Reference profiles
From a desktop GPU to a data center.
The software is designed to run across a range of hardware. These are reference profiles, not recommendations, offers, or commitments that any specific hardware is available or exportable.
The software is in development. No software license or hardware is offered for sale on this site. GDLG is a software company. We do not sell, manufacture, or distribute hardware, and we do not have any hardware partnership to announce.
Compatibility examples
One software profile, several hardware profiles.
The same software runs on every profile. A hardware profile decides model size, quantization, batch size, and concurrency. No performance or concurrency numbers are published until they are measured.
| Profile | Example hardware | Memory | Typical use |
|---|---|---|---|
| Desktop | NVIDIA RTX 5090 | 32 GB GDDR7 | Individual analyst, development, and smaller local models. |
| Compact | NVIDIA DGX Spark | 128 GB unified | Development, edge evaluation, and small teams. |
| Deskside | NVIDIA DGX Station | 748 GB coherent (252 GB HBM3e + 496 GB LPDDR5X) | Long-document analysis, several sector pipelines, and full-sector recomputation at the desk. |
| Data center | NVIDIA H200 | 141 GB HBM3e per GPU | Configurations intended for larger models and multi-user deployments. |
| Data center | NVIDIA B300 | 288 GB HBM3e per GPU | Configurations intended for larger models and multi-user deployments. |
| Data center | NVIDIA GB300 | 288 GB HBM3e per GPU | Grace-Blackwell platform for larger deployments. |
Memory figures are hardware vendor specifications and may vary by configuration. Data-center figures are per GPU. Other CUDA-capable devices may also work; performance varies by hardware.
What these profiles are
Examples of hardware the software is designed to run on. They help a customer size a machine for a legitimate use case. They are not a product catalogue and do not mean that GDLG has the hardware, sells it, or can ship it.
What these profiles are not
They are not recommendations, offers, price lists, or commitments that any specific configuration is available or exportable. We do not recommend specific controlled hardware or assist in procurement. Hardware selection and acquisition are solely the customer's independent decision. A CPU-only license is available where GPU use is not authorized.
NVIDIA, RTX, DGX, H200, B300, and GB300 are names or trademarks of NVIDIA Corporation. They appear here as compatibility examples only. GDLG is not affiliated with or endorsed by NVIDIA.
Where the planned hardware configuration, destination, end user, or end use requires a U.S. government export license or other government authorization, we may require a copy of that license or authorization before enabling GPU features. Where no license is required, we may accept a written no-license opinion from your counsel, subject to our review and acknowledgement. We may ask our counsel to review the opinion if there is any ambiguity, controlled configuration, high-risk destination, or red flag. A hardware vendor's written authorization may be requested as supporting evidence, but it is not a substitute for a government license where one is required. If the required evidence is not provided, we may refuse the GPU-enabled license or offer a CPU-only license. A CPU-only license does not authorize GPU use. We do not obtain the government license on the customer's behalf.
Some of these configurations are subject to U.S. export control and sanctions laws. Customers must not resell, re-export, or divert hardware or software to prohibited destinations or end users, including mainland China, Hong Kong, Macau, embargoed countries or regions, restricted parties, or prohibited end uses. We do not obtain BIS export licenses on behalf of customers. The customer and the hardware vendor are responsible for determining whether a license is required and for obtaining it. See the Export Compliance Statement.