We are announcing the general availability of the Azure FX-series Virtual Machines available in three regions. Azure FX-series Virtual Machines—based on the 2nd Generation Intel Xeon Scalable Processor—feature a high-performing central processing unit (CPU) clock speed per single core of up to 4 GHz (all core turbo), 21 GB memory per vCPU, and local temporary SSD disks. The high-frequency CPU and memory capacity for FX-series Virtual Machines is designed for workloads demanding very high single-threaded and lightly threaded performance, such as Electronic Design Automation (EDA).
More Info: PL-900: Microsoft Power Platform Fundamentals
Azure FX-series Virtual Machines are now generally available in the West US 2, West Europe, and Japan East regions, and will come to more Azure regions to meet customer demand.
Fast and cost-efficient FX virtual machine for EDA workloads
EDA is a collection of complex software tools and workflows to design semiconductor integrated circuits. EDA customers have several key challenges:
◉ EDA simulations are compute-intensive jobs with each simulation running from hours to days depending on the complexity of the design. Additionally, some EDA processes are single-threaded. As a result, customers need fast CPU clock speed and a large memory footprint to accelerate their silicon design into fabrication and testing.
◉ EDA software license costs can be several times that of infrastructure costs and often licensed per CPU core. Consequently, EDA users want to lower their licensing costs by maximizing the performance of each CPU core, thereby enabling them to use a smaller number of CPU cores.
◉ EDA designs usually start with different groups working on smaller subsets of a larger design. Small subsets and blocks then need to be merged for additional verification checks at multiple points throughout the process. That verification run, especially in back-end design, demand a high memory footprint.
The Azure FX-series Virtual Machine is designed to solve these EDA workload challenges. The series is powered by Intel Xeon Scalable Processors with a single-core frequency of up to 4.1 GHz, with a base frequency of 3.4 GHz, and an all-core-turbo frequency of 4.0GHz. It delivers up to 19 percent better performance per CPU core over current generation Compute Optimized Virtual Machines. When customers need additional per-CPU core performance, they can turn off hyper-threading by contacting the customer support team. With 1 TB of total memory and 21 GB RAM per virtual CPU core, FX-series Virtual Machines can meet a majority of front-end and back-end design needs. In addition, all the FX-series Virtual Machine sizes come with a local SSD so that memory can read data locally instead of accessing a slower remote Network File System (NFS).
To demonstrate FX-series Virtual Machine performance on the real EDA workloads, we ran an EDA backend physical verification task on E48sv4 and F48mds separately. It took 12.7 hours to complete the workload on E48sv4 versus 11.5 hours on F48mds Virtual Machine. In short, FX-series Virtual Machines shorten the EDA simulation time by 10 percent.
0 comments:
Post a Comment