EC2 Accelerated Computing Instances

AWS EC2 Accelerated Computing Instances

Accelerated computing instances are mostly utilized by companies who make use of the following:

– Hardware accelerators

– Co-processors

– Functions

– Floating point number calculations

– Graphics processing

– Data pattern matching,

These operations are even performed more efficiently and effectively than what other software running on CPUs are capable of.

 

P3:

They are the newest generation of general-purpose GPU instances.

Its Features Include:

• 8 NVIDIA Tesla V100 GPUs [pairing 5,120 CUDA Cores + 640 Tensor Cores]
• Intel Xeon E5-2686 v4 [High frequency] processors (p3.2xlarge+ p3.8xlarge + p3.16xlarge)
• 5 GHz Intel Xeon P-8175M [High frequency] processors (p3dn.24xlarge)
• NVLink: peer-to-peer GPU communication
• Offer 100 Gbps aggregate network bandwidth
• EFA: p3dn.24xlarge instances

 

Instance	GPUs	vCPU	Mem (GiB)	GPU Mem (GiB)	GPU P2P	Storage (GB)	Dedicated EBS Bandwidth	Networking Performance
p3.2xlarge	1	8	61	16	–	EBS-Only	1.5 Gbps	Up to 10 Gigabit
p3.8xlarge	4	32	244	64	NVLink	EBS-Only	7 Gbps	10 Gigabit
p3.16xlarge	8	64	488	128	NVLink	EBS-Only	14 Gbps	25 Gigabit
p3dn.24xlarge	8	96	768	256	NVLink	2 x 900 NVMe SSD	19 Gbps	100 Gigabit

Every instance has the bellow specs:

• Intel AVX + Intel AVX2 + Intel Turbo
• It is EBS Optimized
• It has Enhanced Networking

 

p3.2xlarge + p3.8xlarge + p3.16xlarge: include 2.3 GHz and 2.7 GHz Intel Xeon E5-2686 v4 processors.

p3dn.24xlarge: 2.5 GHz and 3.1 GHz Intel Xeon P-8175M processors + supports Intel AVX-512.

p3dn.24xlarge: EFA that allows for HPC apps that use the MPI and ML apps through NCCL for scaling to thousands of GPUs.

Used for the below Cases:

– Machine/Deep learning

– High performance computing

– Computational fluid dynamics

– Computational finance

– Seismic analysis

– Speech recognition

– Autonomous vehicles

– Drug discovery

 

 

P2:

Those instances are made specifically for general-purpose GPU compute apps.

Its Features Include:

• Intel Xeon E5-2686 v4 processors (High freq.)
• NVIDIA K80 GPUs (High-performance) (2,496 parallel processing cores + 12GiB GPU memory)
• Works with GPUDirect™ peer-to-peer GPU communications
• Offers Enhanced Networking through ENA (25 Gbps aggregate network bandwidth) within a Placement Group
• It is by default EBS-optimized for no extra cost

Instance	GPUs	vCPU	Mem (GiB)	GPU Memory (GiB)	Network Performance
p2.xlarge	1	4	61	12	High
p2.8xlarge	8	32	488	96	10 Gigabit
p2.16xlarge	16	64	732	192	25 Gigabit

Every instance has the bellow specs:

• 2.3 GHz + 2.7 GHz Intel Xeon E5-2686 v4 Processor
• Intel AVX + Intel AVX2 + Intel Turbo
• It is EBS Optimized
• It has Enhanced Networking

Used in the following Cases:

– Machine learning

– High performance databases

– Computational fluid dynamics

– Computational finance

– Seismic analysis

– Molecular modeling

– Genomics

– Rendering

– Other server-side GPU compute workloads

 

Inf1:

These instances are made up from the ground up for the goal of supporting machine learning inference apps.

Its Features Include:

• 16 Inferentia Chips
• Neuron SDK
• Second Gen Intel® Xeon® Scalable processors (High Frequency)
• 100 Gbps networking

Instance size	vCPUs	Memory (GiB)	Storage	Inferentia chips	Inferentia chip-to-chip interconnect	Network bandwidth	EBS bandwidth
inf1.xlarge	4	8	EBS only	1	N/A	Up to 25 Gbps	Up to 4.75 Gbps
inf1.2xlarge	8	16	EBS only	1	N/A	Up to 25 Gbps	Up to 4.75 Gbps
inf1.6xlarge	24	48	EBS only	4	Yes	25 Gbps	4.75 Gbps
inf1.24xlarge	96	192	EBS only	16	Yes	100 Gbps	19 Gbps

 

Used in the below Cases:

– Recommendation engines

– Forecasting

– Image and video analysis

– Advanced text analytics

– Document analysis

– Voice

– Conversational agents

– Translation

– Transcription

– Fraud detection

 

G4:

These instances are developed for the sake of aiding the acceleration of machine learning inference + graphics-intensive workloads.

