Guide4 sections2 min read
The short answer
If your code needs double precision (FP64), rent a data center GPU such as the A100, H100, H200 or AMD MI300X. GeForce and most workstation GPUs run FP64 at a small fraction of their single-precision rate.
Check the precision your code uses
Many simulation codes in CFD, finite elements, climate and quantum chemistry rely on FP64. Others, including many molecular dynamics packages, run in single or mixed precision and work well on cheaper GPUs.
Read your software's documentation, or profile a short run, before choosing the GPU class.
Bandwidth and memory often matter more than peak compute
Many scientific kernels are limited by memory bandwidth. HBM GPUs such as the A100 SXM (2,039 GB/s), H100 SXM (3,350 GB/s), H200 (4,800 GB/s) and MI300X (5,300 GB/s) move data much faster than GDDR cards.
Large meshes and datasets also need memory: 80 GB to 192 GB on a single GPU reduces the need to split a problem across GPUs.
Pay the price you choose
If the best offer is above your budget, place a bid at the price you are willing to pay per GPU-hour. The bid is matched when an offer meets it.
During pre-launch, prices shown on EverXCloud are indicative and become rentable on launch day. The table above is read from the live exchange each time the page is generated.
Live prices on EverXCloud
| GPU | Memory | From, per GPU-hour |
|---|---|---|
| NVIDIA H100 SXM 80GB | 80 GB | $2.88 |
| NVIDIA A100 SXM 80GB | 80 GB | $1.79 |
| NVIDIA H200 141GB | 141 GB | $4.19 |
| AMD Instinct MI300X 192GB | 192 GB | $4.16 |
Best live offer on the exchange, excluding tax. Prices are indicative during pre-launch.
Frequently asked questions
- Can I use an RTX 4090 for scientific computing?
- Yes for single or mixed precision codes. For double-precision (FP64) codes, data center GPUs such as the A100 or H100 are much faster.
- Which GPU has the most memory bandwidth for HPC?
- Among the GPUs listed on EverXCloud, the AMD MI300X has 5,300 GB/s and the NVIDIA H200 4,800 GB/s; the GB200 NVL reaches 8,000 GB/s per GPU.
- Are the prices shown final?
- No. During pre-launch, EverXCloud prices are indicative. They are shown per GPU-hour, excluding tax, and become rentable on launch day.