gpu deep learning server configuration recommendations

Not I want to recommend, is the customer demand, must be dual 96-core CPU, 6 card 4090 server, I also do not know what to say, also said want to on the A800, first experience 96-core CPU ride 4090 how, 500,000 within the budget really want to spend ah.

　　This machine allows me to match, or according to customer requirements to give the configuration, 400,000 is worth it?

　　1, the box: 4U rackmount chassis 178mm × 437mm × 737mm (H × W × D)

　　2, CPU: 96 core 2.4GHz AMD EPYC 9654

　　3, memory: 64GB DDR5-4800 ECC REG

　　4、System Disk: 500GB M.2 NVMe SSD Enterprise SSD

　　5、Data drive: 18TB 256MB 7200 RPM SATA enterprise hard drive

　　6、GPU: NVIDIA Geforce RTX 4090 all-in-one liquid-cooled version

　　7、CUDA environment: CUDA driver, CUDA toolkit, CUDA Samples

　　8、Application software: provide TensorFlow, Pytorch and other software installation services and technical support

　　9、Wind Tiger Information Machine Management System.

　　Support the creation of administrator accounts and a number of ordinary user accounts, account permissions settings, when submitting multiple jobs

　　Support for job status monitoring, queuing, priority, deletion and other management functions.

　　10、OS: Ubuntu 20.04 LTS 64-bit Server Edition

　　11、Motherboard platform.

　　Support dual-way AMD EPYC 9004 series processors, 24 DDR5 4800MHz ECC REG DIMM. 1 M.2 2260, 2280, 22110 SSD hard drive bays, four 2.5" hot-swappable U.2 hard drive bays, 20 2.5" hot-swappable SATA3 hard drive bays, six 3.5" hot-swappable SATA3 hard drive bays (optional). Supports RAID 0, 1, 5, 10. 9 (x16) PCI-E 5.0 full height service slots, supports up to 8 double-wide GPU cards. On-board ASPEED AST2600 integrated graphics card, on-board dual-port 10GbE network card, independent IPMI 2.0 management interface, four 2000W redundant Platinum power supplies with rails.