GPU IMCs have the capability to operate at a 1:1 ratio with GDDR5, GDDR6, and GDDR7 VRAM.
GPU IMCs have the capability to operate at a 1:1 ratio with GDDR5, GDDR6, and GDDR7 VRAM.
Darkbreeze's interesting write-up on DDR5 memory and CPU IMCs got me thinking about GPU IMCs. Can GPU IMCs run at a 1:1 ratio w/GDDR memory? It would seem impossible considering the quad-pumped MT/s speeds of even GDDR5 memory. If the GPU IMC is running much slower than the GDDR5, GDDR6, GDDR7 MT/s does that mean the memory is waiting for the GPU IMC most of the time (which I suppose is better than the inverse)?
This claim is unrealistic, as graphics card memory controllers typically operate at a dynamic rate rather than a fixed ratio. CPUs are designed for fast, sequential handling of intricate and diverse tasks, while graphics cards excel at high-speed parallel processing of straightforward, repetitive operations. They can manage complex sequential work but are not built for it.
I understand your concern. Based on the load, GPU IMCs and VRAM appear to throttle. However, if VRAM operates at its full speed, could the GPU IMC run at a significantly reduced frequency, potentially even below gear 4 in Intel IMCs?