KX Software Usage Terms vRAM

Within this document, the terms “we”, “us”, “our” and “ourselves” refers to the Supplier defined in the Order Form and “you”, “your” and “yourselves” refers to the Customer defined in the Order Form.

1. Usage Reporting Terms

Usage Report. You will provide us with a usage report (“Usage Report”) every three (3) months from the Effective Date in respect of the preceding three-month period (“Report Period”) containing the following information:

(i) a list of each public cloud, private cloud or on-premises environment in which the Licensed Software ran on, or had access to, a GPU during the Report Period. For each environment, identify the relevant Named Host, Virtual Machine, Container, Kubernetes cluster or Pod, GPU Partition and GPU, and the dates on which each was in use; and

(ii) the vRAM Capacity (defined in section 2) of each item listed under (i), recorded every hour on the hour on each of the 5 Business Days immediately before the end of the Report Period (the “vRAM Measurement Period”), when the relevant environment is running and under similar operating conditions as you use the Licensed Software in the ordinary course.

Usage Report will: (a) identify each environment reported; (b) state the highest vRAM Capacity configured for the Licensed Software in that environment at any time during the vRAM Measurement Period; and (c) confirm that the information is complete and accurate. For the purposes of section 2, the vRAM Capacity reported in this way will be treated as the vRAM Capacity of that environment at each hour of the vRAM Measurement Period. You will provide the configuration records that support the figure reported. If those records, or any other information provided under this section 1, show a higher vRAM Capacity, we will use the higher figure. Each Usage Report will be provided to us within 15 days of the end of the Report Period, and you will subsequently provide any other information and technical reports reasonably requested by us to verify the Usage Report.

Where you have purchased Managed Services, you will grant us such access as is necessary to allow us to generate the Usage Reports and information set out above.

Completed reports should be sent to: kx-self-audit-reports@kx.com.

2. vRAM Capacity Definition and Calculation

vRAM Capacity is the GPU memory provisioned to, and available for use by, the Licensed Software, measured in gigabytes (GiB). vRAM Capacity is defined as follows for each target environment:

• Physical GPU (Bare Metal or Named Host) – The total on-board memory of each GPU that is visible to, or assigned to, any process of the Licensed Software.
• Virtual Machine with GPU passthrough – The total on-board memory of each passed-through GPU.
• NVIDIA vGPU – The frame-buffer size of each vGPU profile assigned to the Virtual Machine.
• NVIDIA MIG – The memory of each MIG instance assigned to the Licensed Software, not the memory of the parent GPU.
• Docker – The memory of each GPU, or GPU Partition, exposed to the Container, whether or not a memory limit is set inside it.
• Kubernetes – The memory of each GPU, or GPU Partition, requested by or allocated to the Pods running the Licensed Software.
• Time-sliced or shared GPU (no memory isolation) – The total on-board memory of the GPU, because no memory limit separates the Licensed Software from other workloads.
• Shared GPU with enforced memory limit – The enforced limit (for example under NVIDIA MPS or a Kubernetes device plugin), if it is set and can be verified.

Where a GPU and CPU draw on a shared memory pool (unified or coherent memory), vRAM Capacity is the portion of that pool provisioned for use as GPU memory by the Licensed Software.

“GPUs” are defined as graphics processing units or other accelerators that provide dedicated or high-bandwidth memory to the Licensed Software.

“GPU Partitions” are defined as defined shares of a GPU, including a MIG instance or a vGPU profile.

“Containers” are defined as packages of applications and execution environments running the applicable Licensed Software.

“Pods” are defined as collections of closely related or tightly coupled Containers.

“Cgroups” (abbreviated from Control Groups) are defined as a Linux kernel feature that limits, accounts for, and isolates the resource usage (CPU, memory, disk I/O, etc.) of a collection of processes.

Calculation of vRAM Capacity. We will calculate your vRAM Capacity for the purposes of assessing your usage by averaging the five (5) highest peaks of vRAM Capacity taken across the vRAM Measurement Period. Each GPU and GPU Partition is counted once, even if it is visible to more than one Container or Pod.

Relationship to RAM. These terms apply only to vRAM Capacity. Where an Order Form also has Licensed Software on a RAM Capacity basis, it is measured, reported and charged separately under its own Software Usage Terms. Unless the Order Form states otherwise, vRAM Capacity does not count towards or reduce your RAM Capacity, and RAM Capacity does not count towards or reduce your vRAM Capacity.

3. Usage Review Policy.

In the event that a Usage Report reveals that your vRAM Capacity exceeds the level for which you are licensed in the Order Form, you will be charged for the additional vRAM Capacity with effect from the end of the relevant Report Period to the end of the Initial Subscription Term or Renewal Subscription Term (as applicable) using the rates specified in the Order Form (and if no rate is stated, then a new Order Form shall be required). If you consistently exceed your licensed vRAM Capacity, then we reserve the right to charge you for the additional vRAM Capacity from the commencement of the relevant Report Period.