Would this work for GPUs?

[Computernaut]

CPUs come as little squares the fit into a motherboard socket, what I'm wondering is, could this work for GPUs? Not only would this not require graphics cards to come on their own "motherboard" but the cooler could be much larger and not be confined to a flat slab on top of the card, this could also allow the card to have more contact points.

[TheGlenlivet]

2 minutes ago, Computernaut said:

CPUs come as little squares the fit into a motherboard socket, what I'm wondering is, could this work for GPUs? Not only would this not require graphics cards to come on their own "motherboard" but the cooler could be much larger and not be confined to a flat slab on top of the card, this could also allow the card to have more contact points.

The RAM and power Systems are all soldered to the card with the GPU so swapping the GPU ,if that's the goal here, wouldn't get you as much of an increase in performance as changing CPUs on a motherboard.

If better cooling is the goal, there are aftermarket cooling systems for most current GPUs.

[DocSwag]

Theoretically yes. A socket would take up a lot more space though making the GPU significantly thicker, plus you might have issues with VRAM. Then you would need removable VRAM, which would take up even more space and cost more money., as well as make cooling way more difficult.

[sof006]

I would say good idea on paper and it's sort of been done in the past. Motherboards used to have integrated graphics built into them but it got replaced by the GPU being built into the CPU itself.

 

Onboard graphics tends to be very underpowered. 

 

Graphics cards on the other hand are powerful because the manufacturer makes their own PCB.

[Computernaut]

22 hours ago, DocSwag said:

Theoretically yes. A socket would take up a lot more space though making the GPU significantly thicker, plus you might have issues with VRAM. Then you would need removable VRAM, which would take up even more space and cost more money., as well as make cooling way more difficult.

It wouldn't be on it's own card it would be part of the motherboard, it might look something like a server motherboard with two CPUs. I very much like the idea of removable VRAM.

[DocSwag]

3 minutes ago, Computernaut said:

It wouldn't be on it's own card it would be part of the motherboard, it might look something like a server motherboard with two CPUs. I very much like the idea of removable VRAM.

Not sure I completely see what you're saying... VRAM takes up a lot of space so it's not as simple as socketing it. Plus different GPUs have different size busses requiring different amounts of VRAM chips

[Computernaut]

1 hour ago, DocSwag said:

Not sure I completely see what you're saying... VRAM takes up a lot of space so it's not as simple as socketing it. Plus different GPUs have different size busses requiring different amounts of VRAM chips

CPUs were probably the same once upon a time. What I'm proposing is a standardization on GPU hardware.  CPU's were made "modular" for compatibility, as were RAM and others, I'm surprised this hasn't happened to GPUs. As graphics cards become more powerful, the will -- one way or another -- become bigger, they will also generate more heat and therefore require bigger coolers.

[DocSwag]

14 hours ago, Computernaut said:

CPUs were probably the same once upon a time. What I'm proposing is a standardization on GPU hardware.  CPU's were made "modular" for compatibility, as were RAM and others, I'm surprised this hasn't happened to GPUs. As graphics cards become more powerful, the will -- one way or another -- become bigger, they will also generate more heat and therefore require bigger coolers.

The problem is, manufacturers will never implement it. You're looking at putting R&D into making it modular, which costs money, for what gains?

 

Plus, there's tons of standardization issues. Different gpus often have different size busses, as I mentioned before, requiring differing amounts of vram chips unless you want to disable a bunch of vram at a huge performance penalty. Cpus didn't have this issue since most are just 2 channels of 64 bit memory busses, and there isn't much difference except at the top. Gpus being more memory bandwidth hungry causes this issue.

 

As well, gpu vram is way more power hungry than dram ever has been, which is a result of gpu's need for memory bandwidth. So you need coolers that also can cool this modular vram, which is kinda difficult... Sort of see where I'm going?

 

So with a gpu upgrade, you're also gonna need to upgrade the vram, and since you're upgrading the vram, you need a different sized cooler to cool it too, so now you've just upgraded every part of the gpu except the pcb to get an upgrade. I'm pretty sure the costs of that would come close to those of just buying a whole new gpu, what with the costs of the modular stuff.

 

Maybe one day we will see modular gpus, but I just don't see it happening right now. One day, cpus and gpus are supposed to approach each other and become the same thing (I think there's like a whole theory on this), but that day isn't today and right now cpus and gpus require very different things, resulting in these differences.

[Computernaut]

Oh, okay.