Hosting
Web HostingWordPress HostingReseller HostingGPU ServersServers
GPU Dedicated ServersVPS HostingVPS in Saudi ArabiaVPS in QatarVPS in the UAEDedicated ServersDomainsAdd-ons
Business EmailSSL CertificatesAboutContactGet startedHome/GPU Servers
NVIDIA L40S, RTX Pro and Quadro on Xeon Platinum and EPYC, hosted in the Kingdom or in India. Single-tenant, unlimited bandwidth, twelve configurations from SAR 3,243.75 a month.
Hosted in Saudi Arabia or India, your choice at order. Single-tenant and billed monthly with no minimum term - the whole card is yours, with no shared allocation and no neighbour competing for VRAM.
Inference, light fine-tuning, CAD and rendering workloads.
Billed monthly, no minimum term
Inference at volume, with memory headroom for bigger batches.
Billed monthly, no minimum term
The same GPU with half as much RAM again for data-heavy pipelines.
Billed monthly, no minimum term
Fine-tuning smaller models without queueing for shared cloud capacity.
Billed monthly, no minimum term
More VRAM and system memory for longer training runs.
Billed monthly, no minimum term
Platinum-class CPU alongside a current-generation professional GPU.
Billed monthly, no minimum term
Training workloads that are as hungry for RAM as for VRAM.
Billed monthly, no minimum term
Two cards for parallel training or serving several models at once.
Billed monthly, no minimum term
Dual professional GPUs with half a terabyte of ECC memory.
Billed monthly, no minimum term
The datacentre card built for LLM inference and generative workloads.
Billed monthly, no minimum term
Dual L40S for serving large models, or training mid-sized ones in region.
Billed monthly, no minimum term
128 CPU cores beside two L40S - the largest configuration we sell.
Billed monthly, no minimum term
Prices exclude VAT. Saudi customers are invoiced with 15% VAT. Elsewhere, VAT is added where applicable — 5% in the UAE and Oman, 10% in Bahrain, and Qatar currently levies no VAT. Indian customers are invoiced with 18% GST. Riyal and dollar prices are exact equivalents at the fixed 3.75 SAR peg.
| Plan | GPU | VRAM | CPU | Cores | RAM | Price / month |
|---|---|---|---|---|---|---|
| GPU Dedicated - 01 | NVIDIA Quadro RTX 4000 | 8 GB | Intel Xeon Gold 6246 | 24 | 128 GB | SAR 3,243.75 |
| GPU Dedicated - 02 | NVIDIA RTX 2000 | 16 GB | Intel Xeon Gold 6254 | 36 | 256 GB | SAR 4,031.25 |
| GPU Dedicated - 03 | NVIDIA RTX 2000 | 16 GB | Intel Xeon Gold 6254 | 36 | 384 GB | SAR 4,425 |
| GPU Dedicated - 04 | NVIDIA RTX Pro A4000 | 24 GB | Intel Xeon Gold 6148 | 40 | 256 GB | SAR 4,665 |
| GPU Dedicated - 05 | NVIDIA RTX Pro A4000 | 32 GB | Intel Xeon Gold 6148 | 40 | 384 GB | SAR 5,452.50 |
| GPU Dedicated - 06 | NVIDIA RTX Pro 4500 | 32 GB | Intel Xeon Platinum 8260 | 48 | 256 GB | SAR 6,225 |
| GPU Dedicated - 07 | NVIDIA RTX Pro 4500 | 32 GB | Intel Xeon Platinum 8260 | 48 | 384 GB | SAR 7,807.50 |
| GPU Dedicated - 08 | 2x NVIDIA RTX Pro 4000 | 48 GB | Intel Xeon Gold 6148 | 40 | 512 GB | SAR 8,587.50 |
| GPU Dedicated - 09 | 2x NVIDIA RTX Pro 4500 | 64 GB | Intel Xeon Platinum 8260 | 48 | 512 GB | SAR 9,862.50 |
| GPU Dedicated - 10 | NVIDIA L40S | 48 GB | Intel Xeon Platinum 8260 | 48 | 256 GB | SAR 10,293.75 |
| GPU Dedicated - 11 | 2x NVIDIA L40S | 96 GB | Intel Xeon Platinum 8260 | 48 | 512 GB | SAR 13,275 |
| GPU Dedicated - 12 | 2x NVIDIA L40S | 96 GB | AMD EPYC 7742 | 128 | 512 GB | SAR 16,361.25 |
Every configuration includes 960 GB of SSD plus 5.85 TB of additional SSD, unlimited bandwidth on a 1 Gbit uplink, a dedicated IP and DDoS protection. Monthly prices exclusive of VAT. Riyal and dollar figures are exact equivalents at the fixed 3.75 SAR peg.
The L40S configurations carry 48 GB or 96 GB of VRAM, which is where serving quantised large models stops being a compromise.
24 GB to 96 GB of VRAM with up to 512 GB of ECC system memory, so the data pipeline does not starve the card.
Quadro and RTX Pro cards for CAD, visual effects and engineering simulation.
Two locations - Saudi Arabia and India - chosen at order. Saudi placement keeps AI workloads under the PDPL and inside the country.
Single-tenant hardware. Your model weights and training data are not sitting in a shared tenancy alongside someone else's.
Capacity is allocated to you for the month. You are not bidding for spot instances or waiting for a region to free up.
Configured and secured by the SmartShield team before you get the credentials.
Hourly GPU cloud is the right answer for bursty experiments. The moment you have a model in production, or a training run that goes all month, a dedicated card is usually cheaper and always more predictable - one invoice, no egress surprises, no capacity that vanishes mid-run.
It is set by the VRAM your model needs, not by the CPU. For inference on quantised models under 13B, the 16 GB and 24 GB cards are usually enough. For larger models, or for serving several at once, the L40S configurations at 48 GB and 96 GB are the sensible floor. Tell us the model and we will tell you the smallest card that runs it.
Yes. GPU servers run from two locations — Saudi Arabia and India — and you choose at order. For AI work under Saudi data-residency obligations that matters: your model weights and training data stay inside the Kingdom, on single-tenant hardware, and we confirm the placement in writing.
Latency to an Indian user base, or an existing team and data already there. For a Saudi audience, or anything touching the PDPL, choose the Kingdom. The specification and the price are the same either way, so you are choosing on location alone.
No. These are single-tenant machines and the whole card is yours for the term. The VRAM in the table is what you get, not a slice of it.
We can. Tell us your stack and we will hand the machine over ready to run rather than as a bare OS. Driver and framework versions matter enormously for GPU work, so we would rather agree them with you than guess.
Yes. Operating system is your choice, and Windows licensing is quoted separately.
Monthly, with no minimum term, in riyals, dollars or rupees. Saudi invoices carry 15% VAT shown separately.
Send the model, the batch size and how long the runs take. We will recommend the smallest configuration that does the job rather than the largest one you would buy.