[QUOTE="The Turbanator, post: 55778473, member: 1101"Thank you for the explanation. I guess it's a matter of pricing then? Or is the latency question different? I ask because I thought the price of VRAM is what makes GDDR not suitable for use as DRAM, but with a sub-$350 GPU expected to launch with 12GB of GDDR6, I have to question why it wouldn't just be used as general RAM in your average system.[/QUOTE]
Single-pool is rarely used in PC architectures because the performance penalty (yes, latency) is huge communicating from main memory to the GPU. On consoles this problem is mitigated because the parts (motherboard, RAM, CPU, GPU) are all on the same board, and usually the CPU/GPU are merged into an APU, closely surrounded on all sides by the RAM modules. The literal length of the traces matters here, compared to a PC where the GPU would have to make a transfer over the PCI express bus and just basically travel way, way farther.
An example of single-pool still used is Intel integrated graphics on laptops, where they typically reserve some main memory. These GPUs are...extremely slow.
As to why GDDR is not used as system memory? Simply because the bandwidth doesn't add anything to general computational use. Traditional DDR is cheaper but still provides tons of bandwith for CPU operations, and it doesn't have to always share that bandwidth directly with the GPU. This saves motherboards from having to be designed with more signal lines (DDR4: 64bit, GDDR6X 3080: 320bits), which would crowd out a lot of the space on the average ATX motherboard and possibly require RAM to be placed closer to the CPU (causing problems for heatsink designs).