Microsoft Releases Benchmark Tests to Prove AMD EPYC Milan-X CPUs Will Help Shape Their Future

Yesterday, AMD released details about the new EPYC Milan-X CPUs, loaded with 3D V-Cache, as well as AMD's Instinct MI200 and the company's plans for the next-gen Zen 4 technology. During the event, AMD did not reveal specifications for the EPYC Milan-X chips. Microsoft, however, was quick to bring plenty of details about AMD's new CPU and what 3D V-Cache will accomplish with its performance on Azure HBv3 VMs. Several benchmarks were released by Microsoft showcasing the next-gen AMD CPU.

Why should Microsoft care about AMD's future? Microsoft plans to utilize EPYC Milan-X CPUs to help power their own Azure HBv3 Series VM technology, something that is primarily based on two EPYC 7V73X CPUs. Both processors individually deliver as high as 64 Zen 3 cores (128 cores total) for a single server. What Microsoft will do is use eight cores from each individual server which will be reserved to power the "Azure hypervisor and other orchestration routines." This process will give Microsoft's clients a total of five different configurations offering varying core counts—16, 32, 64, 96, and 120 cores—while the EPYC 7V73X processing clock speeds as high as 3.5GHz.

Milan-X features up to 768MB of L3 cache (L3 + 3D V-Cache) per chip, so a dual-socket configuration delivers up to 1.5GB of L3 cache per system, or in Microsoft's case, per VM. Logically, the L3 allocation will depend on the setup. For example, the 16-core VM has access to 96MB per core, whereas the 32-core setup drops to 48MB per core. At any rate, AMD's EPYC Milan-X CPUs L3 capacity represents a 3x upgrade over current Milan chips, or a 6x improvement over the previous Rome processors.

— Tom's Hardware

Microsoft has not altered the remainder of the Azure HBv3 hardware. The Azure HBv3 still offers 4458GBs of memory and 350GBps of bandwidth, which uses STREAM TRIAD to measure the speeds (358 GB/s of throughput). Adding additional hardware, such as Mellanox ConnectX-6 NIC and two 900GB NVMe SSD to reach high Ethernet connection speeds (200 Gbps) and read/write speeds between 2.9 and 6.9 Gbps for the memory.

Microsoft reports that Milan-X, also known as EPYC 7V73X, offers as low as 50% less latency on memory than AMD's current EPYC 7V13. Due to AMD transferring the memory controller into the CPU, it offers a huge increase when talking about memory latency performance. However, Microsoft is quick to point out that their results are not completely based on Milan-X improving the DRAM access latencies.

According to Microsoft, large caches allow for higher cache hit rates and create a combination of L3 and DRAM latencies for improved effective real-world results. Due to the way AMD stacked the L3 cache, the width of the L3 latency distribution has expanded. Nonetheless, Microsoft believes that Milan-X should have an L3 memory latency in the same ballpark as Milan. In a worst-case scenario, Milan-X may present slightly slower L3 latency.

via Tom's Hardware

Compared to its predecessor, AMD's EPYC Milan-X CPUs outperformed Milan by 77% using a 64VM configuration, as well as 257% higher than Skylake using "the f1_racecar_140 model on Ansys Fluent 2021 R1." When using the combustor_830m model along with 128VM configuration, AMD's Milan-X performed better than Milan by 16% and Skylake by 131%. When using the OpenFOAM Motorbike benchmark test, Milan-X performed 60% and 305% better than Milan and Skylake using a VM configuration respectively.

Microsoft intends to continue to offer consumers an increase in performance on a linear level, a "gold standard in HPC, ... when performance increases linearly with cost in comparison to one VM." With these benchmark reports, Microsoft attempts to prove that with lower VM cost and higher solution times, using AMD's new EPYC Milan-X CPU technology is a win-win situation for Microsoft customers.