A shape is a template that determines the number of OCPUs , amount of memory, and other resources that are allocated to an instance. Compute shapes are available with AMD processors, Intel processors, and Arm-based processors.
This topic provides basic information about the shapes that are available for bare metal instances, virtual machines (VMs), and dedicated virtual machine hosts. For information about service limits, see Compute Limits. For information about protecting data on NVMe devices, see Protecting Data on NVMe Devices.
Note
When a new region becomes available, it might take a few weeks before host capacity also becomes available.
To obtain a list of shapes available to you, run the ListShapes operation.
Oracle measures compute resource pricing differently. The Oracle CPU (OCPU) represents physical CPU cores and is the unit of measurement for CPUs on x86 CPUs (AMD and Intel) and Arm CPUs (OCI Ampere Compute). A virtual CPU (vCPU), the industry-standard for measuring compute resources, represents one execution thread of a physical CPU core.
Most CPU architectures, including x86, runs two threads per physical core, so one OCPU is the equal of two vCPUs for x86-based compute. For OCI Compute, the minimum unit of provisioning starts from one OCPU on both X86 (Intel and AMD) and OCI Ampere Compute processors.
The following are the provisioning units for compute instances:
1 OCPU on Arm A1 (Compute) = 1 core on Arm A1 (Compute) or 1 vCPU
1 OCPU on Arm A2 (Compute) = 2 cores on Arm A2 (Compute) or 2 vCPUs
A flexible shape is a shape that lets you customize the number of OCPUs and the amount of memory when launching or resizing your VM. When you create a VM instance using a flexible shape, you select the number of OCPUs and the amount of memory that you need for the workloads that run on the instance. The network bandwidth and number of VNICs scale proportionately with the number of OCPUs. This flexibility lets you build VMs that match your workload, enabling you to optimize performance and minimize cost.
The flexible shapes are:
VM.Standard3.Flex (Intel)
VM.Standard.E4.Flex (AMD)
VM.Standard.E5.Flex (AMD)
VM.Standard.A1.Flex (Altra processor from Ampere)
VM.Standard.A2.Flex (AmpereOne processor from Ampere)
VM.DenseIO.E4.Flex (AMD)
VM.Optimized3.Flex (Intel)
Flexible memory is also available on flexible shapes. The amount of memory allowed is based on the number of OCPUs selected.
For standard and optimized flexible shapes, the ratio of memory to OCPUs depends on the shape.
1 GB or a value matching the number of OCPUs, whichever is greater
64 GB per OCPU, up to 1049 GB total
VM.Standard.A1.Flex
76
(OCPU is 1 core of an Altra processor)
1 GB or a value matching the number of OCPUs, whichever is greater
64 GB per OCPU, up to 472 GB total
VM.Standard.A2.Flex
78
(OCPU is 2 cores of an AmpereOne processor)
1 GB or a value matching the number of OCPUs, whichever is greater
64 GB per OCPU, up to 946 GB total
VM.Optimized3.Flex
18
1 GB or a value matching the number of OCPUs, whichever is greater
64 GB per OCPU, up to 256 GB total
For dense I/O flexible shapes, the following configurations are available:
8 OCPUs, 128 GB memory
16 OCPUs, 256 GB memory
32 OCPUs, 512 GB memory
These resources are billed at a per-second granularity with a one-minute minimum.
Optimize your costs by choosing the shape that matches your workload and by changing the
shape when your workload changes. For example, you can configure the VM to maximize
compute processing power by choosing a low core-to-memory ratio. Or, for applications
like in-memory databases or big data processing engines, configure an instance with a
high core-to-memory ratio. Modify the OCPUs and memory as your workload changes, scaling
up to increase performance or scaling down to reduce costs.
Supported Images 🔗
Most platform images are compatible with flexible shapes. Use a platform image that was published after the flexible shape was released (for release dates, see the Compute release notes).
Custom images are also supported, depending on the image. You must add flexible shape compatibility to the custom image, and then test the image on the flexible shape to ensure that it actually works on the shape.
Supported Regions 🔗
For a list of supported regions, see the compute instance service limits. As host capacity becomes available in additional regions, the list is updated.
