Reply 20 of 28, by CharlieFoxtrot
- Rank
- Member
shevalier wrote on 2023-10-18, 07:45:Dual channel makes sense for any parallel process that accesses memory past the CPU, for example loading textures into video mem […]
CharlieFoxtrot wrote on 2023-10-18, 04:21:That dual channel memory controller is completely useless on these, unless you use the GF4MX integrated graphics available on some boards. The whole idea of the solution was to reserve the other channel for IGP.
Dual channel makes sense for any parallel process that accesses memory past the CPU, for example loading textures into video memory during intensive exchange with the memory of the CPU itself.
Or copying from the network to disk under similar conditions.
The difference is not +100, but a modest +10-15.
Plus prefetch in the northbridge (sometimes called L3 cache). Another 10 percent productivity.
At its peak, the difference between nF2 and KT400/600 reached 30 percent. Sometimes even 0, but on average - around 15 percent.
But in those days this was the difference (relatively speaking) between Athlone 2200+ and 2500+.
Which also cost money.
Most of the nForce2 edge over KT400/600 comes from the other factors of the chipset, not from the ability to use dual channel, unless you use IGP.
Here is an old test about the nForce dual channel vs KT400 and performance difference for most tests is low single digits. Where the difference is bigger, it is explained in the article. I'd say 15% is quite a stretch in real world scenarios, especially gaming.
http://ixbtlabs.com/articles2/nforce2-1vs2cha … nels/index.html
The problem with Athlon XP was that @200 is FSB output 3,2GB/s while dual channel 1:1 provides double the memory bandwidth. That is also the reason why IGP gets by far the biggest benefit from it as CPU can't saturate the 6,4GB/s maximum effectively that is available to it.