Pooling those RTX cards together would definately be a game changer. Running distributed inferance across 96GB of separat cards is such a headache compared to having it all unified. The 6000 Pro would simplify everythng.
Those GPUs are on more than one continent and they have regular uses from their owners, so that's a hard *nope*. I agree on the RTX 6000 Pro, but we need to get to a point where not having one is a hard stop before I can ask for one. I suspect one of the 128GB unified systems - either DGX Spark or the much cheaper AMD thing with the long confusing name, would be the choice. We could get four of the GMKtek machines for the price of one big GPU, that would free all of us to move in parallel ...
Pooling those RTX cards together would definately be a game changer. Running distributed inferance across 96GB of separat cards is such a headache compared to having it all unified. The 6000 Pro would simplify everythng.
Those GPUs are on more than one continent and they have regular uses from their owners, so that's a hard *nope*. I agree on the RTX 6000 Pro, but we need to get to a point where not having one is a hard stop before I can ask for one. I suspect one of the 128GB unified systems - either DGX Spark or the much cheaper AMD thing with the long confusing name, would be the choice. We could get four of the GMKtek machines for the price of one big GPU, that would free all of us to move in parallel ...