Network bandwidth is based on expected bandwidth for traffic within a VCN. To determine which physical NICs are active for a shape, refer to the network bandwidth specifications in the following tables. If the network bandwidth is listed as "2 x <bandwidth> Gbps," it means that both NIC 0 and NIC 1 are active.
For bare metal instances, optionally configure advanced BIOS settings, such as disabling simultaneous multithreading, disabling cores, or optimizing the NUMA settings.
Standard Shapes 🔗
Designed for general purpose workloads and suitable for a wide range of applications and use cases. Standard shapes provide a balance of cores, memory, and network resources. Standard shapes are available with Intel, AMD, and Arm-based processors.
These are the bare metal standard series:
BM.Standard3: X9-based standard compute. Processor: Intel Xeon Platinum 8358. Base frequency 2.6 GHz, max turbo frequency 3.4 GHz.
BM.Standard.E4: E4-based standard compute. Processor: AMD EPYC 7J13. Base frequency 2.55 GHz, max boost frequency 3.5 GHz.
BM.Standard.E5: E5-based standard compute. Processor: AMD EPYC 9J14. Base frequency 2.4 GHz, max boost frequency 3.7 GHz.
BM.Standard.A1:
OCI Ampere A1 Compute Arm-based standard compute. Each OCPU corresponds to a single hardware execution thread. Processor: Ampere Altra Q80-30. Max frequency 3.0 GHz.
Shape
OCPU
Memory (GB)
Local Disk
Max Network Bandwidth
Max VNICs Total: Linux
Max VNICs Total: Windows
BM.Standard3.64
64
1024
Block storage only
2 x 50 Gbps
256
129 (1 on the first physical NIC, 128 on the second)
BM.Standard.E4.128
128
2048
Block storage only
2 x 50 Gbps
256
129 (1 on the first physical NIC, 128 on the second)
BM.Standard.E5.192
192
2304
Block storage only
1 x 100 Gbps
256
129 (1 on the first physical NIC, 128 on the second)
Modern CPUs transition to a power-saving state (called c-states) when the CPU is idle or underutilized. These c-states start at C0, which is the normal CPU operating mode (the CPU is 100% activated). The higher the c-state, the deeper the sleep mode into which the CPU transitions. The sleep modes work by cutting the clock signal and power from idle units inside the CPU, thereby reducing the energy use. As the CPU transitions to higher c-states (deeper sleep states), the longer it takes to wake up the units that are shut down. This is an undesirable side effect of c-states transitions, because it can slow down a demanding application.
Fortunately, the hypervisor on standard VM shapes manages this complexity for the end user by preventing transitions to deeper sleep states even when the CPU is underutilized. In addition, it disables c-states when it sees sustained high utilization. When c-states are disabled, the CPU operates in C0 state, where all cores are active at base frequency. Each processor manufacturer names the maximum frequency per core differently; for Intel the maximum frequency is named max turbo frequency, and for AMD it is called max boost frequency. This maximum frequency is realized by the respective CPU's built-in algorithms when the processor is running in C0 state under normal but sustained load.
Currently, the hypervisor does not allow the client operating system running in the instance to manage the c-states using kernel command line options. The client always shows the base frequency, even when the hypervisor is running the processor at the maximum frequency advertised by the processor.
Dense I/O Shapes 🔗
Designed for large databases, big data workloads, and applications that require high-performance local storage. DenseIO shapes include locally-attached NVMe-based SSDs.
This is the bare metal dense I/O series:
BM.DenseIO.E4: E4-based dense I/O compute. Processor: AMD EPYC 7J13. Base frequency 2.55 GHz, max boost frequency 3.5 GHz.
BM.DenseIO.E5: E5-based dense I/O compute. Processor: AMD EPYC 9J14. Base frequency 2.4 GHz, max boost frequency 3.7 GHz.
