GPU machine types
This document outlines the NVIDIA GPU models available on Compute Engine,which you can use to accelerate machine learning (ML), data processing, andgraphics-intensive workloads on your virtual machine (VM) instances. Thisdocument also details which GPUs come pre-attached to accelerator-optimizedmachine series such as A4X, A4, A3, A2, G4, and G2, and which GPUs you can attachto N1 general-purpose instances.
Use this document to compare the performance, memory, and features of differentGPU models. For a more detailed overview of the accelerator-optimized machinefamily, including information on CPU platforms, storage options, and networkingcapabilities, and to find the specific machine type that matches your workload,seeAccelerator-optimized machine family.
For more information about GPUs on Compute Engine, seeAbout GPUs.
To view available regions and zones for GPUs on Compute Engine, seeGPUs regions and zone availability.
GPU machine types
Compute Engine offers different machine types to support your variousworkloads.
Some machine types supportNVIDIA RTX Virtual Workstations (vWS).When you create an instance that uses NVIDIA RTX Virtual Workstation,Compute Engine automatically adds a vWS license. For information about pricingfor virtual workstations, see theGPU pricing page.
| GPU machine types | |||
|---|---|---|---|
| AI and ML workloads | Graphics and visualization | Other GPU workloads | |
| Accelerator-optimized A series machine types are designed for high performance computing (HPC), artificial intelligence (AI), and machine learning (ML) workloads. The later generation A series are ideal for pre-training and fine-tuning foundation models that involves large clusters of accelerators, while the A2 series can be used for training smaller models and single host inference. For these machine types, the GPU model is automatically attached to the instance. | Accelerator-optimized G series machine types are designed for workloads such as NVIDIA Omniverse simulation workloads, graphics-intensive applications, video transcoding, and virtual desktops. These machine types supportNVIDIA RTX Virtual Workstations (vWS). The G series can also be used for training smaller models and for single-host inference. For these machine types, the GPU model is automatically attached to the instance. | For N1 general-purpose machine types, except for the N1 shared-core ( | |
| The following GPU models can be attached to N1 general-purpose machine types:
| ||
You can also use some GPU machine types onAI Hypercomputer. AI Hypercomputer is asupercomputing system that is optimized to support your artificial intelligence(AI) and machine learning (ML) workloads. This option is recommended for creating adensely allocated, performance-optimized infrastructure that has integrationsfor Google Kubernetes Engine (GKE) and Slurm schedulers.
A4X machine series
A4X accelerator-optimized machine types use NVIDIA GB200 Grace Blackwell Superchips (nvidia-gb200) and are ideal for foundation model training and serving.
A4X is an exascale platform based onNVIDIA GB200 NVL72. Each machine has two sockets with NVIDIA Grace CPUs with Arm Neoverse V2 cores. These CPUs are connected to four NVIDIA B200 Blackwell GPUs with fast chip-to-chip (NVLink-C2C) communication.
Tip: When provisioning A4X instances, you mustreserve capacity to create instances and cluster. You can then create instances that use the features and services available from AI Hypercomputer. For more information, seeDeployment options overview in the AI Hypercomputer documentation.| Attached NVIDIA GB200 Grace Blackwell Superchips | |||||||
|---|---|---|---|---|---|---|---|
| Machine type | vCPU count1 | Instance memory (GB) | Attached Local SSD (GiB) | Physical NIC count | Maximum network bandwidth (Gbps)2 | GPU count | GPU memory3 (GB HBM3e) |
a4x-highgpu-4g | 140 | 884 | 12,000 | 6 | 2,000 | 4 | 744 |
1A vCPU is implemented as a single hardware hyper-thread on one ofthe availableCPU platforms.
2Maximum egress bandwidth cannot exceed the number given. Actualegress bandwidth depends on the destination IP address and other factors.For more information about network bandwidth,seeNetwork bandwidth.
