Launch pricing · limited time
Catalog
Every accelerator you'd actually use.
Reserve any GPU below by the hour, day, or month. Multi-node pods up to 1,024 GPUs available on request. Introductory rates apply to the first wave of customers and will increase as capacity fills.
Most popularReserve
NVIDIA H100 80GB SXM5
A strong default for sustained training workloads.
- VRAM
- 80 GB HBM3
- Fabric
- NVLink
- Throughput
- 1,979 TFLOPS FP16
$1.99
$2.89
per GPU / hour · launch rate
NewReserve
NVIDIA H200 141GB
Suited to large-parameter inference and long-context serving.
- VRAM
- 141 GB HBM3e
- Fabric
- NVLink
- Throughput
- Optimized for inference
$2.69
$3.90
per GPU / hour · launch rate
NVIDIA A100 80GB
A workhorse for fine-tuning, distillation, and batch inference.
- VRAM
- 80 GB HBM2e
- Fabric
- NVLink
- Throughput
- 624 TFLOPS FP16
$0.99
$1.44
per GPU / hour · launch rate
NVIDIA L40S
Cost-efficient inference, rendering, and multimodal serving.
- VRAM
- 48 GB GDDR6
- Fabric
- PCIe Gen4
- Throughput
- 362 TFLOPS FP16
$0.59
$0.86
per GPU / hour · launch rate
NVIDIA RTX 4090 24GB
A good fit for prototyping, LoRA, and small-model R&D.
- VRAM
- 24 GB GDDR6X
- Fabric
- PCIe Gen4
- Throughput
- 82.6 TFLOPS FP32
$0.24
$0.35
per GPU / hour · launch rate
LimitedReserve
AMD MI300X 192GB
High single-GPU memory, useful for large-parameter inference.
- VRAM
- 192 GB HBM3
- Fabric
- Infinity Fabric
- Throughput
- 1,307 TFLOPS FP16
$2.29
$3.32
per GPU / hour · launch rate