Operating consumer GPUs on an enterprise server

Economical thinking

Starline engineers have built a powerful GPU server. To do this, they implanted consumer GPUs into enterprise hardware. This means that four comparatively inexpensive RTX 4090 graphics cards from MSI can be used for 3D rendering, AI ML applications, scientific simulations or even decryption. The advantage: the price of the gaming cards is significantly lower than that of professional GPU cards from the enterprise sector.

Four MSI RTX 4090 in the GPU enclosure from Netstor

Our technicians installed the four cards in the Netstor NA265A-G4 and connected the GPU memory to the server via a PCI-SAS connection. Another advantage of this combination is that the server to be used can be a small 1U server with any configuration.

However, there is one drawback. Applications with high data throughput requirements could reach their limits. This is because the PCI-SAS connection transfers with 16 lanes and therefore at a moderate 31.5 GByte per second.