Request for Extension of Time Limit and Increased RAM Allocation for Model Execution

Hi,

I am writing to discuss the constraints we are facing regarding the time limit and memory allocation for our model execution. These limitations significantly hinder our model's performance, especially when dealing with high-resolution cases.

To address this issue, I kindly request the following:

Extension of Time Limit: We propose extending the current time limit allocated for model execution to 10-15 minutes. This additional time will greatly enhance our model's ability to process intricate cases and produce more accurate results.

Increased RAM Allocation: The high-resolution cases require more memory resources. Therefore, we would appreciate an increase in the allocated RAM, which will reduce memory-related bottlenecks and improve overall performance.

We understand the potential implications on system resources and are open to discussing alternative suggestions or compromises. Our goal is to overcome the current limitations and optimize our model's performance.

Thank you for considering our request. We look forward to your response and the possibility of finding a suitable solution.

Dear syz,

thank you for bringing up the concerns regarding the computing resources.

Regarding the time limit, I understand that the initial allocation may have felt restrictive. In response to your input, we've extended the time limit to a more accommodating 15 minutes per case. This should provide ample time for tasks such as CT and MR image registration as well as (multi-modal) segmentation.

Now, concerning the memory constraints:

• While we can't increase the allocated GPU memory, each execution job is already equipped with a Nvidia T4 card with 16 GB of RAM. This allocation should suffice for most inference processes. If you find that more GPU memory is necessary, I recommend exploring optimization techniques such as shifting tensors to the CPU once you finish inference and considering the use of half-precision floating-point format where appropriate.

• For CPU memory, the maximum capacity stands at 32 GB, and you have the flexibility to set this parameter when creating a new algorithm. This limit aligns with standards applied in similar challenges, some of which involve even larger input files. To make the most of this allocation, consider employing memory-efficient data structures, such as 8- or 16-bit structures, as alternatives to 64-bit formats. Additionally, each execution job includes 225 GB of NVMe space designed for paging, offering an additional resource to address memory needs if they arise.

Hope this helps, but please don't hesitate to reach out if you have any further questions or require additional assistance.

Best, Gasper