Its worth saying a GPU dock is just an eGPU. Often they use mobile GPUs and will make something more streamlined or portable. They often include a USB hub.

Most eGPU chassis are the same tech but often larger so you can install a larger desktop gpu, and have more space for cooling and even a dedicated PSU.

GPU docks have their own usefulness but you are likely to get more performance for the price by getting your own chassis and card, but less mobility and more bulky.

There are plenty of chassis available if you go looking. Some haven’t been updated to newer models but remain available.

A chassis is fairly simple - its basically a bit of mother board with pcie and a thunderbolt 3+ connection. Thunderbolt 3 remains powerful enough for most uses, and ones with dedicated PSU will work with newer cards. I think the lack of new products reflects the lack of needing to change the products at the moment.

I haven’t seen much about thunderbolt 4 reducing the overhead. It might do, but there are fundamental constraints in these devices as this is basically converting pcie to thunderbolt and transferring over a distance - thats not going to ever match a direct pcie connection into a motherboard no matter how fast Thunderbolt 4 or 5 are. Thunderbolt 3 may not be the main bottleneck.

The Razer Core X is still available for example. And there are loads of smaller companies woth offerings.

I think just the highest end cards would be out of reach for the popular existing chassis but there is not going to be much market for pairing cards costing 1000s with a laptop when you are far better getting a desktop. So there may not be the market to make lots of new thunderbolt 4 chassis with PSUs.

The most important consideration is your laptops ports and it’s cpu. You will need Thunderbolt 3, 4 or 5 or USB 4 to get high enough transfer speeds and bandwidth between your eGPU and the laptop. You also need a decent CPU to get the full benefits - an eGPU paired with an old or low powered CPU may mean you dont get the full benefits of the eGPU as your CPU is still a bottleneck in running the software or games that would make use of the eGPU.

Then the eGPU chassis you choose will have specific limitations in terms of size of card that will fit. You need to check these carefully to ensure the chassis can fit and support the card you want. The bigger and better the chassis the more expensive it will be. Were talking a couple of hundred pounds / dollars on top of the card price.

But in theory there isn’t a limit on the cards you can use. Any GPUs that fits the chassis would work as its a standard pcie slot. However i would contend that if you want to use a top end card like a 5090 youre better off getting an actual PC to enjoy the full performance. If youre spending 1000s on a GPU it should be paired with a high end laptop or far better in an actual desktop to get the benefit. You also need to ensure the chassis can provide enough power to the card you want.

You lose about 10-15% of the cards functionality in the overhead of the eGPU. Thats because as fast as thunderbolt and usb4 are, you are transfering that to/from a pcie slot in the eGPU chassis and also transferring data over an distance via a cable compared to a gpu plugged directly into pcie on a motherboard for a PC, with direct connection to the CPU and rest of the motherboard. Newer thunderbolt and newer chassis might have lower overheads but they will never be able to completely match direct plug into a motherboard.

So yes eGPUs work, if your device can support it, and you can get big performance boosts. There isn’t a limit on the GPU but you should probably not go too high end as you’d be wasting money. A low end GPU would likely out perform any integrated card or graphics for most laptops and a mid range card would likely give excellent performance if paired with a decent specced laptop. But any eGPU set up cannot match the Max performance of the card in a dedicated desktop set up.

Edit: I know you have a surface but in case others read this and have a Mac - eGPUs wont work with Apples M1/M2 CPU chips. There is no way around this. AMD and Intel chips do although newer is better.