Just a few weeks ago, Amazon announced the AMD Epyc server thanks to Amazon Web Services (AWS). Now the company announces a new hardware platform for A1 just-generated instruction – Amazon Graviton processor.
He writes about the company:
We are currently launching an EC2-powered based Arm-based AWS Graviton processor. By building arm around arms and customized silicon use, A1 instances are optimized for performance and cost. Scale-out loading is great for workloads, where you can share payloads in a few cases. This includes containers for microservices, web servers, development environments and caches for fleets.
You can change the instances that use the scripting code without revoking their applications A1, but if you run your code natively, you will need to rebuild an A1 statement.
Everything we know about gravity we know here. It is based on the Cortex-A72, with a maximum clock speed of 2.3GHz. AWS VP James Hamilton writes:
These new examples cost 45% less and include 170 different instances supported by AWS, Intel-based z1d instances, a 4.0 GHz permanent frequency alert, 12-TB memory instance, F1 instance Family 8 Must be Gate Array Programs, P3 instance with NVIDIA Tesla V100 GPU and new M5a and R5a instances with AMD EPYC processors. No other cloud is offered.
In the new AWS design, Arm-based A1 instances are available in 5 different types of instances, with 1 core, 2 GB memory, 16 memory and 32 GB memory.
It's not clear exactly what kind of Amazon worked on the CPU. According to the Registry, 16 vCPU of each SoC are four sets, 2MB L2 cache per quad. Each individual core has 32KB L1 data cache and 48KB L1 instruction cache, that is, the standard Cortex-A72 configuration. A VCPU is mapped to one core of each CPU. The overall performance of the registry reports is quite referential. In some cases, 16 Graviton kernels were not capable of combining 5 Xeon E5-2697v4 5 colors, a Broadwell-class CPU.
Almost won AMD
It is easy to forget, but more than once, AMD is making a big bet ARM CPUs, not x86. AMD announced its K12 CPU in 2015, a pre-reserved label for "moniker-x86 chips" "K", which was chosen for the future nuclear ARM. K12, we said, would share a broad resource and development strategy with Zen. The platform x86-ARM platform was an ambitious project, the project called Skybridge. According to The Register, Amazon and AMD collaborated in 2015, "AMD failed with all Amazon landmarks milestones".
We do not know the milestone that AMD did not meet, but we chose several. The company's Skybridge project was abolished (we speculated that GF's manufacturing problems worsened at that time), and its Cortex-A57 CPU, Opteron A1100, announced in 2014 but was not sold until 2016. If we take it into account, we have guessed the problems with A1100. In future, AMD explained part of A1100's reason that the development of the ARM services infrastructure was not necessarily strong enough, and the software pilots needed more work to do. It could have been really true – ARM servers have marketed a few more years, than originally expected, but Amazon did not seem to want to wait. The company dumped AMD and purchased Annapurna Labs for design work.
This means why AMD overthrows K12. Lisa Su has focused on the semi-customized parts, and as a dedicated provider, Amazon's ARM company could definitely be the target. Skybridge, Seattle, or both, SkyDown, to show that AMD could market ARM IP, started a new cracked piece based on the same architectural buildings as x86 CPU. When AMD missed Amazon, ARM was able to lower its design and that's why Jim Keller was able to take part in the company. All this is speculation, to make it clear – the timeline is pretty well suited.
Unlike the AMD, this is the first time the ARM servers have been deployed in the cloud sequences, and therefore the only reason for the significant evolution of the ecosystem.
Now read: Amazon launches Windows App on Windows Store, Amazon May Split HQ2: One Virginia, One Maybe New York, and Amazon tried to sell ICE.