Azure HPC Cache Updates: New Caching Option, Discounted Pricing, and More!

New Releases 

Azure HPC Cache Premium Read-Write 

We’re excited to announce the preview of Azure HPC Cache Premium Read-Write. This next generation of premium caching for high-performance computing workloads is designed to provide high-bandwidth and low-latency access to files. Azure compute clients are provided with read and write performance like what they would experience from a local NVMe drive.

 

Premium HPC Cache provides lower latency than the Standard HPC Cache for your compute-intensive enterprise workloads. You can provision up to 84 TB of capacity in a single cache and point thousands of compute clients at the cache to get up to 20 GB/s of read throughput.  Premium HPC Cache’s highlights include:

• Increased Read Throughput: up to 20 GB/sec against an 80+ TiB dataset  
• Reduced Latency: 150 µsec for reads, 1 msec for writes
• Increased Write Throughput: 24.5% increase in write operations
• IOPS Scalability: 170,000 random 4 KiB writes; 450,000 random 4 KiB reads
  

With our lowest-latency cache, Premium HPC Cache provides our best file-based performance to meet your time-sensitive workloads like media rendering, simulations for genomics and financial models, as well as chip design. 

 

Getting Started with Premium Read-Write 

Azure HPC Cache Premium Read-Write is currently available as a Public Preview in select regions. If you are interested in participating in the preview, you can create a new cache and choose the Premium (Preview) option. 

 

Overview of Current and New Offerings

Attribute	Read-Write Standard Caching	Read-Only Caching	Read-Write Premium Caching
Throughput SKU	2, 4, or 8 GB/sec	4.5, 9, or 16 GB/sec	5, 10, or 20 GB/sec
Write Throughput	1.1, 2.2, 4.4 GB/sec	N/A GB/sec (write-through)	2.3, 4.6, 9.2 GB/sec
Read IOPS	62.5K, 125K, 250K ops/sec	160K, 320K, 480K ops/sec	500K, 1M, 2M ops/sec
Write IOPS	16.75K, 33.5K, 67K ops/sec	N/A ops/sec (write-through)	135K, 270K, 540K ops/sec
Cache sizes	3, 6, or 12 TB for 2 GB/sec  6, 12, or 24 TB for 4 GB/sec  12, 24, or 48 TB for 8 GB/sec	21 TB for 4.5 GB/sec  42 TB for 9 GB/sec  84 TB for 16 GB/sec	21 TB for 5 GB/sec  42 TB for 10 GB/sec  84 TB for 20 GB/sec
Maximum number of storage targets	20	20	20
Compatible storage target types	Azure Blob,  NFS (on-premises),   ADLS-NFS (NFSv3-enabled Azure Blob)	NFS (on-premises),  ADLS-NFS (NFSv3-enabled Azure Blob)	Azure Blob,  NFS (on-premises),   ADLS-NFS (NFSv3-enabled Azure Blob)
Caching styles	Read caching or read-write caching	Read caching only	Read-write caching
Cache can be stopped to save cost when not needed	Yes	No	No

*Results were without use of the priming feature. Additional details on priming will be included in our next blog post. 

 

Lower Pricing for Azure HPC Cache - Standard 

While it seems like the cost of everything is going up, we’re happy to report a price drop. As part of our commitment to provide the most cost-effective performance cache, we’re excited to share that we have dropped Azure HPC Cache – Standard prices by up to 33% percent in some regions. The new pricing is effective immediately. 

 

This enables cost-conscious verticals like media and entertainment to meet strict rendering deadlines while spending less on caching. Life sciences customers who rely on grants to fund their genomics research can now stretch their funds further. And chip designers can run their EDA workloads at a lower cost and still maintain the high performance for their tools repositories that they’ve come to expect. 

 

Terraform 

Terraform, an open-source software tool created by HashiCorp, provides an orchestration layer to deploy Azure resources. Using Terraform, you can deploy your own vdbench test system with all the required resources to run a performance benchmark. To try this yourself, the HPC Cache team has created a vdbench NFS-backed storage cache Terraform recipe to deploy a Premium 5G cluster and 30 clients. You can additionally run your own benchmarks against a 5G storage cache. Instructions and examples can be found on HPC Cache GitHub.