uff, writing a GPU driver yourself? Are you sure?
I'd recommend asking yourself: How many device drivers for complex DMA devices with multiple high-rate transports have you written so far? I ask because you might be significantly underestimating the complexity of this.
In fact, writing GPUs and GPU drivers is so hard, ST bought the GPU core and the right to ship their drivers from Vivante.
Maybe you can figure out whether GPU drivers for other OSes than Linux exist from there (quite possible – QNX is pretty popular with people who use embedded processors to build e.g. car displays). But: almost all embedded GPU vendors (and most desktop CPU vendors, too) have but a small open-source kernel shim that does a bit of the low-level "data transfer" logic. The actual graphics card driver lives in closed-source binary objects that resist reverse engineering pretty well. The GPU vendors have interest in not sharing the way their GPUs can be used publicly, because that gives insight into their architectural choices. So, everyone uses the OpenGL/Vulcan/DirectX/... abstractions that the userland drivers offer.
Also: Essentially, the usage of a GPU like this suggests you need a general purpose operating system. The data you'll be sending to the GPU will be quite large, so it's likely you need a file system or at least a very capable networking subsystem; these things use memory in ways that can effectively only be leveraged if you have a system that allows your software to deal with virtual memory, and you need a complex system to queue data transfers and handle interrupts.
Really, not using Linux or a similarly capable OS is probably a non-starter if you want to use that GPU.
You might not want that GPU, though. Maybe you need something that is way less powerful, and it doesn't matter to you that it uses a lot more electrical power to do less? Then, there's quite a few simple devices like SPI-attached framebuffers, but honestly, all these have been superseeded by FPGAs implementing simpler GPUs.
So, I think there's a large amount of requirements engineering that you don't tell us about: you're asking about where to get GPUs from for which you can write drivers, but you don't tell us what they need to do.
