The Cloud Regions Map is a Complimentary Tool
for Exploring Regional CSP Service Availability
Regions with larger CMRG are shown with larger icons. Positive and zero CMGR values are shown by a solid circle, negative CMGR values are shown by a solid square. Every active CSP region has a CMGR value.
Circle size denotes the number of accelerator classes available in a region. A small empty circle indicates that no accelerators are present in that region.
CNA (Custom Neural Accelerator) is designed to optimally accelerate specific classes of deep learning (DL) and other artificial intelligence (AI) algorithms.
- FPGA (Field Programmable Gate Array) is designed to be programmed at a hardware level to optimize performance for specific algorithms that are not well suited to GPUs or general-purpose processors.
- GPU (Graphics Processing Unit) is designed to accelerate graphics rendering and other highly parallel matrix math calculations, like high-performance computing (HPC).
- MOE (Math Offload Engine) is designed to accelerate broad classes of math calculations, much like a GPU but without the graphics bits.
- Transcoder is designed to assist in media processing, such as transcoding video from one format or size to another.
We record many details about tracked CSPs’ IaaS accelerator options. We show accelerator chip vendor brands and general model numbers of the chips offered by tracked CSPs in each region. See below for a complete list of accelerator manufacturers and brands currently deployed by tracked CSPs.
Circle size denotes the number of accelerator brands available in a region. A small empty circle indicates that no accelerators are present in that region.
- Aliyun NPU 800
- AMD FirePro S7150
- AMD MI25
- AMD Radeon Pro V520
- AMD Radeon Instinct MI25
- AMD / Xilinx Alveo U250
- AMD / Xilinx Alveo U30
- AMD / Xilinx Virtex Utrascale + VU9P
- AWS Elastic Graphics
- AWS Elastic Inference
- AWS Inferentia
- Google Cloud TPU v2
- Google Cloud TPU v3
- Intel Arria GX 10
- Intel H3C XG310
- Intel Habana Gaudi HF-205
- NVIDIA GRID K520
- NVIDIA Tensor Core Ampere A10
- NVIDIA Tensor Core Ampere A100
- NVIDIA Tensor Core Ampere A10G
- NVIDIA Tensor Core Turing T4
- NVIDIA Tensor Core Turing T4G
- NVIDIA Tensor Core Volta V100
- NVIDIA Tesla Kepler K80
- NVIDIA Tesla Maxwell M40
- NVIDIA Tesla Maxwell M60
- NVIDIA Tesla Pascal P100
- NVIDIA Tesla Pascal P4
- NVIDIA Tesla Pascal P40
We record many details about tracked CSP’s IaaS processor options. We show processor chip vendor brands and the generations of those brands offered in each region. See below for a complete list of processor brands currently deployed by tracked CSPs and a list of which processor generation code names we compare to each other.
More processor brands available in a region are shown with larger circles.
Intel Xeon Ice Lake
Intel Xeon Cooper Lake
Intel Xeon Cascade Lake
Intel Xeon Skylake
Intel Xeon Broadwell
Intel Xeon Haswell
Intel Xeon Intel Xeon Sandy/Ivy Bridge
AMD EPYC "Milan"
AMD EPYC "Rome"
AMD EPYC "Naples"
AWS Graviton 3
AWS Graviton 2
Zones, also known as availability zones, are physically separated locations within a CSP's region. By providing separate zones, CSP's can mitigate failures at any one zone within a region. Failures can result from issues with software and hardware, but also from natural events such as floods or fires. Some regions may have only one zone, but larger regions can have five or more regions to support their customers.
Circle size is relative to the number of zones in the region. A larger circle represents more zones.
Consistent with Zones count, this measure identifies all the regions with more than one zone, thereby having the ability to offer redundancy within the region. Redundancy is important for companies wanting to mitigate their applications for failures within any given zone.
Circle size is relative to the number of zones in the region. Unlike the zone count, the redundancy only shows a large circle for any region with more than one zone within the region.
The Compute Cost Index allows from easy comparison between regions for general price. Any product may vary from one region to another, but this index provides a reasonable guide for general price differences between regions. The price is based on the geometric mean of the on-demand price in US Dollars for 8-core compute instances within a given region. The prices are measured across all cloud providers for easy comparability. The lowest value is 1.0. All other regions are priced relative to that lowest-priced region.
Circle size is relative to the calculated price. A larger circle represents a higher price.