Its Features Include:

• Second Generation Intel Xeon Scalable processors
• NVIDIA T4 Tensor Core (GPUs)
• 100 Gbps networking throughput
• 8 TB local NVMe storage

	Instance	GPUs	vCPU	Mem (GB)	GPU Memory (GB)	Instance Storage (GB)	Network Performance (Gbps)
Single GPU VMs	g4dn.xlarge	1	4	16	16	125	Up to 25
	g4dn.2xlarge	1	8	32	16	225	Up to 25
	g4dn.4xlarge	1	16	64	16	225	Up to 25
	g4dn.8xlarge	1	32	128	16	1×900	50
	g4dn.16xlarge	1	64	256	16	1×900	50
Multi GPU VMs	g4dn.12xlarge	4	48	192	64	1×900	50
	g4dn.metal*	8	96	384	128	2×900	100

Every instance has the below specs:

• Cascade Lake 24C processors (2.5 GHz)
• Intel AVX + Intel AVX2 + Intel AVX-512 + Intel Turbo
• It is EBS Optimized
• It has Enhanced Networking

Used in the following Cases:

Machine learning inference for apps such as:

– The addition of metadata to an image

– Object detection

– Recommender systems

– Automated speech recognition

– Language translation

Those instances offer a truly cost-efficient platform for the sake of building + running graphics-intensive apps, including:

– Remote graphics workstations

– Video transcoding

– Photo-realistic design

– Game streaming in the cloud

 

G3:

They are specifically developed for graphics-intensive apps.

Its Features Include:

• Intel Xeon E5-2686 v4 processors (High Frequency)
• NVIDIA Tesla M60 GPUs, every one of which has 2048 parallel processing cores + 8 GiB video memory
• Allows for NVIDIA GRID Virtual Workstation features
• Supports 4 monitors + resolutions 4096×2160
• Every GPU found in the instance is licensed for 1 Concurrent Connected User
• Allows NVIDIA GRID Virtual Application abilities for app virtualization software
• Supports 25 concurrent users/GPU
• Every one of the GPU features an on-board hardware video encoder
• Supports 10 H.265 1080p30 streams + 18 H.264 1080p30 streams
• Allows low-latency frame capture + encoding
• Experiences with high-quality interactive streaming
• Enhanced Networking: ENA + 25 Gbps of aggregate network bandwidth (Placement Group)

Instance	GPUs	vCPU	Mem (GiB)	GPU Memory (GiB)	Network Performance
g3s.xlarge	1	4	30.5	8	Up to 10 Gigabit
g3.4xlarge	1	16	122	8	Up to 10 Gigabit
g3.8xlarge	2	32	244	16	10 Gigabit
g3.16xlarge	4	64	488	32	25 Gigabit

Every instance has the below specs:

• 2.3 GHz + 2.7 GHz  Intel Xeon E5-2686 v4 Processor
• Intel AVX + Intel AVX2 + Intel Turbo
• It is EBS Optimized
• It has Enhanced Networking

Used in the below Cases:

– 3D visualizations

– Graphics-intensive remote workstation

– 3D rendering

– App streaming

– Video encoding

– Other server-side graphics workloads.

 

F1:

This type of instances provides customizable hardware acceleration going along with some FPGAs.

Its Instances Features Include:

• Intel Xeon E5-2686 v4 (High frequency) processors
• NVMe SSD Storage
• Enhanced Networking

Its FPGA Features Include:

• Xilinx Virtex UltraScale+ (VU9P FPGAs)
• 64 GiB of ECC-protected memory (4x DDR4)
• PCI-Express (x16 interface)
• About 2.5 mill logic elements
• About 6,800 DSP engines
• FPGA Developer (AMI)

Instance	FPGAs	vCPU	Mem (GiB)	SSD Storage (GB)	Networking Performance
f1.2xlarge	1	8	122	470	Up to 10 Gigabit
f1.4xlarge	2	16	244	940	Up to 10 Gigabit
f1.16xlarge	8	64	976	4 x 940	25 Gigabit

f1.16xlarge instances: PCI-e fabric allows the FPGAs to share the exact memory space + communicate with one another throughout the fabric at 12 Gbps in every direction.

Every instance has the below specs:

• 2.3 GHz + 2.7 GHz Intel Xeon E5-2686 v4 Processor
• Intel AVX + Intel AVX2 + Intel Turbo
• It is EBS Optimized
• It has Enhanced Networking

Used in the following Cases:

– Genomics research

– Financial analytics

– Real-time video processing

– Big data search + analysis + security

ec2 storage optimized instances