24GB GDDR6
Massive graphics memory for large models and datasets.
Powerful performance for AI training, AI inference, rendering, compute-intensive workloads and multimedia.
Run advanced workloads directly in the cloud, without buying and operating your own hardware. Our GPU instances are built for businesses, developers and creators who demand high performance, low latency and full control.
Modern hardware: Nvidia L4 (PCIe Gen 4) with 24 GB GDDR6 and the energy-efficient Ada Lovelace architecture.
Flexible instances: scale up or down as needed.
Local infrastructure: low latency within the EU, with servers located in Umeå, Sweden.
Transparent pricing: the most performance for your money.
Massive graphics memory for large models and datasets.
Full control over the entire GPU, no sharing with other customers.
Compute power for AI and rendering.
Specialized for AI inference, AI training and video processing.
2.5x better performance per watt compared to previous generations.
Data center in Umeå, Sweden.
The L4 family consists of GPU servers with Nvidia L4 GPUs, AMD EPYC 7713 CPUs, DDR4 RAM and high-IOPS local NVMe drives in RAID.
l4.gpu1
SEK 3.91€0.3555 / hour
or
SEK 2,627€239 / month
l4.gpu2
SEK 4.28€0.3891 / hour
or
SEK 2,875€261 / month
l4.gpu3
SEK 4.82€0.4382 / hour
or
SEK 3,239€294 / month
l4.dual-gpu
SEK 10.69€0.9718 / hour
or
SEK 7,187€653 / month
Based on 672 hours per month. If you use the same instance for a full month, the cost never exceeds the fixed monthly price.
1 IPv4 address and 1 IPv6 address are included per instance.
All operations and data storage take place in our data center in Umeå, within the EU. No data leaves the country, and you know where your data and your users' data is.
We know our customers prioritize speed, which is why all data is stored on enterprise-grade NVMe SSDs. Faster and more energy efficient! GPU instances use local NVMe storage in RAID.
Internet connectivity for our services is delivered by us through our own AS number. A stable and secure connection with low latency.
A cloud GPU instance is a virtual machine at a cloud provider with a GPU (graphics card) attached to the virtual machine. The GPU can be either a shared vGPU or a full GPU via PCIe passthrough.
A regular CPU-based VPS is good for general-purpose applications, but some workloads require the parallel processing power of a GPU. With a GPU instance you can:
In short: CPU for general work, GPU for accelerated computing.
All our GPU instances come with dedicated NVMe storage local to the server. This disk is used as scratch disk or primary storage and delivers extremely high IOPS and low latency compared to traditional network storage. For workloads such as AI training, data processing and video processing, this means much faster access to your data.
We offer flexible hourly billing where you only pay for the time your instance is running. For customers who want long-term operation, we also show the maximum monthly price. Your registered payment card is charged automatically and there are no lock-in periods: you can start and stop your GPU instances whenever you want.
Businesses can apply for invoicing as a payment method by submitting a support ticket in Hexabyte Cloud.
You get full root access to your GPU instance and can install your own drivers, libraries and software. This means you can set up anything from Docker containers with PyTorch/TensorFlow to full virtual workstations (vWS) with, for example, Remote Desktop or NICE DCV.
Support is handled through support tickets directly in Hexabyte Cloud, and you can write in both Swedish and English. Our technicians are based in Sweden and help you with operational questions, instance management and troubleshooting.
No. All our GPU plans include only dedicated NVIDIA L4 GPUs, attached via PCIe passthrough. This means you do not share GPU resources with other customers, unlike some cloud providers that offer “shared vGPU” with lower performance. With us, performance is predictable and guaranteed.
Absolutely. The NVIDIA L4 is excellent for both AI inference and model training. The card supports FP32, FP16 and INT8, which makes it efficient for both training and optimized inference. You can use popular frameworks such as TensorFlow, PyTorch and JAX without any problems.