AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |
Back to Blog
![]() ![]() RAM is still behind PCIe with both bandwidth and latency penalty on access when compared to VRAM. If a page is allocated but not used, it will be moved to RAM to free up space in VRAM for other pages that may be more used. There is an algorithm in HBCC to determine whether page in RAM is allocated and used or allocated and not used. It does not tell a game anything and as far as we know does not accept guidance from the game either. In desktop/consumer space HBCC can and will use RAM as the larger memory space and VRAM as a cache. What HBCC does not solve is latency and bandwidth which are important for real-time applications like games. This has quite nice implications for compute and some workstation applications where data sets are tens or hundreds of GBs. It treats VRAM as the last-level cache and allows having the larger memory set somewhere else - system memory, Flash drive (like on Radeon Pro SSG) or even networked storage. Then there is some renaming like VRAM to HBC. Primary technical change is enlarged address space - 49-bit if my memory serves right - plus the HBC controller itself to manage the data flow. ![]() What HBCC does is rework how Vega treats and manages memory. Click to expand.Do you have a link to the whitepaper? ![]()
0 Comments
Read More
Leave a Reply. |