Ah, FEA. I'm assuming you are doing the FEA through Solidworks Simulation? Unfortunately, I think Solidworks still does the FEA heavy-lifting on the CPU side, only using the GPU for visualising. As this means you can be CPU-limited, you would need to ensure you have the best performing CPU you can fit into your budget AND as much RAM as you can squeeze in, too.
As you are running FEA at max. tessellation, this can massively increase the computational cost. The bottom line is more tessellation, higher granularity, more interacting parts in the simulation. If your mesh size increases by a factor of n in (x, y, z) then even sparse matrix solvers will increase in complexity between O(n x n) and O(n x n x n) typically - depends on the solver and mesh quality. It's not just the number of elements that causes the performance hit, it's how those increased number of elements interact with each other and that can easily blow-up the solver calculation time. If you can use a different FEA with GPU solver support then there are some substantial gains to be had.
Gaming class GPUs (like the 4070Ti) are not really recommended for FEA and engineering simulation where accuracy and solver performance is essential. For starters, they are designed for peak performance rather than longevity and accuracy. Also, gaming GPUs have terrible FP64 performance compared to a workstation class GPU. The 4070Ti has ~0.05TFLOPS of FP64 performance. The RTX A4000 has ~0.39TFLOPS (almost 8x more FP64 performance). Games don't need the accuracy and are designed to be run at FP32 precision. FEA, CAD and engineering require FP64 if accuracy is important to you.
Additionally, you pay a bit extra for workstation cards as you are paying for certified cards and drivers that have had extra testing against industry standard software. The workstation GPUs also have improved performance for things like line drawing, virtualisation, layering, and are designed to run 24/7 without error. They also have ECC which means the memory can self-correct 1-bit memory errors and ensure data integrity and accuracy. They are designed for peak fault tolerance and stability as opposed to peak performance (like gaming GPUs).