Home / NVIDIA GPUs / H200 NVL 141GB
Hopper · 141GB HBM3e · For servers
NVIDIA H200 NVL 141GB GPU, in Egypt & the UAE.
NVIDIA’s Hopper data-centre GPU as a PCIe card: 141GB of HBM3e at 4.8 TB/s, with NVLink bridges for 2 or 4 cards. Built for training, fine-tuning and serving large models in your servers, factory-new with a 12-month replacement warranty.
141GB HBM3e
4.8 TB/s of memory bandwidth on every card.
NVLink bridges
Link 2 or 4 cards at 900 GB/s per GPU.
Passive, for servers
Up to 600W, cooled by the server’s fans. Not for desktops.
12-month replacement
Factory-new, fit-checked by our engineers, delivered in Egypt and the UAE.
What it’s for
For the largest models you run yourself.
HBM3e memory is about 2.7 times faster than the RTX PRO 6000’s GDDR7, which is what training and long-context inference need.
Training and fine-tuning
Train and fine-tune large models, with NVLink to split the work across cards.
Large-model inference
A 70B model at FP8 with long context and many users per card, and larger models across 2 or 4 linked cards.
HPC and science
30 TFLOPS of FP64, and 60 TFLOPS on FP64 Tensor Cores, for simulation and research.
Share one GPU
Up to 7 isolated MIG instances of 16.5GB each.
Confidential computing
Protects data and models while they are in use on the GPU.
Vision pipelines
7 NVDEC and 7 JPEG decoders for video and image AI.
Before you buy
A data-centre GPU, not a workstation card.
Passive cooling
There are no fans on the card. It needs a server built to cool 600W passive GPUs, such as NVIDIA-Certified and MGX systems.
No display outputs
The H200 NVL is for compute only. For graphics and rendering, choose the RTX PRO 6000.
Export requirements
H200 supply is subject to export requirements. We confirm availability and any end-use paperwork on your quote.
Specifications
NVIDIA H200 NVL specifications.
NVIDIA’s published specifications for the H200 NVL PCIe card.
GPU: NVIDIA H200 NVL
| Architecture | NVIDIA Hopper |
|---|---|
| GPU memory | 141GB HBM3e |
| Memory bandwidth | 4.8 TB/s |
| FP64 | 30 TFLOPS |
| FP64 Tensor Core | 60 TFLOPS |
| FP32 | 60 TFLOPS |
| TF32 Tensor Core | 835 TFLOPS* |
| BF16 / FP16 Tensor Core | 1,671 TFLOPS* |
| FP8 Tensor Core | 3,341 TFLOPS* |
| INT8 Tensor Core | 3,341 TOPS* |
| Multi-Instance GPU | Up to 7 instances @ 16.5GB |
| Decoders | 7 NVDEC, 7 JPEG |
* With sparsity.
Card and installation
| Max power | Up to 600W (configurable) |
|---|---|
| Form factor | PCIe, dual-slot, passive |
| Interface | PCIe 5.0 x16 (128 GB/s) |
| NVLink | 2- or 4-way bridge, 900 GB/s per GPU |
| Confidential computing | Supported |
| Servers | NVIDIA MGX and NVIDIA-Certified systems with up to 8 GPUs |
| Warranty | 12-month Mega Build Up replacement warranty |
Want it ready-built? See our H200 8-GPU server.
How we deliver
From first question to a working GPU.
01
Tell us your system
Share the server or workstation model and what you want to run on it.
02
We check the fit
Slots, power connectors, power supply, airflow and drivers, before you order.
03
Source factory-new
New, genuine NVIDIA GPUs at today’s price.
04
Deliver or install
Delivered across Egypt and the UAE, or installed and tested by our engineers.
05
12-month replacement
If a card fails within 12 months, we replace it.
Compare NVIDIA GPUs
Four GPUs, from 32GB to 141GB.
Efficient AI and design
RTX PRO 4500 Blackwell
32GB GDDR7
Up to 896 GB/s · 200W workstation or 165W server card
Large models on a desk
RTX PRO 5000 Blackwell
72GB GDDR7
1,344 GB/s · 300W blower · AI workstations
Flagship RTX PRO
RTX PRO 6000 Blackwell
96GB GDDR7
Up to 1.8 TB/s · workstations and servers
Training and HPC
NVIDIA H200 NVL
141GB HBM3e
4.8 TB/s · NVLink · servers only
You are here
Can I use the H200 NVL in a workstation?
No. It is passively cooled, relies on server fans and has no display outputs. For a workstation, choose the RTX PRO 6000 or the RTX PRO 5000 72GB.
H200 NVL or RTX PRO 6000?
Choose the H200 NVL for training, fine-tuning large models, long-context inference and FP64 science: it has 141GB of HBM3e at 4.8 TB/s and NVLink. Choose the RTX PRO 6000 for inference mixed with rendering, video and graphics.
What do the NVLink bridges do?
They connect 2 or 4 cards at 900 GB/s per GPU, about seven times PCIe 5.0, so AI frameworks can split a model and its training across the linked cards.
Which servers support it?
Servers designed for passive 600W GPUs, such as NVIDIA MGX and NVIDIA-Certified systems, with up to 8 cards. Tell us your server model, or see our ready-built H200 8-GPU server.
Is the H200 available in Egypt and the UAE?
We supply it subject to export requirements. Your quote confirms availability, lead time and any end-use paperwork.
What warranty do the GPUs have?
Every GPU comes with a 12-month Mega Build Up replacement warranty. If a card fails, our engineers diagnose it and we replace it. See our warranty services.
How do I get a price?
GPU prices change often, so we quote today’s price. Message us on WhatsApp or answer three quick questions in the form on this page, and an engineer replies within 24 working hours with the price and availability of the H200 NVL.
Get a quote
Get today’s H200 NVL price.
Answer three quick questions and an engineer replies within 24 working hours with today’s price, availability and a fit check for your system.
- No obligation, no sales pressure
- Fit checked by an engineer before you order
- 12-month replacement warranty
Step 1 of 3
Talk to an AI infrastructure engineer.
Call, WhatsApp or email our team and we’ll reply within 24 working hours.
