Ampere® Altra® Max and our Relentless Focus on Scaling Out


Naren Nayak – Sr. Director, Application Engineering

A fundamental sales pitch of the public cloud is that the cost to rent one CPU for a thousand hours is equivalent to renting a thousand CPUs for one hour. This ability to scale horizontally or ‘out’ in an elastic manner has been a big reason for the continuing growth in the cloud market. Performance can be exponentially improved by running applications over many CPUs simultaneously with no increase in total price.

Over the last decade, software has been re-architected to take advantage of this ability. Containerization and the move to microservice-based applications is one such paradigm to have come out of the “cloudification” of software. Microservices make it easier to scale out while drastically lowering the development-to-deployment time through innovations like DevOps and CI/CD. And this will only accelerate: IDC estimates that 80% of workloads will shift to containers/microservices by 2023.

Microservice-based usages and workloads can benefit greatly from increased core density, a trait the Ampere® Altra® already exhibits.

Ampere® Altra® delivers 80 cores with industry-leading performance that scales to provide the highest compute density in the cloud. In my previous blog, I showed the Ampere® Altra® outperforming the highest-end AMD EPYC processor across a variety of workloads, from integer benchmarks and in-memory caching to web serving and media encoding. Its performance lead against the Intel Xeon Scalable Processor family is even larger.

We’ve promised a rapid and predictable product cadence, something cloud service providers consider to be prerequisites to continuing their rapid growth trajectory. Ampere® Altra® Max is our next big leap forward and it is on track to ramp into production in just a matter of months. Ampere® Altra® Max is built for scale out while continuing our industry-leading core count streak – 128 cores this time. With 60% more cores, Ampere® Altra® Max is a significant improvement in scalability and is a perfect fit for cloud native architecture. It’s hard to imagine that less than four years ago, the industry-leading server processor had 32 cores!

We recently benchmarked some cloud workloads that are typically containerized on Ampere® Altra® Max in our labs and are excited to provide a preview.

Encryption

Our first example is encryption, a fundamental task in many cloud workloads. The Advanced Encryption Standard (AES) is a fast and secure symmetric block cipher that is widely used in the industry and by the US government. Our results show that Ampere® Altra® Max can encrypt 57% more blocks than Ampere® Altra® on AES-256-gcm with key sizes ranging from 16 to 16K bytes.

Media Encoding

Software media encoding continues to be a popular workload in the cloud and media companies like Netflix use the elasticity of the cloud very effectively to transcode video. h.264 is the most popular video codec used online today. Encoding media is a balance between quality and throughput – you want to encode as many clips as you can with a specific combination of profile, resolution, and bitrate.

The Ampere® Altra® Max can encode 55% more aggregate frames per second than Ampere® Altra®, which, in its own right, has already been demonstrated to lead the current x86 processors.

NGINX – Web Server

Most cloud applications have a front-end web server that fields incoming requests. According to Netcraft, 34% of all web sites run NGINX, a high performance, lightweight web server that can also be used as a reverse proxy, load balancer, mail proxy, and HTTP cache. The performance metric with web servers is usually throughput or requests per second under a specified Service Level Agreement (SLA) like a 99 percentile (or p.99) latency. In our testing, with a p.99 latency SLA of 10ms, Ampere® Altra® Max throughput was 51% higher than on Ampere® Altra®.

As you can see, not only does Ampere® Altra® Max deliver a massive increase in core count, but that scalability also translates into real performance gains of over 50%. With our rapid product cadence and groundbreaking annual performance increase, Ampere® is delivering a new class of processors that excels on modern workloads. The gains seen easily outpace those being delivered by other CPUs.

At Ampere®, we are firmly committed to the scale out philosophy in addition to the other cloud tenets – predictable high performance and energy efficiency. The Ampere® Altra® Max embodies all of those qualities and more. Stay tuned to hear more about the most performant and energy efficient cloud native processor over the coming months.



Related Post