Quad Channel vs Dual

[trents]

I just found this awesome article. It confirms my 2400Mhz recommendation.

http://www.anandtech.com/show/7364/memory-scaling-on-haswell/10

His conclusions are based on results from a single game: DIRT 3. I think we need a broader spectrum of testing than that in order to confirm what he is saying. Besides, his conclusions are at odds with what a lot of us have experienced in our own experimentation with higher memory speeds.

 

[ATMINSIDE]

I just found this awesome article. It confirms my 2400Mhz recommendation.

http://www.anandtech.com/show/7364/memory-scaling-on-haswell/10

To pull an excerpt from this:

For discrete GPU users, recommending any kit over another is a tough call. In light of daily workloads, a good DDR3-1866 C9 MHz kit will hit the curve on the right spot to remain cost effective. Users with a few extra dollars in their back pocket might look towards 2133 C9/2400 C10, which moves a little up the curve and has the potential should a game come out that is heavily memory dependent.

This falls in line with the majority of our recommendations from the regulars here at OCF.
We will almost always recommend staying in an 1866 or 2133 MHz CL9 kit.

 

[Woomack]

Anand results are good only for single examples on stock CPU/memory controller clock. In real most enthusiasts ( count like 3% users ) are overclocking everything. You can't say that one memory is faster than any other without comparing memory controller speed or additional factors like memory density. Anand was comparing double sided kits and single sided kits in one article without telling which is which. It's just lack of knowledge which is later being repeated on various forums. Actually most memory tests/reviews in the web are pretty bad.

On DDR3 any double sided memory kit 2133 - 2600 will beat every available single sided kit , including 3300+. Also new cpus like haswell at higher clock will perform better so when you compare memory, you have to use the same CPU/cache clock.
On DDR4 memory latency doesn't affect general performance so much. Internal bus is wider and cpu cache is fast/large what helps a lot. In this case the biggest factor is memory controller/cache speed. If you raise cache clock on X99 platforms to about 4GHz then you barely see any difference between 2666 and 3200+ kits. DDR4 has no performance issues related with memory density. In this case more cores = higher bandwidth so 5820K will have lower max bandwidth than the 5960X.

One more thing is memory controller scalling - read it as memory can run at higher clock but performance is not raising much. AMD are barely scalling above ~2000, haswell above ~2600, haswell-e above ~3000. Still we see much higher clocked memory available in stores.