Amazon Launches ARM-Based Custom Graviton Processor



[ad_1]

This site may generate affiliate commissions from the links on this page. Terms of use.
servers

It has been only a few weeks since Amazon announced that it would offer optional Epyc AMD servers for Amazon Web Services (AWS). The company is now announcing a new type of hardware platform for its newly created A1 instances: the Amazon Graviton processor.

The company writes:

Today, we are launching EC2 instances based on Arm-based AWS Graviton processors. Built around Arm cores and making extensive use of custom silicon, A1 instances are optimized in terms of performance and cost. They are ideal for scalable workloads in which you can share the load on a group of smaller instances. This includes containerized microservices, web servers, development environments, and caching fleets.

ARM-Graviton

Instances that use script code can move their applications to A1 without rewriting, but if your code runs natively, you will have to rebuild it for an A1 instance.

The Graviton processor

Here is what we know about the Graviton so far. It is based on the Cortex-A72, with a maximum clock speed of 2.3 GHz. James Hamilton, Vice President of AWS, writes:

These new instances offer up to 45% lower costs and will join the 170 different types of instances supported by AWS, ranging from Intel-based z1d instances that provide a maintained base frequency of 4.0 GHz , a 12TB memory instance, the F1 family instance with up to 8 on-site programmable interface matrices, P3 instances with NVIDIA Tesla V100 GPUs and the new M5a and R5a instances with AMD EPYC processors . No other cloud offer even comes close.

ARM-Graviton-Small

graviton

The new AWS-based Arm based A1 instances are available in 5 different instance types, from a core with 2 GB of memory and up to 16 cores with 32 GB of memory.

The type of custom work that Amazon does on the processor is unclear. According to the register, the 16 vCPU constituting each SoC are divided into groups of four, with a L2 cache of 2 MB shared between each quad. Each core has a 32-KB L1 data cache and a 48-KB L1 instruction instruction cache, a standard Cortex-A72 configuration. A vCPU is mapped to a physical processor core. The registry indicates that the overall performance is quite varied. In some cases, 16 Graviton kernels could not match 5 kernels of a Broadwell-clbad Xeon E5-2697v4.

Almost an AMD victory

It's easy to forget, but once upon a time, AMD was betting heavily on ARMS Processors, not x86. When AMD announced its K12 processor in 2015, it chose to give the nickname "K" – a label previously reserved for x86 chips – to a future ARM core. K12, we were told, would share important resources and a development strategy with Zen. It was planned to create an ambitious common joint x86-ARM platform, called Project Skybridge. According to The Register, Amazon and AMD worked closely together in 2015, until "AMD failed to meet all performance milestones set by Amazon."

AMD failed to reach certain milestones, but we were able to choose several. The company's Skybridge project was suddenly canceled (we had then badumed that GF's manufacturing problems could have contributed to this problem) and its Cortex-A57 processor, the Opteron A1100, announced in 2014, does not have the same effect. was not delivered until 2016. If we had to guess, we would guess that the problems were with the A1100. At the time, AMD had explained that the A1100 had been delayed in part because the infrastructure needed to deploy ARM servers was not as strong as it needed to be, and additional efforts were needed to rectify the stack of software. That may be quite true – ARM servers took years to come on the market longer than expected – but Amazon apparently did not want to wait. The company abandoned AMD and purchased Annapurna Labs for its design work.

It may also explain why AMD has shelved K12. Lisa Su focused on semi-custom parts, and serving as a vendor dedicated to Amazon's ARM activities could certainly suit this target. In this speech, Skybridge, Seattle or both were test balls to demonstrate that AMD could put ARM IP on the market before launching a new custom part based on some of the same architectural building blocks as its x86 processor. When AMD lost the Amazon opportunity, it may have moved on beyond its ARM concept, which may explain why Jim Keller left the company in its place. All this is only speculation, to be clear: it corresponds quite well to the chronology.

Whatever the result with AMD, this is the first time we've seen ARM servers deployed in cloud instances, which represents a significant advance for the ecosystem for this reason alone.

Now read: Amazon launches Alexa application in the Windows Store, Amazon May Split HQ2: a company in Virginia, another in New York and Amazon is trying to sell to ICE a defective facial recognition technology

[ad_2]
Source link