Northbridge speed and low RAM timing significance on AMD platforms

[xd_1771 [OCN]]

Feel free to discuss here or you may join me at the original thread/discussion on Overclock.net

The rules/details of this article:
-Timings change between 6-8-6 and 9-9-9
-Northbridge speed changes between 2700Mhz and 3000Mhz
-Memory speed is constant 1600Mhz, the most optimal RAM frequency for AMD. This is not a RAM speed vs. timings comparison. There are no variances in this to keep everything fair.
-Everything else (bus clock, CPU clock, etc.) is also constant.
-The program used to test is "MaxxMem".

So, 20 minutes later:
1600Mhz 6-8-6, 2700Mhz northbridge:

1600Mhz 6-8-6, 3000Mhz northbridge:

1600Mhz 9-9-9, 2700Mhz northbridge:

1600Mhz 9-9-9, 3000Mhz northbridge:

1. Does a higher NB provide a benefit?
The answer is yes. You can see in both cases there's about a 500-600MB/s improvement when considering NB speed. Increasing NB speed is one of the ways to provide the biggest performance increases on AMD platforms.
2. Does higher NB speed provide more benefits over tighter timings?
Looks like the answer is also yes. At 3000Mhz 1600 9-9-9 I could whip up better MaxxMem results than at 2700Mhz 1600 6-8-6, so NB speed provides more benefits; of course I couldn't really prove what is equal and what is not (i.e. NB is 300Mhz higher is equal to how many timings lower?) but just so you know it can provide somewhat more benefit. Of course, do consider that you have 5-6ns more latency with CL9 vs CL6-8-6.
3. Do tighter timings still provide benefits?
Of course they will still provide good benefits on AMD. You can see the tighter timings provide a 400-500MB/s benefit in the benchmarks, and the latency is much lower.
4. So, in essence, NB speed and tighter timings scale about the same way?
Yes, so if you are running 1600 CL6 @ 2700-2800NB vs 1600 CL9 @ 3000Mhz NB, you get similar results +/- 100-200MB/s due to similar scaling patterns.
5. Which scales slightly better?
NB speed would scale slightly better over RAM timings/speed. So if you can get a higher NB speed but will have to slightly sacrifice RAM speed/timings, chances are you will get better bandwidth from this, but your latency will suffer.
6. Latency vs. bandwidth? Which is better?
I guess it really depends on what you do. On bandwidth intensive apps such as video editing, certain benchmarks, if you have an SSD, etc. the extra bandwidth provided would probably prove to be a lot better. However when you consider online gaming/ping, boot/app load times, etc. latencies would probably have more effect if the bandwidth is already fairly high. Remember, lowering timings can result in both extra bandwidth and latency. You may also want to look at individual values between different timings/NB speeds, i.e. read bandwidth vs. write bandwidth, depending on what you do.
7. Now what?
In conclusion both are significant and you should go for the highest NB possible and the lowest timings possible, because both give you the greatest results. Of course, lots of memory out there won't reach 1600 @ lower timings than just CL9, and lots of CPU IMCs out there (pretty much everything non-x6) won't reach 3000NB. So if you're looking for absolute best memory performance, these are the things to consider.

Notes:
-CPU speed appears to be around 1200Mhz, this is due to Cool'n'Quiet, something I can successfully apply to my CPU despite overclock.
-Actual CPU speed (during load) is 4.05Ghz in these tests.
-Tests were done on Windows 7 64-bit, on my "Benchmarking" user account which is separate from my main-use user account in that it does not have any startup programs, basic GUI, has more services disabled... basically, more suited for benchmarking.