Shape
OCPU
Memory (GB)
Local Disk
Max Network Bandwidth
Max VNICs Total: Linux
Max VNICs Total: Windows
BM.DenseIO.E4.128
128
2048
54.4 TB NVMe SSD Storage (8 drives)
2 x 50 Gbps
256
129 (1 on the first physical NIC, 128 on the second)
BM.DenseIO.E5.1281
128
1536
81.6TB NVMe SSD Storage (12 x 6.8TB drives)
1 x 100Gbps
256
129 (1 on the first physical NIC, 128 on the second)
Designed for hardware-accelerated workloads. GPU shapes include Intel or AMD CPUs and NVIDIA graphics processors.
Some bare metal GPU shapes support cluster networking.
These are the bare metal GPU series:
BM.GPU2: X7-based GPU compute.
GPU: NVIDIA Tesla P100 16 GB
CPU: Intel Xeon Platinum 8167M. Base frequency 2.0 GHz, max turbo frequency 2.4 GHz.
BM.GPU3: X7-based GPU compute.
GPU: NVIDIA Tesla V100 16 GB
CPU: Intel Xeon Platinum 8167M. Base frequency 2.0 GHz, max turbo frequency 2.4 GHz.
BM.GPU4: E3-based GPU compute.
GPU: NVIDIA A100 40 GB
CPU: AMD EPYC 7542. Base frequency 2.9 GHz, max boost frequency 3.4 GHz.
BM.GPU.A10: X9-based GPU compute.
GPU: NVIDIA A10 24 GB
CPU: Intel Xeon Platinum 8358. Base frequency 2.6 GHz, max turbo frequency 3.4 GHz.
BM.GPU.A100: E4-based GPU compute.
GPU: NVIDIA A100 80 GB
CPU: AMD EPYC 7J13. Base frequency 2.55 GHz, max boost frequency 3.7 GHz.
BM.GPU.H100.8: X10-based GPU compute.
GPU: 8x H100 80 GB
CPU: Intel Sapphire Rapids 8480+ 2x 56c. Base frequency 2 GHz, max boost frequency 3.8 GHz.
BM.GPU.MI300X.8: X10-based GPU compute.
GPU: 8x MI300X 192 GB
CPU: Intel Sapphire Rapids 8480+ 2x 56c. Base frequency 2 GHz, max boost frequency 3.8 GHz.
BM.GPU.L40S.4
GPU: 4x L40S 48 GB
CPU: 2 x 56-core Intel Sapphire Rapids 8480+
BM.GPU.H200.8
GPU: 8x NVIDIA H200 Tensor Core GPUs 141 GB
CPU: 2 x 56-core Intel Sapphire Rapids 8480+
Shape
OCPU
GPU Memory (GB)
CPU Memory (GB)
Local Disk
Max Network Bandwidth
Max VNICs Total: Linux
Max VNICs Total: Windows
BM.GPU2.2
(GPU: 2xP100)
28
32
192
Block storage only
2 x 25 Gbps
28
15 (1 on the first physical NIC, 14 on the second)
BM.GPU3.8
(GPU: 8xV100)
52
128
768
Block storage only
2 x 25 Gbps
52
27 (1 on the first physical NIC, 26 on the second)
BM.GPU4.8
(GPU: 8xA100)
64
320
2048
27.2 TB NVMe SSD (4 drives)
1 x 50 Gbps
8 x 200 Gbps RDMA
64
Windows images are not supported on this shape.
BM.GPU.A10.4
(GPU: 4xA10)
64
96
1024
7.68 TB NVMe SSD (2 drives)
2 x 50 Gbps
256
Windows images are not supported on this shape.
BM.GPU.A100-v2.8
(GPU: 8xA100)
128
640
2048
27.2 TB NVMe SSD (4 drives)
2 x 50 Gbps
16 x 100 Gbps RDMA
256
Windows images are not supported on this shape.
BM.GPU.H100.8
(GPU: 8xH100)
112
640
2048
16 x 3.84 TB NVMe
1 x 100 Gbps
8 x 2 x 200 Gbps RDMA
256
Windows images are not supported on this shape.
BM.GPU.MI300X.8
112
1536
2048
8 x 3.84 GB NVMe
1 x 100 Gbps
8 x 1 x 400 Gbps RDMA
256
Windows images are not supported on this shape.
BM.GPU.L40S.4
112
192
1024
2 x 3.84 TB NVMe
1 x 200 Gbps
800 Gbps RDMA
256
Windows images are not supported on this shape.
