Does the use of two A10 vs. one A30 in terms of GPU memory have a disadvantage for the AI/ML calculation in terms of the size of the model (>24GB)? I see that two A10s should be more powerful than one A30 in terms of computing power and, according to Release Notes - NVIDIA Docs, one unified memory for the A10 provides a comprehensive memory address range. That sounds to me like two A10s come across as a combined, stronger GPU from a vServer perspective and the shared GPU memory can be assigned to a vServer as desired. Am I seeing it that way correctly?
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| 2 NVIDIA A2 GPUs | 0 | 493 | September 16, 2022 | |
| How A30 GPU is faster than A10 GPU? | 3 | 5594 | July 5, 2022 | |
| Physical RAM and GPU VRAM capacity matching? | 0 | 743 | July 11, 2023 | |
| Max M10 boards per server | 4 | 4855 | January 16, 2019 | |
| Maximize memory available | 1 | 747 | May 29, 2013 | |
| Same gpu with different memory size | 1 | 654 | January 12, 2022 | |
| 10*A6000 or 10*A40 for training large language models? | 0 | 1059 | January 10, 2023 | |
| A10 GPU memory was only 22731M | 5 | 1116 | January 27, 2022 | |
| Windows 10 using ~1 GB of memory for all GPUs (WDDM) | 3 | 5689 | October 22, 2017 | |
| Memory exception | 2 | 23 | July 24, 2024 |