Ram: Why is Dual channel only 10% faster than single channel if throughput is double?

[JEDI]

With dual channel, you're doubling the xfer speed from memory to CPU and vice cersa.

so why do benchmarks say only a 10% increase (at best)?

 

[themisfit610]

Because for most desktop tasks, memory bandwidth is not a significant bottleneck 🙂

Derek

 

[Mwing]

bottleneck should be hdd on most system right?

so if it is ssd, dual channel ram should have bigger impact?

 

[taltamir]

bottleneck should be hdd on most system right?

so if it is ssd, dual channel ram should have bigger impact?

bottle neckS INCLUDE the HDD.
they also include many other components...

 

[taltamir]

bottleneck should be hdd on most system right?

so if it is ssd, dual channel ram should have bigger impact?

bottle neckS INCLUDE the HDD.
they also include many other components...

Besides, there is absolutely no reason for 2x increase in ram speed to provide a 2x increase to performance....

Lets say that to perform a task you need to spend 30ms performing calculations on the CPU and 20ms getting data from ram. Total time to completion is then 50ms.
Now, double the ram speed and your task takes 30ms of CPU calculations and 10ms of ram data access... 40ms total. aka 20% speed increase. You have literally cut in half the time needed for the ram to be accessed, yet it didn't cut in half your time to complete the task, this is normal and sensible...

This is actually why I don't advocate SSDs for people wanting faster boot time, I find that the SSD has a fairly minor effect on it, since most of it is spent waiting on the mobo to initialize the bios, etc.
SSDs are great though, they vastly improve install times of programs and games, and vastly improve my loading times in games (I want to play, not watch a "loading" screen)

 

[Dufus]

Here's some RAM with a memory clock of 400MHz and theoretical dual channel bandwidth of 12.8GB/s. The bandwidth gain over single channel is ~85%.

Just because the RAM's theoretical bandwidth is double doesn't necessarily mean the memory controller, CPU and interface can support it.

 

[IntelUser2000]

With dual channel, you're doubling the xfer speed from memory to CPU and vice cersa.

so why do benchmarks say only a 10% increase (at best)?

Simple. 2x increase in performance with 2x bandwidth would mean the design is flawed because its 100% bandwidth bound.

Think of it this way. 2x faster car won't net you 2x faster speed because in regular roads, you'll be caught by the traffic lights. I see lots of times where a slower moving car ends up going just as fast as the speeding car because the traffic lights prevent the faster one from going at its full potential.

Compare that to jogging at 2x speed of walking. Jogging would surely result in arriving nearly 2x faster because a block is vast to a person jogging/walking.

 

[sub.mesa]

Check some real-world benchmarks and you will see the cheaper slowest memory is only a few percent slower than the most expensive fastest memory available; for real world applications the memory throughput and latencies are not a bottleneck at all.

Thus, it's better to pick more but slower memory than less but faster memory.