BM.GPU.H200.8
(GPU: 8xH200)
112
1128
3072
8 x 3.84 TB NVMe
1 x 200 Gbps
8 x 400 Gbps RDMA
256
Windows images are not supported on this shape.
HPC and Optimized Shapes 🔗
Designed for high-performance computing workloads that require high frequency processor cores. Bare metal HPC and optimized shapes support cluster networking.
This is the bare metal optimized series:
BM.Optimized3: Processor: Intel Xeon 6354. Base frequency 3.0 GHz, max turbo frequency 3.6 GHz.
BM.HPC.E5: Processor: AMD EPYC 9J14. Base frequency 2.4 GHz, max boost frequency 3.7 GHz.
Shape
OCPU
Memory (GB)
Local Disk
Max Network Bandwidth
Max VNICs Total: Linux
Max VNICs Total: Windows
BM.Optimized3.36
36
512
3.84 TB NVMe SSD (1 drive)
2 X 50 Gbps
1 X 100 Gbps RDMA
256
129
BM.HPC.E5.1441
144
768
3.84 TB NVMe SSD (1 drive)
1 x 100Gbps
1 x 100Gbps RDMA
256
129 (1 on the first physical NIC, 128 on the second)
Network bandwidth is based on expected bandwidth for traffic within a VCN.
Standard Shapes 🔗
Designed for general purpose workloads and suitable for a wide range of applications and use cases. Standard shapes provide a balance of cores, memory, and network resources. Standard shapes are available with Intel, AMD, and Arm-based processors.
These are the VM standard series:
VM.Standard3: X9-based standard compute. Processor: Intel Xeon Platinum 8358. Base frequency 2.6 GHz, max turbo frequency 3.4 GHz.
VM.Standard.E2.1.Micro: E2-based, E3-based, or E4-based standard compute. Oracle Cloud Infrastructure assigns one of the following processors:
AMD EPYC 7551. Base frequency 2.0 GHz, max boost frequency 3.0 GHz.
AMD EPYC 7742. Base frequency 2.25 GHz, max boost frequency 3.4 GHz.
AMD EPYC 7J13. Base frequency 2.55 GHz, max boost frequency 3.5 GHz.
VM.Standard.E4: E4-based standard compute. Processor: AMD EPYC 7J13. Base frequency 2.55 GHz, max boost frequency 3.5 GHz.
VM.Standard.E5: E5-based standard compute. Processor: AMD EPYC 9J14. Base frequency 2.4 GHz, max boost frequency 3.7 GHz.
VM.Standard.A1:
OCI Ampere A1 Compute Arm-based standard compute. Each OCPU corresponds to a single hardware execution thread. Processor: Ampere Altra Q80-30. Max frequency 3.0 GHz.
VM.Standard.A2:
OCI Ampere A2 Compute Arm-based standard compute. Each OCPU corresponds to two hardware execution threads (2 cores). Processor: Ampere AmpereOne A160-30. Max frequency 3.0 GHz.
Designed for large databases, big data workloads, and applications that require high-performance local storage. DenseIO shapes include locally-attached NVMe-based SSDs.
This is the VM dense I/O series:
VM.DenseIO.E4: E4-based dense I/O compute. Processor: AMD EPYC 7J13. Base frequency 2.55 GHz, max boost frequency 3.5 GHz.
VM.DenseIO.E5: E5-based dense I/O compute. Processor: AMD EPYC 9J14. Base frequency 2.4 GHz, max boost frequency 3.7 GHz.
Designed for hardware-accelerated workloads. GPU shapes include Intel or AMD CPUs and NVIDIA graphics processors.
These are the VM GPU series:
VM.GPU2: X7-based GPU compute.
GPU: NVIDIA Tesla P100 16 GB
CPU: Intel Xeon Platinum 8167M. Base frequency 2.0 GHz, max turbo frequency 2.4 GHz.
VM.GPU3: X7-based GPU compute.
GPU: NVIDIA Tesla V100 16 GB
CPU: Intel Xeon Platinum 8167M. Base frequency 2.0 GHz, max turbo frequency 2.4 GHz.
