Oracle continues to build on its offering of AMD instances in Oracle Cloud Infrastructure (OCI). The new OCI Compute E5 instances are based on the 4th generation AMD EPYC processors and come in three different flavors. We briefly introduce them to you in this article.
The partnership between Oracle and AMD is now five years old. During what was then called Oracle OpenWorld (now CloudWorld) in 2018, the companies announced the first generation of OCI instances based on AMD processors. That generation went by the name E2. Today, in the fifth year of the partnership, Oracle announces OCI Compute E5 instances. The plan is to make them available to customers in the second half of 2023.
OCI Compute E5 Standard, Dense-IO and HPC
The OCI Compute E5 instances will be available in three variants. There is OCI Compute E5 Standard, for everyday workloads. There is also OCI Compute E5 Dense-IO for hosting and working on large databases, such as for big data projects. Finally, there is OCI Compute E5 HPC, for high performance compute workloads such as AI training, analytics, simulations and cluster file systems. Below you can see the features of the different instances conveniently laid out in a diagram:
The performance of the E5 instances is a lot better than the E4 instances, OCI’s previous generation of AMD instances. Furthermore, you can see above that the instances are also available in different types. Oracle calls these compute shapes. With E5 Standard and E5 Dense-IO, you have the choice of bare metal and Flex VM. With E5 HPC, you can only choose bare metal. The latter is not surprising, because with HPC workloads you want/need to be as close to the metal as possible. That’s when you get the best performance. Hence, E5 HPC has RDMA as a feature. With this you bypass the CPU and address the much faster memory directly. That’s what you want if your main concern is performance.
Looking also briefly at the other features (besides RDMA) offered by Oracle, we see most of them in the standard instances. Most of them speak for themselves. Burstable means that you can scale the number of cores up and down, shielded instances are provided with extra security, confidential computing makes it possible to do encryption in memory and capacity reservations means that there is always capacity available should the need arise. Think of disaster recovery, among other things. For this, having the capacity, even though you’re not using it, is crucial.
The only feature that may not quite speak for itself is pre-emptible. These are instances where no guarantees of availability are needed and given. They can be deployed for something else at any time. This makes them relatively inexpensive, but also suitable for a limited number of use cases. Oracle mentions test environments in its own documentation as an example of a use case. These can always be resumed later.
Flex VM compute shape for ‘right-sizing’
Flex VM is an interesting concept to dwell on a little longer. Leo Leung, VP of Product and Strategy at Oracle, regularly talks about ‘right-sizing’ with Flex VM during a briefing we had with him about today’s launch. What you can do with this is use exactly the number of compute cores you need for a workload. This is how Oracle is trying to differentiate itself. Other cloud instance providers operate according to what Leung calls T-shirt sizing. These offer a more limited number of fixed choices, where you purchase a range of cores, not a specific number. That means there’s a chance you’re paying for cores you don’t use. Oracle wants to avoid that, according to Leung, by offering Flex VM.
The savings for customers using AMD instances in conjunction with Flex VM are substantial, Leung points out. He talks about OCI customers together already saving about $40 million a year. That’s a hefty sum, although we can’t really judge whether this is really a big savings in relative terms. After all, that also has to do, among other things, with the number of customers we are talking about and what the total amount is on which these savings are realized.
All in all, with today’s announcement, Oracle expands the range of OCI instances a little further. While it is a natural progression (from E4 to E5), it is one that gives customers more options once again. In any case, if the claimed price/performance of the new OCI Compute E5 instances based on the 4th generation AMD EPYC processors is true, it is another big step in that area.