HBv4-series VMs are optimized for various HPC workloads such as computational fluid dynamics, finite element analysis, frontend and backend EDA, rendering, molecular dynamics, computational geoscience, weather simulation, and financial risk analysis. HBv4 VMs feature up to 176 AMD EPYC™ 9V33X ("Genoa-X") CPU cores with AMD's 3D V-Cache, clock frequencies up to 3.7 GHz, and no simultaneous multithreading. HBv4-series VMs also provide 768 GB of RAM, 2.3 GB L3 cache. The 2.3 GB L3 cache per VM can deliver up to 5.7 TB/s of bandwidth to amplify up to 780 GB/s of bandwidth from DRAM, for a blended average of 1.2 TB/s of effective memory bandwidth across a broad range of customer workloads. The VMs also provide up to 12 GB/s (reads) and 7 GB/s (writes) of block device SSD performance.
All HBv4-series VMs feature 400 Gb/s NDR InfiniBand from NVIDIA Networking to enable supercomputer-scale MPI workloads. These VMs are connected in a non-blocking fat tree for optimized and consistent RDMA performance. NDR continues to support features like Adaptive Routing and the Dynamically Connected Transport (DCT). This newest generation of InfiniBand also brings greater support for offload of MPI collectives, optimized real-world latencies due to congestion control intelligence, and enhanced adaptive routing capabilities. These features enhance application performance, scalability, and consistency, and their usage is recommended.
1Temp disk speed often differs between RR (Random Read) and RW (Random Write) operations. RR operations are typically faster than RW operations. The RW speed is usually slower than the RR speed on series where only the RR speed value is listed.
Storage capacity is shown in units of GiB or 1024^3 bytes. When you compare disks measured in GB (1000^3 bytes) to disks measured in GiB (1024^3) remember that capacity numbers given in GiB may appear smaller. For example, 1023 GiB = 1098.4 GB.
Disk throughput is measured in input/output operations per second (IOPS) and MBps where MBps = 10^6 bytes/sec.
1Some sizes support bursting to temporarily increase disk performance. Burst speeds can be maintained for up to 30 minutes at a time.
Storage capacity is shown in units of GiB or 1024^3 bytes. When you compare disks measured in GB (1000^3 bytes) to disks measured in GiB (1024^3) remember that capacity numbers given in GiB may appear smaller. For example, 1023 GiB = 1098.4 GB.
Disk throughput is measured in input/output operations per second (IOPS) and MBps where MBps = 10^6 bytes/sec.
Data disks can operate in cached or uncached modes. For cached data disk operation, the host cache mode is set to ReadOnly or ReadWrite. For uncached data disk operation, the host cache mode is set to None.
Expected network bandwidth is the maximum aggregated bandwidth allocated per VM type across all NICs, for all destinations. For more information, see Virtual machine network bandwidth
Upper limits aren't guaranteed. Limits offer guidance for selecting the right VM type for the intended application. Actual network performance will depend on several factors including network congestion, application loads, and network settings. For information on optimizing network throughput, see Optimize network throughput for Azure virtual machines.
To achieve the expected network performance on Linux or Windows, you may need to select a specific version or optimize your VM. For more information, see Bandwidth/Throughput testing (NTTTCP).
Accelerator (GPUs, FPGAs, etc.) info for each size