I remember reading about the GDDR versus DDR series a few years ago, and the most significant improvement of GDDR over DDR was:
1. You can activate a memory bank and issue a read request in the same clock cycle. In DDR they are each lagged by 1 clock because there is a collision.
2. In GDDR you can write and read the same memory location at the same time. This isn’t as useful anymore because GPU cache is much improved over the ‘old days’ and you ‘shouldn’t’ have to read the same location you are writing.
I tried doing a quick google search and I could only confirm #1, so #2 might not be correct but I couldn’t find anything saying #2 isn’t possible.
ar 2. punktu viss skaidrs, bet īsti nesaprotu pirmā punkta jēgu: “you can activate this feature”, tātad tas nozīmē, ka atsevišķi jāaktivizē šī iespēja? par ko šī memory bank atbild?
lasot tālāk, ir atrodams skaidrojums:
DDR can make one read or one write operation at one time.
GDDR can make both.
For example: While getting the data from computer’s main memory for the next frame, GPU can access the memory and do the operations about the current frame.
If it was standard DDR, either GPU or PCI-Express should wait in order to get exclusive access to framebuffer memory.
This way with the same speed GDDR memory can act 2 times faster.
bet,
[Edit: Sorry guys, misinformation. It was VRAM. Not GDDR…]
.
