Ampere In stock

NVIDIA RTX 3090

The cheapest way to get 24GB of VRAM under a CUDA runtime.

Best suited to

Hobby projects
Inference prototypes
Learning and coursework

Available regions

Region	Location	Typical latency
us-east-1	US East (Ashburn, VA)	12 ms
eu-west-1	EU West (Dublin, IE)	84 ms

Hourly rate

$0.29/hr

Billed per second. $6.96 for a full day, $212 if left running a month.

Deploy this GPU All pricing

Full specification

ArchitectureAmpere
VRAM24 GB
Memory typeGDDR6X
Memory bandwidth936 GB/s
FP16 dense142 TFLOPS
FP8 densen/a
CUDA cores10,496
Tensor cores328 (3rd gen)
Board power350W
InterconnectNVLink Bridge (pairs)
Host vCPU8
Host RAM48 GB
NVMe scratch300 GB
Network50 Gbps Ethernet
MIG partitioningNo

Other Ampere cards

A100 80GB

80 GB HBM2e

A100 40GB

40 GB HBM2

RTX A6000

48 GB GDDR6 ECC