80 GB+ VRAM GPUs Cloud Pricing
Train models with hundreds of billions of parameters. Serve 70B+ models at full precision. Run multi-node distributed training with NVLink interconnects. This tier — A100 80 GB, H100, H200, MI300X, and Blackwell-generation GPUs — uses high-bandwidth memory (HBM2e/HBM3/HBM3e) that delivers 2–5x the bandwidth of GDDR, directly impacting training throughput and large-batch inference speed.
80 GB+ VRAM GPUs Available in the Cloud
A100 PCIe 80GB
ServerA100 SXM
B100
ServerB200
Gaudi 2
ServerGaudi 3
ServerGB200
ServerGB300
ServerGH200
H100 NVL
ServerH100 PCIe
ServerH100 SXM
ServerH200
HGX B300
ServerMax 1550
ServerMI250
ServerMI250X
ServerMI300A
ServerMI300X
MI325X
ServerMI355X
ServerRTX PRO 6000
ServerSample 80 GB+ VRAM GPUs Pricing
| Provider | Price / hr |
|---|---|
$1.68/hr 1× | |
$1.99/hr 1× | |
$2.85/hr 1× | |
$3.00/hr 8× | |
$3.60/hr 4× | |
$8.00/hr 8× | |
$8.68/hr 1× | |
$10.00/hr 4× | |
$3.00/hr 1× |
Showing 9 of 795 price points. Visit individual GPU pages above for full pricing.
Frequently Asked Questions
Do I need 80 GB+ VRAM?
If you're training models larger than 13B parameters at full precision, running inference on 70B+ models, or need large batch sizes for production throughput, 80 GB+ is recommended. For smaller workloads, 24–48 GB GPUs are more cost-effective.
What is the difference between HBM and GDDR memory?
HBM (High Bandwidth Memory) provides 2–5x the bandwidth of GDDR6X, which directly impacts training throughput and large-model inference speed. HBM is standard on datacenter GPUs (A100, H100, MI300X) while consumer GPUs use GDDR.