Google Cloud Platform (GCP) announced the coming availability of its Arm-based instance, the Tau T2A, last week (currently available in preview) to address the ever-expanding needs of customers developing and deploying scale-out, cloud-native workloads. What does this announcement mean for enterprise IT? Does landing this final major cloud player fully validate Arm in the enterprise? And what motivated Google to jump on the Arm bandwagon? We'll address this a little more in the following paragraphs.
What was announced
The T2A virtual machine (VM) is part of the GCP Tau scale-out instance family. Tau is targeted at those cloud-native applications that run containerized or in VMs that don't require extreme compute resources. The Tau family was deployed initially with AMD EPYC (T2D) with fixed configurations to offer this instance type optimized for cost and scale-out performance.
The Tau T2A VM is based on Ampere Computing’s Altra CPU. It’s important to note that Azure announced Ampere and instances are GA at Oracle Cloud Infrastructure, as well as several Chinese clouds (including TikTok parent, ByteDance).
To motivate developers and customers, GCP offers a free 8-core, 32G RAM instance of T2A through general availability.
How Google is positioning T2A
One of the things I find with Arm announcements is that sometimes the "why would I use this?' question isn't fully answered. It's almost as if an assumption is made that enterprise IT professionals would fully understand the price-performance benefits of Arm and workload affinity.
Through the briefings Moor Insights & Strategy Patrick Moorhead and I received, as well as the various public statements from Google, it is refreshing to see the company is helping guide its customers. As mentioned, T2A is a VM targeting those scale-out workloads that don't require maximum compute resources at the individual instance level. Unsurprisingly, one of the supporting blogs from Google discusses optimizations for the Google Kubernetes Engine (GKE), Google's container environment.
A valuable capability of GKE is its multi-architecture support. So, containerized workloads can run in an x86 and Arm environment simultaneously. While this has many practical benefits, it also makes it easier for IT organizations to dip their collective toes in the “Arm” water, so to speak. It is capabilities such as this (not unique to GCP) that allow for organizations to deploy on Arm seamlessly.
It should be noted that T2A also runs the Google Container-optimized OS. So, organizations utilizing the popular Docker containers can expect full support.
Google has also enabled its Batch and Dataflow cloud services to run on T2A. These two services that target batch processing and streaming analytics respectively benefit from the Tau family's scale-out nature and T2A in particular.
While Google provides good guidance for its customers considering exploring or deploying on Arm, the use of T2A can be far broader. Independent of Google, Ampere has developed a robust ecosystem of partners, spanning the operating system to the workload. Functions like serverless caching via Momento, SLURM workload scheduling via SchedMD, and HPC through Rescale – are all optimized workloads for Ampere. And there are many more.
A few more details on T2A
Google is careful in how it positions its VM instances. When the company released its Tau VM family last year, it was very clear in positioning these as cost-effective, scale-out VMs. As one would expect with “cost-effective,” some options customers may prefer are lacking, such as local SSD support and higher bandwidth networking (32G supported in T2A v. 100G in other instances). Further, once a customer is locked into a T2A VM size (vCPU and RAM), they cannot dynamically add more resources.
Given the workloads targeted, the above makes sense, as customers look to distribute applications across many "good enough" performing VMs that don’t require maximum network throughput.
I like that GCP drives all of its specialized value into the Tau family, including T2A. The security measures, optimizations around memory (NUMA), network optimizations, etc.. that GCP has developed are all lit up in T2A. This level of support should assures customers utilizing T2A that these instances enjoy the same level of support as the highest performing compute engines.
Has Arm arrived in the enterprise?
The quick and simple answer is yes, though not for every workload. GCP announcing Arm-based instances rounds out support from all the major CSPs. This widespread support hasn't happened because Arm is cool or trendy. Nor has it happened as an exercise to drive better pricing from the x86 players. Arm is being deployed because CSPs can deliver equal or better performance for specific workloads at a lower cost and power envelope. Period. This is basic economics.
While Arm is not going to replace x86 to run virtualized infrastructure on VMware anytime soon, there are still use cases where Arm is a good fit. In its blog promoting T2A, one of GCP's reference customers is Harvard University. The school runs several compute-intensive workloads on SLURM VirtualFlow, and T2A allows it to run tens of thousands of VMs in parallel, reducing compute time significantly. But here’s the key to what Harvard had to say – the migration to T2A was done with minimal effort. Such is the beauty of cloud-native development. The cost and time savings will be immediately recognized.
I like this Harvard reference because it reminds us that Arm is not just for the digitally born companies that have never had an on-premises datacenter. It's for any company embarking on a digital transformation or modernization project.
Further proof of Arm's move into the datacenter can be seen in HPE's announcement of the upcoming ProLiant RL300 Gen11 server based on Ampere's Altra CPU. This is the first mainstream server that HPE has announced ahead of its Gen11 launch, and I expect the market will see competitors roll out its servers in time.
Is T2A just a competitive response from Google?
I don’t believe that Google is interested in investing in and rolling out an Arm-based instance to be like every other cloud provider. GCP is run by many intelligent people who firmly understand its customers' wants and needs.
As a company, Google has deep roots in silicon design, development, and optimizations. It's no secret that the company works with CPU vendors to deliver Google compute-optimized platforms. I think GCP has done its due diligence in ensuring the Ampere CPU could and would meet its particular and the needs of its customers.
I believe my only question is around the longer-term strategy for Google and Arm. There are two camps in the CSP space: those that design its silicon (i.e., AWS Graviton) and those that deploy Ampere. Given Google's history in silicon development, could we see a custom chip in the future? It is a scenario that is entirely plausible.
Google rounds out support for Arm from the major CSPs with its Tau T2A VM offering, based on Ampere Computing’s Altra CPU. While the company is last to market in this regard, it has done a thorough job of positioning Arm relative to x86 and target workloads.
I believe this is just the beginning for Arm at GCP and suspect the company will eventually roll Arm offerings into other compute engine offerings over time. But I think it will do this in a very measured way, looking for areas where Arm can offer a differentiated experience for customers.
It's a good time to be a proponent of Arm. And a better day to be an investor of Ampere Computing. There is no doubt that Arm is here to stay. Not as a cheap alternative to x86, but as an architecture that can be optimized for many workloads, with the ability to lead in raw performance, price-performance, and performance-per-watt, at a time when each of these measures are so critical.
Note: Moor Insights & Strategy writers and editors may have contributed to this article.