The funny thing is this thing actually happened. There was a chip called the Lucid Hydra 200 that allowed the user to combine a Nvidia card and an AMD card in multi-GPU configuration. They did it by intercepting the Direct3D calls and selectively delegating them to each GPU.

It goes without saying that it didn't work anywhere near satisfactory levels, but kudos on them for trying.
https://www-sweclockers-com.translate.goog/te … &_x_tr_pto=wapp

PS: Yes I know I am some years late.

kurkosdr wrote on 2023-02-03, 19:30:

The funny thing is this thing actually happened. There was a chip called the Lucid Hydra 200 that allowed the user to combine a […]
Show full quote

The funny thing is this thing actually happened. There was a chip called the Lucid Hydra 200 that allowed the user to combine a Nvidia card and an AMD card in multi-GPU configuration. They did it by intercepting the Direct3D calls and selectively delegating them to each GPU.

It goes without saying that it didn't work anywhere near satisfactory levels, but kudos on them for trying.
https://www-sweclockers-com.translate.goog/te … &_x_tr_pto=wapp

PS: Yes I know I am some years late.

This was meant to be an open source/platform agnostic approach to multi GPU rendering that had some buzz for a while but ultimately never went anywhere.

The graphics chips were not on the same PCB, they were separate GPUs that worked independently, and with the idea that they could also work parallel.

I think what killed it is two things

1) Multi GPU gaming just died
2) Multi GPU rendering and compute went down a different path with solutions that are either proprietary (like CUDA) or just run through virtualization or with parallel workloads and don't require any kind of multi GPU synchronization.

mothergoose729 wrote on 2023-02-03, 19:48:

This was meant to be an open source/platform agnostic approach to multi GPU rendering that had some buzz for a while but ultimat […]
Show full quote

kurkosdr wrote on 2023-02-03, 19:30:

The funny thing is this thing actually happened. There was a chip called the Lucid Hydra 200 that allowed the user to combine a […]
Show full quote

The funny thing is this thing actually happened. There was a chip called the Lucid Hydra 200 that allowed the user to combine a Nvidia card and an AMD card in multi-GPU configuration. They did it by intercepting the Direct3D calls and selectively delegating them to each GPU.

It goes without saying that it didn't work anywhere near satisfactory levels, but kudos on them for trying.
https://www-sweclockers-com.translate.goog/te … &_x_tr_pto=wapp

PS: Yes I know I am some years late.

This was meant to be an open source/platform agnostic approach to multi GPU rendering that had some buzz for a while but ultimately never went anywhere.

The graphics chips were not on the same PCB, they were separate GPUs that worked independently, and with the idea that they could also work parallel.

I think what killed it is two things

1) Multi GPU gaming just died
2) Multi GPU rendering and compute went down a different path with solutions that are either proprietary (like CUDA) or just run through virtualization or with parallel workloads and don't require any kind of multi GPU synchronization.

Even back when Multi-GPU gaming was a thing, nobody in their right mind would buy one GPU from AMD and another one from NVIDIA to run them in multi-GPU setup. The Lucid Hydra is one of these cases of engineers preoccupied with whether they could, they didn't stop to think if they should. But it's a real-world product that's basically BitchinFast 3D 2000 (save for the fact the chips aren't on the same PCB), and that brings a smile to my face.

Even in rendering scenarios, it makes sense to stick to a single vendor because there is not much benefit in not doing so (save for marginal cost savings) and many pitfalls if you don't.

Tetrium wrote on 2018-03-21, 20:49:

I still want a boxed one 🤣

I want one with the box .. then I can display both on the wall !!

kurkosdr wrote on 2023-02-03, 21:54:

mothergoose729 wrote on 2023-02-03, 19:48:

This was meant to be an open source/platform agnostic approach to multi GPU rendering that had some buzz for a while but ultimat […]
Show full quote

kurkosdr wrote on 2023-02-03, 19:30:

The funny thing is this thing actually happened. There was a chip called the Lucid Hydra 200 that allowed the user to combine a […]
Show full quote

The funny thing is this thing actually happened. There was a chip called the Lucid Hydra 200 that allowed the user to combine a Nvidia card and an AMD card in multi-GPU configuration. They did it by intercepting the Direct3D calls and selectively delegating them to each GPU.

It goes without saying that it didn't work anywhere near satisfactory levels, but kudos on them for trying.
https://www-sweclockers-com.translate.goog/te … &_x_tr_pto=wapp

PS: Yes I know I am some years late.

This was meant to be an open source/platform agnostic approach to multi GPU rendering that had some buzz for a while but ultimately never went anywhere.

The graphics chips were not on the same PCB, they were separate GPUs that worked independently, and with the idea that they could also work parallel.

I think what killed it is two things

1) Multi GPU gaming just died
2) Multi GPU rendering and compute went down a different path with solutions that are either proprietary (like CUDA) or just run through virtualization or with parallel workloads and don't require any kind of multi GPU synchronization.

Even back when Multi-GPU gaming was a thing, nobody in their right mind would buy one GPU from AMD and another one from NVIDIA to run them in multi-GPU setup. The Lucid Hydra is one of these cases of engineers preoccupied with whether they could, they didn't stop to think if they should. But it's a real-world product that's basically BitchinFast 3D 2000 (save for the fact the chips aren't on the same PCB), and that brings a smile to my face.

Even in rendering scenarios, it makes sense to stick to a single vendor because there is not much benefit in not doing so (save for marginal cost savings) and many pitfalls if you don't.

Hardware nerds were definitely interested in it. There was this abstract potential that you could keep your old GPU as you upgrades and just add a new GPU to your computer utilizing the power of both. I was one such nerd between maybe 2011-2014. This would come up maybe every six months on web forums like OCN and anandtech.

There was one partiular combo of drivers for ATI and nVidia that let you have both in your system without argument and to then use OpenCL on both. Or just CUDA on the nVid, something else on the AMD/ATi.. Anyhoo, it was with that lashup I found I could use a 9400GT PCI for PhysX and game on a HD4650. didn't try it on much tho.

However, though it has not been greatly leveraged yet, I see it could be possible with Vulkan to have dissimilar cards just doing as much Vulkanning as each can. CPU too if it's got gfx.