3GPU memory is the memory on a GPU device that can be used fortemporary storage of data. It is separate from the instance's memory and isspecifically designed to handle the higher bandwidth demands of yourgraphics-intensive workloads.
A4 machine series
A4 accelerator-optimizedmachine types haveNVIDIA B200 Blackwell GPUs(nvidia-b200) attached and are ideal for foundation modeltraining and serving.
| Attached NVIDIA B200 Blackwell GPUs | |||||||
|---|---|---|---|---|---|---|---|
| Machine type | vCPU count1 | Instance memory (GB) | Attached Local SSD (GiB) | Physical NIC count | Maximum network bandwidth (Gbps)2 | GPU count | GPU memory3 (GB HBM3e) |
a4-highgpu-8g | 224 | 3,968 | 12,000 | 10 | 3,600 | 8 | 1,440 |
1A vCPU is implemented as a single hardware hyper-thread on one ofthe availableCPU platforms.
2Maximum egress bandwidth cannot exceed the number given. Actualegress bandwidth depends on the destination IP address and other factors.For more information about network bandwidth, seeNetwork bandwidth.
3GPU memory is the memory on a GPU device that can be used fortemporary storage of data. It is separate from the instance's memory and isspecifically designed to handle the higher bandwidth demands of yourgraphics-intensive workloads.
A3 machine series
A3 accelerator-optimizedmachine types have NVIDIA H100 SXM or NVIDIA H200 SXM GPUs attached.
A3 Ultra machine type
A3 Ultramachine types haveNVIDIA H200 SXM GPUs(nvidia-h200-141gb) attached and provides the highest networkperformance in the A3 series. A3 Ultra machine types are ideal for foundation model training andserving.
| Attached NVIDIA H200 GPUs | |||||||
|---|---|---|---|---|---|---|---|
| Machine type | vCPU count1 | Instance memory (GB) | Attached Local SSD (GiB) | Physical NIC count | Maximum network bandwidth (Gbps)2 | GPU count | GPU memory3 (GB HBM3e) |
a3-ultragpu-8g | 224 | 2,952 | 12,000 | 10 | 3,600 | 8 | 1128 |
1A vCPU is implemented as a single hardware hyper-thread on one ofthe availableCPU platforms.
2Maximum egress bandwidth cannot exceed the number given. Actualegress bandwidth depends on the destination IP address and other factors.For more information about network bandwidth,seeNetwork bandwidth.
3GPU memory is the memory on a GPU device that can be used fortemporary storage of data. It is separate from the instance's memory and isspecifically designed to handle the higher bandwidth demands of yourgraphics-intensive workloads.
A3 Mega, High, and Edge machine types
To useNVIDIA H100 SXM GPUs,you have the following options:
- A3 Mega: thesemachine types have H100 SXM GPUs (
nvidia-h100-mega-80gb)and are ideal for large-scale training and serving workloads. - A3 High: thesemachine types have H100 SXM GPUs (
nvidia-h100-80gb) and arewell-suited for both training and serving tasks. - A3 Edge: thesemachine types have H100 SXM GPUs (
nvidia-h100-80gb),are designed specifically for serving, and are available in alimited set of regions.
A3 Mega
a3-megagpu-8g machine types, we recommend using a cluster of these instances and deployingwith a scheduler such as Google Kubernetes Engine (GKE) or Slurm. For detailed instructions on either ofthese options, review the following:- To create Google Kubernetes Engine cluster, seeDeploy an A3 Mega cluster with GKE.
