A100 H100 dedicated server The full GPU, your OS, your root.

The H100 is the generation after the A100. As the PCIe GPUs we fit, the A100 has 40 GB of HBM2 at 1,555 GB/s or 80 GB of HBM2e at 1,935 GB/s and a PCIe 4.0 interface. The H100 has 80 GB of HBM2e at 2,039 GB/s, fourth-generation Tensor Cores with FP8, and a PCIe 5.0 interface, and it draws 350 W against 250 or 300 W. All three have ECC, Multi-Instance GPU and no video encoder.

The lowest monthly price for a complete H100 80 GB server is $1,682.67. That machine is Intel Xeon Silver 4110 8 Core 2.10 GHz · 16 GB DDR4 · 2 × SATA-SSD 240 GB (RAID 1) · 1 Gbps, and the GPU is $1,599 of the total. A build with the GPU fitted adds a one-time setup of $29, and the builds table lists every platform with its own price.

The lowest monthly price for a complete A100 server is $772.47 with the 40 GB GPU and $972.47 with the 80 GB GPU, each in the cheapest machine that takes it. The A100 40 GB build is Intel Xeon E5-2620 v4 Octo Core 2.10 GHz · 32 GB DDR4 · 2 × SATA 500 GB (RAID 1) · 1 Gbps. Memory, disks and a faster port are priced on the order form.

PCIe. Our chassis take PCIe add-in GPUs, so the H100 we fit is the 80 GB PCIe version with HBM2e. The SXM5 module, with HBM3 and up to 700 W, needs an HGX baseboard that none of our servers have.

Yes, for models that fit in 80 GB with the training method you use. Full fine-tuning in mixed precision with Adam needs about 16 bytes per parameter before activations, so one H100 fully trains about 5 billion parameters, and LoRA or QLoRA adapters stretch it to much larger models. Past that the job needs several GPUs, which multi-GPU servers covers.

Often, and it is quoted by hand rather than ticked at checkout: chassis clearance, PCIe lanes and the power supply set the count. Both GPUs carry an NVLink bridge connector, but no bridge is on the order form, so GPUs in one server talk over PCIe unless a bridge is quoted. Ask in chat before you order if the job needs one.

Only a machine already racked, ordered as it stands, from instant servers. It costs its monthly price divided by 730 hours, times 1.3, metered by the second until you release it. A build to order, like the rows on this page, is billed by the month.

About 30 minutes after approval when a machine with the GPU is already racked, 4 to 24 hours when we fit a GPU we hold, 5 to 10 working days when the GPU is bought in, and 10 to 15 when it is on allocation. We say which one applies before any payment is taken.

Not in hardware. Neither has an NVENC encoder, so encoding would run in software on the GPU or the processor. They do decode, with 5 NVDEC engines on the A100 and 7 on the H100. For live transcoding pick a GPU with NVENC, such as the L40S.

No. The GPU is passed through to your own operating system with no hypervisor in between, and the machine has one tenant. You can still split it yourself with Multi-Instance GPU into up to seven instances for your own jobs.

The AI on our chat replies within seconds and can look up your account and your server while you talk. Say you want a person and one of us takes over the same conversation.