• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Failure Analysis on GPUs

1 transistor out of 9 billion.. Just imagine that. That is mind blowing stuff especially for those who work in the electronics field. Compare it to your everyday circuit debugging or doing failure analysis on circuits!
 
Just crazy to think a single 980ti has more transistors than people on the planet and they can find the single transistor that's borked..
 
Wonder when we will get GPUs that treat the PCB as a motherboard. They come pre-fitted at the factory but you can buy a replacement PCB for the GPU. Should be easier now with HBM making PCBs more for power. seems a waste to call the whole package dead when the GPU is probably still good, given how much more reliable it apparently is. Could open up a whole new market of GPU motherboards with features, better power etc.
 
The 1st video wasn't really impressive to me. Thought it was going to at least show some actual equipment use.

Will watch 2nd one tonight.
 
Wonder when we will get GPUs that treat the PCB as a motherboard.
I'd speculate; never. Different GPUs have different power requirements, substrate sizes (IE, number of connections) and so on. The one-size-fits-all would be the biggest size for everyone, and mid/low-end GPUs don't need 300+ watt voltage regulators. It's just increased expense for no benefit.

If GPUs are to be replaceable, you'd reasonably need to cap them, for the sake of durability. This adds thermal resistance, vertical height, and cost. Sockets also add vertical height to the board, making it more difficult to fit a decent cooler.

You'd also get locked into a socket layout for an extended period, dampening innovation, or else facing rapid obsolescence. Like with complications like NVLink for example, you can't really foresee something like that and pre-allocate connections for it, in case you need it. When you do have a need, you have to redesign the socket anyhow, so what's the point of even making replaceable GPU boards when they have no chance of lasting any significant time on the market?

Also, socket connections tend to be detrimental for high-speed electrical signals, it might not be feasible.
 
Back
Top