- To create a Slurm cluster, seeDeploy an A3 Mega Slurm cluster.
| Attached NVIDIA H100 GPUs | |||||||
|---|---|---|---|---|---|---|---|
| Machine type | vCPU count1 | Instance memory (GB) | Attached Local SSD (GiB) | Physical NIC count | Maximum network bandwidth (Gbps)2 | GPU count | GPU memory3 (GB HBM3) |
a3-megagpu-8g | 208 | 1,872 | 6,000 | 9 | 1,800 | 8 | 640 |
A3 High
a3-highgpu-1g,a3-highgpu-2g, ora3-highgpu-4g machine types,you must create instances by using Spot VMs orFlex-start VMs. For detailed instructions on these options, review the following:- To create Spot VMs, set the provisioning model to
SPOTwhen youcreate an accelerator-optimized VM. - To create Flex-start VMs, you can use one of the following methods:
- Create a standalone VM and set the provisioning model to
FLEX_STARTwhen youcreate an accelerator-optimized VM. - Create a resize request in a managed instance group (MIG). For instructions, seeCreate a MIG with GPU VMs.
- Create a standalone VM and set the provisioning model to
| Attached NVIDIA H100 GPUs | |||||||
|---|---|---|---|---|---|---|---|
| Machine type | vCPU count1 | Instance memory (GB) | Attached Local SSD (GiB) | Physical NIC count | Maximum network bandwidth (Gbps)2 | GPU count | GPU memory3 (GB HBM3) |
a3-highgpu-1g | 26 | 234 | 750 | 1 | 25 | 1 | 80 |
a3-highgpu-2g | 52 | 468 | 1,500 | 1 | 50 | 2 | 160 |
a3-highgpu-4g | 104 | 936 | 3,000 | 1 | 100 | 4 | 320 |
a3-highgpu-8g | 208 | 1,872 | 6,000 | 5 | 1,000 | 8 | 640 |
A3 Edge
| Attached NVIDIA H100 GPUs | |||||||
|---|---|---|---|---|---|---|---|
| Machine type | vCPU count1 | Instance memory (GB) | Attached Local SSD (GiB) | Physical NIC count | Maximum network bandwidth (Gbps)2 | GPU count | GPU memory3 (GB HBM3) |
a3-edgegpu-8g | 208 | 1,872 | 6,000 | 5 |
| 8 | 640 |
1A vCPU is implemented as a single hardware hyper-thread on one ofthe availableCPU platforms.
2Maximum egress bandwidth cannot exceed the number given. Actualegress bandwidth depends on the destination IP address and other factors.For more information about network bandwidth,seeNetwork bandwidth.
3GPU memory is the memory on a GPU device that can be used fortemporary storage of data. It is separate from the instance's memory and isspecifically designed to handle the higher bandwidth demands of yourgraphics-intensive workloads.
A2 machine series
A2 accelerator-optimizedmachine types haveNVIDIA A100 GPUsattached and are ideal for model fine tuning, large modeland cost optimized inference.
A2 machine series are available in two types:
- A2 Ultra: these machine types have A100 80GB GPUs(
nvidia-a100-80gb) and Local SSD disks attached. - A2 Standard: these machine types have A100 40GB GPUs(
nvidia-tesla-a100) attached. You can also add LocalSSD disks when creating an A2 Standard instance. For the number of disksyou can attach, seeMachine types that require you to choose a number of Local SSD disks.
A2 Ultra
| Attached NVIDIA A100 80GB GPUs | ||||||
|---|---|---|---|---|---|---|
| Machine type | vCPU count1 | Instance memory (GB) | Attached Local SSD (GiB) | Maximum network bandwidth (Gbps)2 | GPU count | GPU memory3 (GB HBM2e) |
a2-ultragpu-1g | 12 | 170 | 375 | 24 | 1 | 80 |
a2-ultragpu-2g | 24 | 340 | 750 | 32 | 2 | 160 |
a2-ultragpu-4g | 48 | 680 | 1,500 | 50 | 4 | 320 |
a2-ultragpu-8g | 96 | 1,360 | 3,000 | 100 | 8 | 640 |
A2 Standard
| Attached NVIDIA A100 40GB GPUs | ||||||
|---|---|---|---|---|---|---|
| Machine type | vCPU count1 | Instance memory (GB) | Local SSD supported | Maximum network bandwidth (Gbps)2 | GPU count | GPU memory3 (GB HBM2) |
a2-highgpu-1g | 12 | 85 | Yes | 24 | 1 | 40 |
a2-highgpu-2g | 24 | 170 | Yes | 32 | 2 | 80 |
a2-highgpu-4g | 48 | 340 | Yes | 50 | 4 | 160 |
a2-highgpu-8g | 96 | 680 | Yes | 100 | 8 | 320 |
a2-megagpu-16g | 96 | 1,360 | Yes | 100 | 16 | 640 |
1A vCPU is implemented as a single hardware hyper-thread on one ofthe availableCPU platforms.
