Node Types
A wide range of GPU options are available, allowing you to select the most optimal compute resource for your workload. If GPU resources are not requested, the workload will be executed on a CPU only node.
Vendor | Generation | Model | VRAM | Label |
NVIDIA | Volta | Tesla V100 | 16 GB | Tesla_V100 |
NVIDIA | Turing | RTX 2080 Ti | 11 GB | GeForce_RTX_2080_Ti |
NVIDIA | Turing | RTX 2060 Super | 8 GB | GeForce_RTX_2060_Super |
NVIDIA | Pascal | GTX 1080 Ti | 11 GB | GeForce_GTX_1080_Ti |
NVIDIA | Pascal | GTX 1070 Ti | 8 GB | GeForce_GTX_1070_Ti |
NVIDIA | Pascal | GTX 1070 | 8 GB | GeForce_GTX_1070 |
NVIDIA | Pascal | P104-100 | 8 GB | P104-100 |
NVIDIA | Pascal | GTX 1060 | 6 GB | GeForce_GTX_1060_6GB |
NVIDIA | Pascal | P106-100 | 6 GB | P106-100 |
Each GPU includes a certain amount of host CPU and RAM, these are included at no additional fee.
GPU Model | vCPU | RAM | Great For |
Tesla V100 NVLINK | 4 Xeon Gold | 32 GB | Deep learning, neural network training, HPC |
Tesla V100 | 3 | 16 GB | AI inference, rendering, batch processing, hashcat |
RTX 2080 Ti | 3 | 16 GB | Machine learning, neural network training, HPC |
RTX 2060 Super | 3 | 16 GB | Machine learning, rendering, batch processing |
GTX 1080 Ti | 1 | 11 GB | Machine learning, rendering, batch processing |
GTX 1070 Ti | 1 | 8 GB | Video transcoding, rendering, batch processing |
GTX 1070 | 1 | 8 GB | Video transcoding, rendering, batch processing |
P104-100 | 0.5 | 8 GB | Batch processing, blockchain compute, hashcat |
GTX 1060 | 0.5 | 6 GB | Video transcoding, batch processing |
P106-100 | 0.5 | 6 GB | Batch processing, blockchain compute |
A workload requesting more resources than allowed for the specific GPU class will have its resources capped to the maximum allowable amount. For example, launching a Pod with a request for 2 1070Tis will have its resource request capped to 2 CPU and 16GB RAM.
CPU Model | RAM per vCPU | Label |
Intel Xeon v1 | 3 GB | xeon |
AMD Epyc Rome | 4 GB | epyc |
Workloads without GPU requests are always scheduled on CPU nodes. If a specific CPU model is not explicitly selected, the scheduler will automatically schedule workloads requesting few CPU cores on Epyc class CPUs, as these perform exceptionally well on single thread workloads.
A combination of resource requests and node affinity is used to select the type and amount of compute for your workload. CoreWeave Cloud relies only on these native Kubernetes methods for resource allocation, allowing maximum flexibilty.
spec:containers:- name: exampleresources:limits:cpu: 3memory: 16Ginvidia.com/gpu: 1affinity:nodeAffinity:requiredDuringSchedulingIgnoredDuringExecution:nodeSelectorTerms:- matchExpressions:- key: gpu.nvidia.com/modeloperator: Invalues:- Tesla_V100
spec:containers:- name: exampleresources:limits:cpu: 2memory: 16Ginvidia.com/gpu: 2affinity:nodeAffinity:requiredDuringSchedulingIgnoredDuringExecution:nodeSelectorTerms:- matchExpressions:- key: gpu.nvidia.com/modeloperator: Invalues:- GeForce_GTX_1070- GeForce_GTX_1070_Ti
spec:containers:- name: exampleresources:limits:cpu: 16memory: 48Giaffinity:nodeAffinity:requiredDuringSchedulingIgnoredDuringExecution:nodeSelectorTerms:- matchExpressions:- key: cpu.coreweave.cloud/familyoperator: Invalues:- xeon
spec:containers:- name: exampleresources:limits:cpu: 1memory: 4Giaffinity:nodeAffinity:requiredDuringSchedulingIgnoredDuringExecution:nodeSelectorTerms:- matchExpressions:- key: cpu.coreweave.cloud/familyoperator: Invalues:- epyc
Kubernetes allows resources to be scheduled with requests and limits. When only limits are specified, the requests are set to the same amount as the limit.
