As I understand it, PCIe 4 has double the throughput of PCIe 3. Thus, PCIe 3 x 16 has the same throughput as PCIe 4 x 8, correct?
Correct.
A RX 6600 with its 8 lanes in my PCIe 3 system would have the equivalent of just 4 lanes throughput, right?
Careful here. You still get 8 lanes of PCIe 3.0 if the card has 8 PCIe 4.0 lanes.
The RX6600 will run at 8x PCIe 3.0 in your board, and will get around ~7.8 GB/s.
Should I rather invest in the RX 6700, which seems to be the cheapest PCIe 4 x 16 graphics card?
That depends on the games you play.
The PCIe bandwidth is not directly scaling how much FPS a GPU can achieve. In some games you will find that running on 4x lanes has a negligible impact while other games will use as much bandwidth as they can get.
This article from TechSpot shows some benchmarks of the same GPU used on different PCIe bus configurations for a variety of games. For most of the games they tested the performance drop going from 16x PCIe 4.0 to 8x PCIe 3.0 was around 5%. For Doom Eternal it was a significant 18% performance drop. So there isn't a general answer and the games you want to play should factor into your decision.
My opinion is that for most games this drop in bandwidth is not going to be enough to cause buyer's remorse.
If you're wondering about a game that wasn't benchmarked you can sort of judge whether it will lose FPS with less bandwidth based on some things about the way the game works and its engine.
- Games that stream models and textures in big open worlds tend to need high bandwidth, uploading data to the GPU takes time and can cause stuttering. Doom Eternal is an exceptional example of this because the engine very aggressively uses streaming instead of keeping unused data in memory.
- Games with loading screens for every area you go in to are doing all of their GPU uploads during the loading screen, so nothing needs to happen as you are playing. A game like Overwatch is a good example of this.