2Maximum egress bandwidth cannot exceed the number given. Actualegress bandwidth depends on the destination IP address and other factors.For more information about network bandwidth,seeNetwork bandwidth.
3GPU memory is the memory on a GPU device that can be used fortemporary storage of data. It is separate from the instance's memory and isspecifically designed to handle the higher bandwidth demands of yourgraphics-intensive workloads.
G4 machine series
G4 accelerator-optimized machine types use NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs (nvidia-rtx-pro-6000) and are suitable for NVIDIA Omniverse simulation workloads, graphics-intensive applications, video transcoding, and virtual desktops. G4 machine types also provide a low-cost solution for performing single host inference and model tuning compared with A series machine types.
A key feature of the G4 series is support for direct GPU peer-to-peer (P2P) communication on multi-GPU machine types (g4-standard-96,g4-standard-192,g4-standard-384). This allows GPUs within the same instance to exchange data directly over the PCIe bus, without involving the CPU host. For more information about G4 GPU peer-to-peer communication, seeG4 GPU peer-to-peer communication.
| Attached NVIDIA RTX PRO 6000 GPUs | |||||||
|---|---|---|---|---|---|---|---|
| Machine type | vCPU count1 | Instance memory (GB) | Maximum Titanium SSD supported (GiB)2 | Physical NIC count | Maximum network bandwidth (Gbps)3 | GPU count | GPU memory4 (GB GDDR7) |
g4-standard-48 | 48 | 180 | 1,500 | 1 | 50 | 1 | 96 |
g4-standard-96 | 96 | 360 | 3,000 | 1 | 100 | 2 | 192 |
g4-standard-192 | 192 | 720 | 6,000 | 1 | 200 | 4 | 384 |
g4-standard-384 | 384 | 1,440 | 12,000 | 2 | 400 | 8 | 768 |
1A vCPU is implemented as a single hardware hyper-thread on one ofthe availableCPU platforms.
2You can add Titanium SSD disks when creating a G4 instance. For the number of disksyou can attach, seeMachine types that require you to choose a number of Local SSD disks.
3Maximum egress bandwidth cannot exceed the number given. Actualegress bandwidth depends on the destination IP address and other factors.SeeNetwork bandwidth.
4GPU memory is the memory on a GPU device that can be used fortemporary storage of data. It is separate from the instance's memory and isspecifically designed to handle the higher bandwidth demands of yourgraphics-intensive workloads.
G2 machine series
G2 accelerator-optimizedmachine types haveNVIDIA L4 GPUsattached and are ideal for cost-optimized inference, graphics-intensive andhigh performance computing workloads.
Each G2 machine type also has a default memory and a custommemory range. The custom memory range defines the amount of memory thatyou can allocate to your instance for each machine type. You can also add LocalSSD disks when creating a G2 instance. For the number of disksyou can attach, seeMachine types that require you to choose a number of Local SSD disks.
