Turbonomic Blog

Unlocking EPYC Performance in Your Cloud Environment

Posted by Rick Ochs on Aug 27, 2020 1:00:00 PM

The cloud is evolving at a furious pace.

Cloud professionals are now facing an unprecedented amount of options in the public cloud. For example, when users deploy a workload on Amazon EC2 service (which is one of the hundreds of services offered by AWS), they are faced with millions of possible combinations of options. AWS offers 77 availability zones (across 26 regions) and over 340 instance types (as of August 2020), some of the instances types are based on Intel CPU, some on AMD and some are ARM-base. On top of that, there are the various OS options, five EBS storage options (the new io2 introduced late August 2020), two tenancy options, more than a dozen of purchasing options (On-Demand, Spot, reservation (RIs and Savings plan) and more.) Like we said, millions of options!

The sheer amount of configuration options when deploying or scaling a VM is no different with Microsoft’s cloud platform - just take a look at the number of choices available when creating a virtual machine on Azure:

AzureVMTypes_image

Selecting a VM size among the many hundreds of choices is a full-time job. With the latest releases of VM types/Instance type families, we have even more performance options at our fingertips. Within the CPU industry, AMD has led the charge on innovation - specifically with their EPYC 2nd generation chips which utilize a new 7 nanometer process to unlock large efficiency and performance gains. These are available for cloud users through Azure Dav4-series and Eav4-series and Amazon EC2 C5a and C5ad instance types.

In this article, we will focus on Azure’s AMD-based EPYC Gen-2 VM types that were released earlier this year including Dav4-series, Eav4-series and others.

Azure CPU Benchmarking

Microsoft’s Azure engineering team assigns performance values to each VM family, called ACUs - or Azure Compute Units. ACUs are essentially distilled benchmarks designed to provide comparisons of performance between different families. These values factor in elements such as multithreading, cache, IPC efficiency, and much more.

The Azure team has placed an ACU value range on AMD’s EPYC Gen 2 family of 230-260, one of the highest ranges for a CPU.

Multithreaded CPUs generally score lower in ACU value, because they have half the number of full, physical cores per vCPU than their non-multithreaded counterparts. In this case, the new AMD EPYC Gen 2 CPUs are multithreaded, but still put down higher ACU values than non-hyperthreaded high power Intel CPUs, such as the highly adopted Dv2 family, which can use up to Intel Skylake chips.

Our independent benchmarking work has validated these numbers – we have noted that Microsoft is quite safe and conservative in their performance evaluations. Interestingly enough, the Dv2 family is noted as a reasonably large ACU range higher than the Dv3 family, despite using the same processors, due to hyperthreading being enabled on the Dv3 family. This ends up reducing average core performance when hyperthreaded cores are counted. Note that it does lessen the per-core cost in the Dv3 family slightly compared to the Dv2 family.

When we compare the cost of the brand new Dv4a family, we find it is roughly 16% higher for the same core count, compared to the older Dv3 family, yet a whopping THIRTY PERCENT faster on average. Per core.

If you apply this extra CPU performance capacity across an enterprise IT environment, you can consistently acquire more application performance with no additional cost. This allows you to run ~30% fewer cores while increasing single threaded performance at the same time! At the enterprise scale with tens of thousands of VMs, that translates into meaningful application performance improvements and significant gains in how much spend is required to achieve true application performance.

Assuring EPYC Performance with Turbonomic

Here at Turbonomic, we believe, above all else, that application performance is paramount. All of our product’s scaling actions are performance actions (including scale down actions for cost). They just so happen to also increase cost efficiency, because a performant VM is not a wasted VM. AMD’s latest EPYC Gen 2 processors unlock a whole new opportunity to achieve high performance within your budget.

AMD’s latest Azure SKUs are available in Turbonomic v7.22.7 and higher and allow our customers to enjoy the benefits of AMD’s latest CPU offerings and the excellent cost-to-performance ratio they provide.

For example, the below image was captured from an older Turbonomic version, which didn’t have support for AMD’s latest offerings yet. In this example, Turbonomic selected to scale the instance to the best Azure SKU available, Standard_F8. The action will provide the VM with 87.4 GHz (corrected) CPU speed and will result in $203.67 monthly savings (0.279*730).

2-Aug-25-2020-08-56-38-33-PM

In comparison, the same VM in Turbonomic v7, which supports the latest Azure AMD-based SKUs, generated a scale action to an EPYC-based Standard_D8as_v4 VM type, which will provide the VM with 90.9 (corrected) GHz CPU speed and result in $214 monthly savings.

3-Aug-25-2020-08-56-38-31-PM

The above example showcases the value of the latest-gen VM types - more performance at a lower cost!

If you want to enjoy the latest cloud SKUs and many more new features and enhancements such as the CPU and Memory time-series graphics you noticed in the image above, make sure to upgrade your Turbonomic server to the latest version of Turbonomic (v7 or higher).

Want to experience the latest Turbonomic version? Click here to request a Turbonomic SaaS Instance for your cloud environment.

Request a Turbonomic SaaS Instance

 

Subscribe Here!

Recent Posts

Posts by Tag

See all