GPU Cloud
H100 SXM, dedicated.
A single NVIDIA H100 SXM GPU, 80 GB of HBM3 at 3.35 TB/s on the Hopper architecture, on an x86 host. $2.19 per hour, billed per minute, the same terms as every other GPUwerk tier.
Rented first come, first served: deploy an H100 when one is free, or join its waitlist in the console while none is.
Specifications
Per node. Every GPUwerk instance, on any tier, is one whole machine to yourself.
| GPU | H100 SXM5, Hopper architecture |
|---|---|
| Memory | 80 GB HBM3 |
| Memory bandwidth | 3.35 TB/s |
| CUDA cores | 16,896 |
| Max power | Up to 700 W |
| MIG | Up to 7 instances of 10 GB |
| Host | x86 host, configuration published once machines are racked |
| Tenancy | One customer per machine, never shared |
GPU figures from NVIDIA H100 and the H100 architecture whitepaper. No tok/s or training throughput figures yet: no unit is in the fleet, and GPUwerk publishes figures only after running them on its own hardware. See the DGX Spark fleet benchmarks for how they are reported.
What would you run on it?
HBM3 at 3.35 TB/s: the most memory bandwidth of any GPUwerk tier, for throughput rather than capacity.
High-throughput serving
Batched inference for many concurrent users, where memory bandwidth sets the tokens per second.
Training and fine-tuning
FP8 Transformer Engine and HBM3 for full fine-tunes and training runs that fit in 80 GB.
Models up to about 70 GB
A 70B-class model at FP8 fits, with little room left for the KV cache; larger models need several GPUs, see the 8× H100 servers.
Billing
| Rate | $2.19 per hour, billed per minute |
|---|---|
| Billing | Per minute, against a prepaid credit balance |
| Fees | No egress fees |
| Commitment | None |
Prices are in USD, excluding applicable tax. Full billing mechanics, including what a stop versus a terminate does, are on the pricing page.
FAQ
Can I rent an H100 now?
Yes, when one is free. Deploy it from the console; while none is free, the console offers a waitlist instead, and people in line are emailed one at a time, oldest first. No machine is held for anyone: whoever deploys first gets it.
What's the host configuration around the GPU?
The host configuration (CPU, system RAM, storage) is published once the first machines are in the fleet. GPUwerk does not publish numbers for a machine it has not installed.
How does billing work on this tier?
The same as every GPUwerk tier: per minute, against a prepaid credit balance, in USD, excluding applicable tax, with no egress fees and no commitment.
Need GPU time today?
A DGX Spark has 128 GB of unified memory and is rentable now, at $0.79/hour billed per minute. First top-up: pay $20, get $30 in credit. See how a DGX Spark compares, or deploy an H100.