| Attached NVIDIA L4 GPUs | |||||||
|---|---|---|---|---|---|---|---|
| Machine type | vCPU count1 | Default instance memory (GB) | Custom instance memory range (GB) | Max Local SSD supported (GiB) | Maximum network bandwidth (Gbps)2 | GPU count | GPU memory3 (GB GDDR6) |
g2-standard-4 | 4 | 16 | 16 to 32 | 375 | 10 | 1 | 24 |
g2-standard-8 | 8 | 32 | 32 to 54 | 375 | 16 | 1 | 24 |
g2-standard-12 | 12 | 48 | 48 to 54 | 375 | 16 | 1 | 24 |
g2-standard-16 | 16 | 64 | 54 to 64 | 375 | 32 | 1 | 24 |
g2-standard-24 | 24 | 96 | 96 to 108 | 750 | 32 | 2 | 48 |
g2-standard-32 | 32 | 128 | 96 to 128 | 375 | 32 | 1 | 24 |
g2-standard-48 | 48 | 192 | 192 to 216 | 1,500 | 50 | 4 | 96 |
g2-standard-96 | 96 | 384 | 384 to 432 | 3,000 | 100 | 8 | 192 |
1A vCPU is implemented as a single hardware hyper-thread on one ofthe availableCPU platforms.
2Maximum egress bandwidth cannot exceed the number given. Actualegress bandwidth depends on the destination IP address and other factors.For more information about network bandwidth,seeNetwork bandwidth.
3GPU memory is the memory on a GPU device that can be used fortemporary storage of data. It is separate from the instance's memory and isspecifically designed to handle the higher bandwidth demands of yourgraphics-intensive workloads.
N1 machine series
You can attach the following GPU models to anN1 machine typewith the exception of theN1 shared-core machine types.
Unlike the machine types in the accelerator-optimized machine series, N1 machinetypes don't come with a set number of attached GPUs. Instead, you specify thenumber of GPUs to attach when creating the instance.
N1 instances with fewer GPUs limit the maximum number of vCPUs. In general, ahigher number of GPUs lets you create instances with a higher number of vCPUsand memory.
N1+T4 GPUs
You can attachNVIDIA T4GPUs to N1 general-purpose instances with the following instance configurations.
| Accelerator type | GPU count | GPU memory1 (GB GDDR6) | vCPU count | Instance memory (GB) | Local SSD supported |
|---|---|---|---|---|---|
nvidia-tesla-t4 ornvidia-tesla-t4-vws | 1 | 16 | 1 to 48 | 1 to 312 | Yes |
| 2 | 32 | 1 to 48 | 1 to 312 | Yes | |
| 4 | 64 | 1 to 96 | 1 to 624 | Yes |
1GPU memory is the memory available on a GPU device that you can usefor temporary data storage. It is separate from the instance's memory and isspecifically designed to handle the higher bandwidth demands of yourgraphics-intensive workloads.
N1+P4 GPUs
You can attachNVIDIA P4GPUs to N1 general-purpose instances with the following instance configurations.
| Accelerator type | GPU count | GPU memory1 (GB GDDR5) | vCPU count | Instance memory (GB) | Local SSD supported2 |
|---|---|---|---|---|---|
nvidia-tesla-p4 ornvidia-tesla-p4-vws | 1 | 8 | 1 to 24 | 1 to 156 | Yes |
| 2 | 16 | 1 to 48 | 1 to 312 | Yes | |
| 4 | 32 | 1 to 96 | 1 to 624 | Yes |
1GPU memory is the memory that is available on a GPU devicethat you can use for temporary data storage. It is separate from the instance'smemory and is specifically designed to handle the higher bandwidth demands ofyour graphics-intensive workloads.
2For instances with attached NVIDIA P4 GPUs, Local SSD disksare only supported in zonesus-central1-c andnorthamerica-northeast1-b.
N1+V100 GPUs
You can attachNVIDIA V100GPUs to N1 general-purpose instances with the following instance configurations.
| Accelerator type | GPU count | GPU memory1 (GB HBM2) | vCPU count | Instance memory (GB) | Local SSD supported2 |
|---|---|---|---|---|---|
nvidia-tesla-v100 | 1 | 16 | 1 to 12 | 1 to 78 | Yes |
| 2 | 32 | 1 to 24 | 1 to 156 | Yes | |
| 4 | 64 | 1 to 48 | 1 to 312 | Yes | |
| 8 | 128 | 1 to 96 | 1 to 624 | Yes |
1GPU memory is the memory available on a GPU device that you can usefor temporary data storage. It is separate from the instance's memory and isspecifically designed to handle the higher bandwidth demands of yourgraphics-intensive workloads.