VM.GPU.A10: X9-based GPU compute.
GPU: NVIDIA A10 24 GB
CPU: Intel Xeon Platinum 8358. Base frequency 2.6 GHz, max turbo frequency 3.4 GHz.
Shape
OCPU
GPU Memory (GB)
CPU Memory (GB)
Local Disk (TB)
Max Network Bandwidth
Max VNICs Total: Linux
Max VNICs Total: Windows
VM.GPU2.1
(GPU: 1xP100)
12
16
72
Block storage only
8 Gbps
12
12
VM.GPU3.1
(GPU: 1xV100)
6
16
90
Block storage only
4 Gbps
6
6
VM.GPU3.2
(GPU: 2xV100)
12
32
180
Block storage only
8 Gbps
12
12
VM.GPU3.4
(GPU: 4xV100)
24
64
360
Block storage only
24.6 Gbps
24
24
VM.GPU.A10.1
(GPU: 1xA10)
15
24
240
Block storage only
24 Gbps
15
15
VM.GPU.A10.2
(GPU: 2xA10)
30
48
480
Block storage only
48 Gbps
24
24
HPC and Optimized Shapes 🔗
Designed for high-performance computing workloads that require high frequency processor cores.
This is the VM optimized series:
VM.Optimized3: Processor: Intel Xeon 6354. Base frequency 3.0 GHz, max turbo frequency 3.6 GHz.
Shape
OCPU
Memory (GB)
Local Disk
Max Network Bandwidth
Max VNICs Total: Linux
Max VNICs Total: Windows
VM.Optimized3.Flex
1 OCPU minimum, 18 OCPU maximum
1 GB minimum, 256 GB maximum
Block storage only
4 Gbps per OCPU, maximum 40 Gbps
2 VNICs per OCPU.
Maximum 24 VNICs.
2 VNICs per OCPU.
Maximum 24 VNICs.
Dedicated Virtual Machine Host Shapes 🔗
Shape
Instance Type
Billed OCPU
Usable OCPU1
Total Memory (GB)3
Usable Memory (GB)1
Supported Shapes for Hosted VMs
DVH.Standard2.522
X7-based VM host
52
48
768
736
VM.Standard2 series
DVH.Standard3.64
X9-based VM host
64
60
1024
960
VM.Standard3 series
DVH.Standard.E2.642
E2-based VM host
64
59
512
480
VM.Standard.E2 series
DVH.Standard.E3.1282
E3-based VM host
128
124
2048
1912
VM.Standard.E3 series
DVH.Standard.E4.128
E4-based VM host
128
124
2048
1912
VM.Standard.E4 series
DVH.Standard.E5.192
E5-based VM host
192
188
2304
2098
VM.Standard.E5 series
DVH.DenseIO2.522
X7-based dense I/O VM host
52
48
768
736
VM.DenseIO2 series
DVH.Optimized3.36
X9-based optimized VM host
36
32
512
472
VM.Optimized3 series
Notes
1: The difference between total and usable OCPUs and memory is caused by the need to reserve OCPUs and memory for hypervisor use.
2: Because this dedicated virtual machine host shape supports hosted VMs that use a previous generation shape series, it is available by request only.
3: For Standard2, Standard.E2, and DenseIO2 shapes, billing is based on OCPUs, not memory. For all other shapes that support flexible hosted VMs, billing is based on OCPUs and memory, which are billed independently.
Previous Generation Shapes 🔗
Tip
Previous generation shapes are still fully supported. However, because the underlying hardware has reached the sustaining phase of its lifecycle, capacity in certain high-demand regions might be limited.
Oracle Cloud Infrastructure periodically releases new generations of Compute shapes. The latest shapes let you take advantage of newer hardware and a better price-performance ratio. When a shape is several years old, and newer generation shapes that are suited for the same purposes are available, the old shape transitions to become a previous generation shape.
Your current utilization is fully supported on the previous generation shape. In certain high demand regions, you may need to plan your utilization growth on a newer generation shape.
Upgrading from a Previous Generation Shape 🔗
To upgrade from a previous generation shape to a current generation shape, you can do the following things: