Talk:SGI O2

From Wikipedia, the free encyclopedia

[edit] UMA = SMA?

I have just noticed that UMA is linked to Shared Memory Architecture. This in incorrect in my view, as the SGI implementation UMA is very different from SMA:

• Firstly, SMA is intended to cut costs. It could be argued that UMA has the same goal, but the main difference with UMA is that it aims to be better than SMA while being cheaper than systems that use the higher end crossbar architecture.

• UMA also intends to reduce bottlenecks by connecting major components to the memory controller with dedicated links, and these themselves are high bandwidth - for example, the link from the memory controller to the GDE consists of dual 64 bit 133 MHz buses, not exactly cheap (for the time).

• Another difference in that unlike SMA, I do not believe UMA poses any restriction on the amount of RAM available to the system, the memory the GPU uses is still the main memory, still accessible. What memory the GPU uses is treated as used system RAM, and is fully dynamic.

I'm not too familiar with UMA or the O2, but I think that this requires some discussion as to whether UMA is the same as SMA. If it isn't I think the link should be removed. Rilak (talk) 07:25, 9 January 2008 (UTC)