2For instances with attached NVIDIA V100 GPUs, Local SSD disksaren't supported inus-east1-c.
N1+P100 GPUs
You can attachNVIDIA P100 GPUsto N1 general-purpose instances with the following instance configurations.
For some NVIDIA P100 GPUs, the maximum CPU and memory available for someconfigurations depends on the zone in which the GPU resource runs.
| Accelerator type | GPU count | GPU memory1 (GB HBM2) | Zone | vCPU count | Instance memory (GB) | Local SSD supported |
|---|---|---|---|---|---|---|
nvidia-tesla-p100 ornvidia-tesla-p100-vws | 1 | 16 | All P100 zones | 1 to 16 | 1 to 104 | Yes |
| 2 | 32 | All P100 zones | 1 to 32 | 1 to 208 | Yes | |
| 4 | 64 | us-east1-c,europe-west1-d,europe-west1-b | 1 to 64 | 1 to 208 | Yes | |
| All other P100 zones | 1 to 96 | 1 to 624 | Yes |
1GPU memory is the memory available on a GPU device that you can usefor temporary data storage. It is separate from the instance's memory and isspecifically designed to handle the higher bandwidth demands of yourgraphics-intensive workloads.
General comparison chart
The following table describes the GPU memory size, feature availability, andideal workload types of different GPU models that are available onCompute Engine.
| GPU model | GPU memory | Interconnect | NVIDIA RTX Virtual Workstation (vWS) support | Best used for |
|---|---|---|---|---|
| GB200 | 186 GB HBM3e @ 8 TBps | NVLink Full Mesh @ 1,800 GBps | Large-scale distributed training and inference of LLMs, Recommenders, HPC | |
| B200 | 180 GB HBM3e @ 8 TBps | NVLink Full Mesh @ 1,800 GBps | Large-scale distributed training and inference of LLMs, Recommenders, HPC | |
| H200 | 141 GB HBM3e @ 4.8 TBps | NVLink Full Mesh @ 900 GBps | Large models with massive data tables for ML Training, Inference, HPC, BERT, DLRM | |
| H100 | 80 GB HBM3 @ 3.35 TBps | NVLink Full Mesh @ 900 GBps | Large models with massive data tables for ML Training, Inference, HPC, BERT, DLRM | |
| A100 80GB | 80 GB HBM2e @ 1.9 TBps | NVLink Full Mesh @ 600 GBps | Large models with massive data tables for ML Training, Inference, HPC, BERT, DLRM | |
| A100 40GB | 40 GB HBM2 @ 1.6 TBps | NVLink Full Mesh @ 600 GBps | ML Training, Inference, HPC | |
| RTX PRO 6000 | 96 GB GDDR7 with ECC @ 1597 GBps | N/A | ML Inference, Training, Remote Visualization Workstations,Video Transcoding, HPC | |
| L4 | 24 GB GDDR6 @ 300 GBps | N/A | ML Inference, Training, Remote Visualization Workstations,Video Transcoding, HPC | |
| T4 | 16 GB GDDR6 @ 320 GBps | N/A | ML Inference, Training, Remote Visualization Workstations, Video Transcoding | |
| V100 | 16 GB HBM2 @ 900 GBps | NVLink Ring @ 300 GBps | ML Training, Inference, HPC | |
| P4 | 8 GB GDDR5 @ 192 GBps | N/A | Remote Visualization Workstations, ML Inference, and Video Transcoding | |
| P100 | 16 GB HBM2 @ 732 GBps | N/A | ML Training, Inference, HPC, Remote Visualization Workstations |
To compare GPU pricing for the different GPU models and regions that areavailable on Compute Engine, seeGPU pricing.
