Great idea! Then the video chip can saturate the bus talking to ram that's 5x slower than it needs to operate all the while choking bandwidth away from the general processor.
I can't wait for this next step!
:lol:
Quite right
Torenzza in 4x4 is a just a lot of aditional marketing and a lot of expencive hardware connected to work theoreticly but not practicly to improve general system performance in any reasonable and noticable amount.
Let's see what HTT3 can do for Torenzza?
Four links, each fullduplex 32bit 2.6GHz providing 10.4GB/s in each direction, 20.8GB/s total.
One link used for connecting the 2 CPUs, two for connecting to the two PPUs(1 link per PPU) and one for connecting to a chipset.
2P K8, 2 IMCs each handling (lets guess) 128bit DDR2-SLI-1066 or 17GB/s. In 2P K8 with NUMA, each CPU will benefit only 10.4GB/s memory bandwidth. It is becouse the limit of HTT3, 10.4GB/s total bandwidth per each direction per link.
Processors used in modern single GPU graphics cards today are handling more than 50GB/s. In the time when Torenzza 4x4 will be available for us the GPU/PPU cards will be dual processor handling more bandwidth each.
Each time they will request data from system RAM, the CPU involved in memory tranffic will suffer for RAM bandiwidth.
So, we can not expect from them to boost the CPU performance in any noticable way.