some AI coprocessors, you’re right. they can’t be used.
Most of what nVidia is offering, while lacking display outputs, can still be used for gaming/VDI/vGPU use, and in general aren’t particularly bad at it, usually quite the opposite. might take a bit of work to get them going, and nVidia hasn’t made it easy with their whole licensing thing. Putting aside the challenges, it’s usually possible to adapt their SXM datacenter GPU modules to work with more common frameworks like PCIe, and then utilize those as virtual GPUs in a virtual desktop. This is not dissimilar to what they’re doing with the geforce now platform.
The big benefit of the chips is that they come with loads of memory, so cutting them up as a vGPU is actually a really optimal use-case. Most games are memory limited, which is to say they need more memory than most GPUs have, so the limit in performance is from the memory limitations, not the graphics processing limitations.
The AI processors that are GPU based, like almost everything that nvidia is using, can mostly be repurposed, but those that are ML-only, like AI coprocessors that show up in CPUs like the AI optimized CPUs from AMD and Intel, can’t really be re-used, but they’re usually not released as a standalone product. Those may still be able to perform physics processing offloads… so maybe not useless? I’m not sure.
Not strictly, there are usualy hurdles to overcome for home usage of datacentre tech, but it’s possible.
An easy example that comes to mind is that the datacentre GPU’s might have cooling configurations that rely on a large amount of airflow from the surrounding cabinet, which you’ll get in a datacentre, but not so much in a personal tower.
Also, noise, in a datacentre, noise is much less of an issue than in a personal computer, so they can eschew noise reduction in favour of performance.
Many people that live near data-centres would disagree, but i mean strictly from a usage point of view, not a broader community perspective.
Not strictly, there are usualy hurdles to overcome for home usage of datacentre tech, but it’s possible.
The hurdles are basically insurmountable with the hardware released after 2024.
The NVL72 for the Blackwell generation cost about $3 million and takes up a single server rack. The power consumption is about 130 kW, and most configurations require dedicated plumbing for the liquid cooling.
To put things in perspective, a residential electrical hookup is usually 50A or 100A for a house, with recommendations that anyone who is going to be charging electric cars should have 100A service. 100A at 240V is 24 kW.
So one server rack uses as much power as the maximum electrical capacity of 5 homes. You’ll never be able to pull that off in an actual residential environment.
Oh, and the newest 2026 generation, the Rubin NVL72s, use something like 230 kW of electrical power, almost twice as much as the previous 2024 generation.
AI chips arnt usable for personal use though, they seem to be limited to be used by the AI datancenters and not your computers
some AI coprocessors, you’re right. they can’t be used.
Most of what nVidia is offering, while lacking display outputs, can still be used for gaming/VDI/vGPU use, and in general aren’t particularly bad at it, usually quite the opposite. might take a bit of work to get them going, and nVidia hasn’t made it easy with their whole licensing thing. Putting aside the challenges, it’s usually possible to adapt their SXM datacenter GPU modules to work with more common frameworks like PCIe, and then utilize those as virtual GPUs in a virtual desktop. This is not dissimilar to what they’re doing with the geforce now platform.
The big benefit of the chips is that they come with loads of memory, so cutting them up as a vGPU is actually a really optimal use-case. Most games are memory limited, which is to say they need more memory than most GPUs have, so the limit in performance is from the memory limitations, not the graphics processing limitations.
The AI processors that are GPU based, like almost everything that nvidia is using, can mostly be repurposed, but those that are ML-only, like AI coprocessors that show up in CPUs like the AI optimized CPUs from AMD and Intel, can’t really be re-used, but they’re usually not released as a standalone product. Those may still be able to perform physics processing offloads… so maybe not useless? I’m not sure.
Not strictly, there are usualy hurdles to overcome for home usage of datacentre tech, but it’s possible.
An easy example that comes to mind is that the datacentre GPU’s might have cooling configurations that rely on a large amount of airflow from the surrounding cabinet, which you’ll get in a datacentre, but not so much in a personal tower.
Also, noise, in a datacentre, noise is much less of an issue than in a personal computer, so they can eschew noise reduction in favour of performance.
Many people that live near data-centres would disagree, but i mean strictly from a usage point of view, not a broader community perspective.
The hurdles are basically insurmountable with the hardware released after 2024.
The NVL72 for the Blackwell generation cost about $3 million and takes up a single server rack. The power consumption is about 130 kW, and most configurations require dedicated plumbing for the liquid cooling.
To put things in perspective, a residential electrical hookup is usually 50A or 100A for a house, with recommendations that anyone who is going to be charging electric cars should have 100A service. 100A at 240V is 24 kW.
So one server rack uses as much power as the maximum electrical capacity of 5 homes. You’ll never be able to pull that off in an actual residential environment.
Oh, and the newest 2026 generation, the Rubin NVL72s, use something like 230 kW of electrical power, almost twice as much as the previous 2024 generation.
i was talking more of the prosumer and close to prosumer units that could theoretically be retrofitted or housed in homelabs.
If datacentres went down en-masse i’d assume there’d be mix of hardware generations there.
Damn though, i didn’t know anything about the stuff you just mentioned, so my estimates of what could be available/usable are probably way off.
Now i have to go and do a deep dive, thanks.