Ultra-Tier GPUs Cloud Pricing
Ultra-tier GPUs are the highest-performance accelerators available for cloud compute. This tier includes the H100, H200, A100 80 GB, MI300X, and Blackwell-generation datacenter GPUs. They feature HBM memory (80–192 GB+), NVLink interconnects, and the highest tensor core throughput. These are the standard for large-scale model training, high-throughput inference serving, and HPC workloads.
Ultra-Tier GPUs Available in the Cloud
A100 PCIe 40GB
A100 PCIe 80GB
ServerA100 SXM 40GB
ServerA100 SXM 80GB
B100
ServerB200
Gaudi 3
ServerGB200
ServerGB300
ServerGH200
H100 NVL
ServerH100 PCIe
ServerH100 SXM
ServerH200 NVL
ServerH200 SXM
HGX B300
ServerMax 1550
ServerMI250
ServerMI250X
ServerMI300A
ServerMI300X
MI355X
ServerRTX 5090
RTX PRO 6000
ServerSample Ultra-Tier GPUs Pricing
| Provider | Price / hr |
|---|---|
$0.810/hr 1× | |
$1.15/hr 4×8× | |
$1.45/hr 8× | |
$1.68/hr 1× | |
$2.45/hr 8× | |
$2.50/hr 2× | |
$3.87/hr 1× | |
$4.54/hr 1× |
Showing 9 of 1087 price points. Visit individual GPU pages above for full pricing.
Frequently Asked Questions
Which ultra-tier GPU should I choose?
For the broadest software compatibility and availability, the H100 is the safest choice. The H200 offers more memory (141 GB) for memory-bound workloads. The MI300X has the most VRAM (192 GB) at a potentially lower price point. Compare current rates above.
Do I need ultra-tier GPUs for LLM training?
For training models above 13B parameters from scratch, ultra-tier GPUs with HBM are strongly recommended. For fine-tuning or training smaller models, high-tier GPUs can be more cost-effective. The key factors are VRAM capacity and inter-GPU bandwidth.