Performance comparison chart
The following table describes the performance specifications of different GPUmodels that are available on Compute Engine.
Compute performance
| GPU model | FP64 | FP32 | FP16 | INT8 |
|---|---|---|---|---|
| GB200 | 90 TFLOPS | 180 TFLOPS | ||
| B200 | 40 TFLOPS | 80 TFLOPS | ||
| H200 | 34 TFLOPS | 67 TFLOPS | ||
| H100 | 34 TFLOPS | 67 TFLOPS | ||
| A100 80GB | 9.7 TFLOPS | 19.5 TFLOPS | ||
| A100 40GB | 9.7 TFLOPS | 19.5 TFLOPS | ||
| L4 | 0.5 TFLOPS1 | 30.3 TFLOPS | ||
| T4 | 0.25 TFLOPS1 | 8.1 TFLOPS | ||
| V100 | 7.8 TFLOPS | 15.7 TFLOPS | ||
| P4 | 0.2 TFLOPS1 | 5.5 TFLOPS | 22 TOPS2 | |
| P100 | 4.7 TFLOPS | 9.3 TFLOPS | 18.7 TFLOPS |
1To allow FP64 code to work correctly, the T4, L4, and P4 GPUarchitecture includes a small number of FP64 hardware units.
2TeraOperations per Second.
Tensor core performance
| GPU model | FP64 | TF32 | Mixed-precision FP16/FP32 | INT8 | INT4 | FP8 |
|---|---|---|---|---|---|---|
| GB200 | 90 TFLOPS | 2,500 TFLOPS2 | 5,000 TFLOPS1, 2 | 10,000 TFLOPS2 | 20,000 TFLOPS2 | 10,000 TFLOPS2 |
| B200 | 40 TFLOPS | 1,100 TFLOPS2 | 4,500 TFLOPS1, 2 | 9,000 TFLOPS2 | 9,000 TFLOPS2 | |
| H200 | 67 TFLOPS | 989 TFLOPS2 | 1,979 TFLOPS1, 2 | 3,958 TOPS2 | 3,958 TFLOPS2 | |
| H100 | 67 TFLOPS | 989 TFLOPS2 | 1,979 TFLOPS1, 2 | 3,958 TOPS2 | 3,958 TFLOPS2 | |
| A100 80GB | 19.5 TFLOPS | 156 TFLOPS | 312 TFLOPS1 | 624 TOPS | 1248 TOPS | |
| A100 40GB | 19.5 TFLOPS | 156 TFLOPS | 312 TFLOPS1 | 624 TOPS | 1248 TOPS | |
| L4 | 120 TFLOPS2 | 242 TFLOPS1, 2 | 485 TOPS2 | 485 TFLOPS2 | T4 | 65 TFLOPS | 130 TOPS | 260 TOPS |
| V100 | 125 TFLOPS | |||||
| P4 | ||||||
| P100 |
1For mixed precision training, NVIDIA GB200, B200, H200, H100,A100, and L4 GPUs also support thebfloat16 data type.
2NVIDIA GB200, B200, H200, H100, and L4 GPUssupport structural sparsity. You can use structural sparsity to double the performanceof your models. The values that are documented apply when using structured sparsity.If you aren't using structured sparsity, the values are halved.
What's next?
- Learn more aboutCompute Engine GPUs.
- CheckGPU regions and zones availability.
- ReviewNetwork bandwidths and GPUs.
- ViewGPU pricing details.
Except as otherwise noted, the content of this page is licensed under theCreative Commons Attribution 4.0 License, and code samples are licensed under theApache 2.0 License. For details, see theGoogle Developers Site Policies. Java is a registered trademark of Oracle and/or its affiliates.
Last updated 2025-12-15 UTC.