When comparing the performance of the 5090 8 card machine and the 4090 8 card machine, the following key points need to be considered:

1. Architecture and process
5090: Expected to use next-generation architectures (e.g., NVIDIA's Blackwell or Hopper) with more advanced processes (e.g., 3nm or 5nm) for higher energy efficiency and performance.

4090: Based on the Ampere architecture, using 7nm or 5nm process, although the performance is strong, but not as good as the next-generation architecture.

2. CUDA Cores vs. Tensor Cores
5090: The number of CUDA cores and Tensor cores is increased, and the performance of AI and deep learning is significantly improved.
4090: Low performance due to a small number of cores.
3. Video memory
5090: Expected to be equipped with GDDR6X or GDDR7 memory with higher capacity and faster bandwidth to increase big data processing capabilities.
4090: The memory capacity and bandwidth are low, and the processing of big data may be limited.
4. Power consumption and heat dissipation
5090: Despite the higher performance, the power consumption may be lower and the cooling requirements will be higher thanks to a more advanced process.
4090: High power consumption and high heat dissipation requirements.
5. Actual performance
5090: In tasks such as deep learning, AI training, and graphics rendering, the performance is expected to be improved by 30%-50% compared with the 4090.
4090: Although the performance is strong, it is not as good as the 5090.
6. Price
5090: Expected to be more expensive, for those on a budget who are looking for extreme performance.
4090: More cost-effective for those who have a limited budget but still need high performance.
summary
5090 8 Card Machine: Stronger performance, suitable for high-end needs, but more expensive.
4090 8 Card Machine: More cost-effective, suitable for users who are on a budget but still need high performance.
The choice is based on a trade-off between budget and performance needs.