NVIDIA H100 GPU
rental in Bangladesh
Hopper · 80 GB HBM3
The NVIDIA H100 is an NVIDIA Hopper data-centre GPU with 80 GB of HBM3 memory. On ComputeBD you rent it by the hour, work in Jupyter in your browser and pay in taka.
Large model training, transformer workloads, HPC
INDICATIVE RATE৳1,047*/GPU hour*estimate, depends on market rates
SPECIFICATIONS
NVIDIA H100 GPU specs
| Memory | 80 GB HBM3 |
|---|---|
| Memory bandwidth | 3.35 TB/s |
| FP16 / BF16 Tensor | 1,979 TFLOPS |
| FP8 Tensor | 3,958 TFLOPS |
| Architecture | Hopper |
| Interconnect | NVLink 900 GB/s |
| Form factor | SXM |
Tensor figures are NVIDIA's published numbers with sparsity (dense is half). Source: H100.
MONTHLY COST
What does the NVIDIA H100 cost per month?
| Usage | Example | Indicative cost |
|---|---|---|
| 10 hours | A weekend experiment | ৳10,470 |
| 40 hours | A thesis or project month | ৳41,880 |
| 100 hours | Regular weekly training | ৳104,700 |
| 176 hours | Weekdays, 8 hours a day | ৳184,272 |
Calculated as ৳1,047 × hours. Billing is per minute, so you pay only for the time the GPU runs.
WHICH ONE SHOULD I PICK?
When to pick the NVIDIA H100 — and when not to
Pick it if: You train or fine-tune transformers seriously, want FP8, or run large-model inference where speed matters. It is the reference GPU most papers and tutorials assume.
Choose another GPU if: Your model needs more than 80 GB (choose H200) or the job is small enough for an L40S or A100 at a lower rate.
What fits in memory (roughly): 32B in FP16, 70B at 8-bit (tight). Check whether your model will run →
The most expensive GPU is not automatically the best. Check that your model fits in memory, how fast you need results, how long it will run and your budget — matching those four gives the lowest cost.
JUPYTER WORKFLOW
Common questions
FAQ
How much does it cost to rent an NVIDIA H100 in Bangladesh?
About ৳1,047 per GPU hour, so roughly ৳41,880 for 40 hours. Billing is by the minute and you pay in taka with bKash, no dollar card needed.
How much memory does the NVIDIA H100 have?
80 GB of HBM3 memory at 3.35 TB/s. As a rough guide it fits: 32B in FP16, 70B at 8-bit (tight).
What is the NVIDIA H100 good for?
Large model training, transformer workloads, HPC. You train or fine-tune transformers seriously, want FP8, or run large-model inference where speed matters. It is the reference GPU most papers and tutorials assume.
Can I use it from Jupyter?
Yes. You open a Jupyter notebook in the browser, choose the GPU, run your code and stop it with one click when you are done — no server setup.
Is this the final launch price?
No. Pricing is indicative and may change with capacity, exchange rate, tax, storage and network usage.
Compare the NVIDIA H100 and other GPUs
NVIDIA H100 vs NVIDIA H200 →NVIDIA H100 vs NVIDIA A100 →NVIDIA L4 GPU · ৳147/hr →NVIDIA A10 GPU · ৳196/hr →NVIDIA L40S GPU · ৳343/hr →NVIDIA A100 GPU · ৳638/hr →NVIDIA H200 GPU · ৳1,390/hr →NVIDIA B200 GPU · ৳1,652/hr →NVIDIA B300 GPU · ৳1,880/hr →GPU cloud in Bangladesh — how it works →Hourly GPU rental — priced in taka →Every GPU price in one table →