to launch this quarter and would be displacing the 2nd Gen EPYC Rome lineup which featured the Zen 2 core architecture.

The AMD EPYC 7543 was spotted by Twitter fellow @Leakbench in the Geekbench 4 database. The chip features the Zen 3 core architecture and is comprised of 32 cores and 64 threads. The system was running a single-socket configuration so it isn't using a 2P design which we saw in the previous benchmarks.

As for clock speeds, the chip has a rated base clock of 2.80 GHz and boosts up to 3.70 GHz which is quite a respectable clock rate for the processor. Based on the clocks, we can assume that this part will have a TDP close to 200W. The CPU also features 256 MB of L3 cache and 16 MB of L2 cache. It confirms that the chip is actually making use of 8 CCD's instead of four. The four CCD 32 core part would be configured for a different SKU with 128 MB of L3 cache.

In terms of performance, the CPU scored 6065 points in single-core and 111379 points in the multi-core tests. For comparison, a dual-chip Intel Xeon Platinum 8276 server with 56 cores & 112 thread processors only manages a score of 4913 points in single-core and 112457 points in the multi-core test. The Intel test platform was also configured with 192 GB of system memory while the AMD 3rd Gen EPYC Milan system was configured with 384 GB of system memory. So just for doing a better comparison, we found a result for a dual Xeon Platinum 8280 server with 56 cores and 112 threads that also featured 384 GB of system memory. This system scored 5048 points in the single-core & a higher 117171 points in the multi-core test.

.

.

Regardless of featuring a higher core and thread count, the single AMD EPYC Milan CPU was just enough to score a better single-core score while coming on par with Intel's top of the line Cascade Lake Platinum Xeon chips. We also have to take into account that this benchmark makes full use of the AVX-512 instruction set that is featured on Intel's CPUs, giving them a higher score advantage though there are not many real-world workloads that utilize AVX-512 instructions.

So in apples to apple comparison, the AMD EPYC Milan CPUs will have a tremendously higher lead and we are just taking general performance without mentioning the better performance/value and perf per watt which would translate to lower TCO when building a Milan server.

Also, 10nm+ clocks aren't going to look great compared to 14nm+ nodes. Clocks and efficiency are a whole different thing and Intel also seems to be lacking in terms of cores with AMD now offering two full generations of server CPUs with up to 64 cores. It is looking grim for Intel's Ice Lake-SP line of server chips but Intel seems to be placing all bets on AVX-512 workloads since that's the only edge they have versus AMD's EPYC processors right now.