Adopting the professional GPUs like the RTX A6000 can be the easiest way to achieve the performance improvement without worrying about other factors. However, the unit cost of $4,650 makes the overall cost exceeding the budget for the performance upgrade.
The unit cost of the professional GPU model is approximately 3 times compare to the GeForce RTX3090 while the performance difference is under 30%.Instead of going for the expensive professional products, implement H3 Platform Falcon 42 series solution to power the servers with RTX 3090s. A little less performance with a great cut in costs.
The MSRP difference between the RTX A6000 and the RTX 3090 is around $3,000 and every node requires 8 GPUs. The number adds up to $24,000 excluding the cost of new server chassis. The overall cost reduction for the organization is significant as the IT system scales out.
The highest TDP that a GPU has was approximately 250~300W. However, most of the high-end GPUs introduced by Nvidia has TDP exceeding 300W, TDP of the GeForce RTX 3090 even comes to 350W and can reach 450W when stressed. This design makes no existing GPU server capable of handling the power supply issue of the RTX 3090.
As a dedicated GPU solution, the Falcon 42 series offers up to 450W power to any PCIe device. In addition, the cooling solution is designed specifically for multiple GPUs, the fans and chassis chamber are aligned so that the airflow can successfully cool down the heat generated by the GPUs.
We successfully added 8 GeForce RTX3090 GPUs to each existing server with Falcon GPU solution and assured that the device temperature stays under control.
IT resource management is critical in any organization and academical institution is no exception. IT staffs need to keep track of every piece of GPU so they can respond immediately when error happens.
H3 management software provides comprehensive system and device information via user friendly graphical interface. Real-time system status and cluster topology helps IT staffs to monitor system health more easily.
IT staffs knows exactly where each GPU is located in the server room. In addition, uniformed management software makes it easier to operate the system, even for people with less IT experience.
""The disaggregated GPU solution helps the school to overcome challenges in IT system upgrade""
The academic institution offers its students GPU accelerated computing systems for data analytical purpose. The institution adopted hundreds of RTX 2080Ti GPUs previously, and is looking to upgrade its GPUs to the Ampere GPUs due to the huge performance improvement. The school decided to go for RTX 3090 for a couple reasons: 1. Given the amount of GPU it needs, adopting the Quadro or Tesla GPUs would easily go over budget. 2. The RTX 3090 provides performance close to the higher-end GPUs with an acceptable price tag.
However, different from Nvidia’s last generation GPUs, there is a great increase in power consumption for the RTX 3090. The GPU server model they had could carry 8 RTX 2080Ti GPUs, but now it could only carry maximum 4 RTX 3090s due to the higher TDP. Moreover, whether the cooling solution could handle the amount of heat generated by the RTX 3090 remains unknown. On the other hand, the institution needs a rack mount solution that could fit into its IT room, custom build workstation is not an option in their case.
By deploying H3 Platform Falcon solution, the institution successfully overcome the technical challenges including: