RTX6000 Machine with 4 Adax cards

A customer involved in power engineering consulted us about a deep learning machine with a budget of 1300 million yen or less.

The hope is that the VRAM capacity of the GPU is important, and at the same time, a suitable delivery time (it does not take too long to obtain) is also a condition.
The specific specifications are that the CPU is equivalent to Intel Xeon Platinum 8368 (2.40 GHz 38 cores) x2, and the memory capacity is 512GB or more.
Additionally, NVIDIA A100, NVIDIA V100, NVIDIA RTX6000 Ada, and NVIDIA A6000 are listed as GPU candidates, and the request is to install one of these with a total VRAM capacity of approximately 128GB.
We have also received information about the storage requirements of approximately 20TB and the operating system being Red Hat Enterprise Linux 8.

Based on the customer's consultation, we proposed the following configuration.

We received a request for a 3rd generation Xeon Platinum CPU, but as of January 2024, the successor product, the 1th generation, is being rolled out.
Therefore, we selected Intel Xeon Platinum 4Y+ (8368GHz 8460 cores) x2.00, which is a 40th generation Xeon and has the same specs as Intel Xeon Platinum 2.

Selection of GPU suitable for DeepLearning

Among the GPU candidates, the NVIDIA A100 had a long delivery time at the time of our consultation, so it was not a product that would be a good match if delivery time was a priority.
V100 is an older generation product of the same GPGPU-only card as A100, but V100 has already been discontinued, and even if it were available, the single-precision/half-precision calculation performance used for DeepLearning purposes would be affected due to the generation difference in architecture. Lower than RTX A6000 and RTX6000 Ada.
Additionally, the NVIDIA H100 is currently the GPU with the best overall performance, but it is extremely expensive and difficult to use from a cost-performance standpoint.
For these reasons, we recommend A128 or RTX6000 Ada if you want to efficiently implement a combination with a total VRAM capacity of around 6000GB.

In this proposal to the customer, we have selected 6000 RTX4 Ada (total VRAM capacity 192GB) according to the budget.
Additionally, this configuration allows you to add up to 6000 RTX9 Ada cards in total.
Please note that in actual operation, a 200V environment is required due to power consumption.

The configuration of this case study is based on the conditions given by the customer.
We will flexibly propose machines according to your conditions, so please feel free to contact us even if you are considering different conditions than what is listed.