E1 TCO Calculator

This calculator estimates the cost impact of deploying the Xsight E1 in an AI training cluster. Pick the scenario that matches your environment, adjust a few inputs to reflect your own deployment, and the tool shows the savings, payback, and return over a four-year window.

Models a large-scale training cluster of the kind a cloud or hyperscale operator runs on the HGX B200 platform. It assumes high, sustained GPU utilization and the ability to put freed capacity back to work or sell it. Use this view if your GPUs rarely sit idle.

Your environment

Number of nodesUtility rate ($/kWh)PUEGPU hourly rate ($/GPU-hr)

Share of freed GPU capacity you expect to use50%

When the E1 takes networking and infrastructure work off your GPUs, it frees up a share of GPU time. You can bank that as lower power and cooling cost, or put it back to work running more jobs. This slider sets how much of the freed capacity you expect to use rather than bank. The tool starts at a typical value for your chosen use case; move it to match your own situation and the results update.

4-year results

Total 4-Year Value Created$17.0M

Payback Period3 months

ROI325.6%

GPU-Hours Freed (4-Year)4,983,467

Where the 4-year value comes from

Total$17.0M

CAPEX savings$3.0M17.6%
OPEX savings$815,8994.8%
Recovered GPU $$13.2M77.6%

Estimates based on a 4-year analysis at the node count you enter, using an 8-GPU HGX B200 (hyperscaler) or DGX B200 (enterprise) reference node. The ROI accounts for both the cost of the E1 cards (CAPEX) and the infrastructure running costs (OPEX). Assumes conservative infrastructure offload from deploying the E1 DPU, with GPU productivity valued using reserved-rate market assumptions. For full methodology and assumptions, enter your email below and our team will follow up.

Want to go deeper on DPUs?

Enter your details to get the ebook in your inbox.