Intel Memory Controllers

[genetix]

Hello, genetix on a row again..

Well, I got a bit curious on RAM since I've been testing all sort of systems and here's the deal. at:

(Please, read the whole thing I write through generations to understand my meaning before commenting back.)

LGA 775/X38 - Specifications of Intel pages states:
* DDR2(/DDR3) - Dual Channel:
- Rated Speed: 10.67 GB/s FSB: 1333MHz / 1066MHz / 800MHz

Now DDR2 was times premium and DDR3 was just upcoming and an 800Mhz DDR2 was 800Mhz 1:1 giving rough 9000-10600MB/s of speed out.

LGA 1366/X58 - Specifications of Intel pages states:
* DDR3 - Triple Channel:
- Rated Speed: 25.6 GB/s -> DDR3-800/1066
- Other: Dependable on Uncore clock entirely. (Ex. 4000Mhz Uncore reflects exactly the bandwidth of DDR3 controller > Your maximum memory bandwidth = Uncore clock)

We were using 1600-2400Mhz DDR3 while the speed was calculated exact half of 'UnCore' Speed of CPU/Motherboard (Ex. 4000Mhz / 2 = 2000Mhz DDR3 speeds) and (ex. DDR3 1700-1950Mhz giving rough 16500-17500MB/s of speed out (NOTE: write speed was the only speed to watch at LGA1366 do not even consider to tell me otherwise as it was 100% limited to UnCore test, if you don't believe me)).

LGA 1155/Z68 - Specifications of Intel pages states:
* DDR3 - Dual-Channel:
- Rated Speed: 21 GB/s -> DDR3-1066/1333
- Other: Pre-setupped 'UnCore' increases through DDR3 setup specifications 1066Mhz -> +2400Mhz.


Now DDR3 maximum official speed is 1066-1333Mhz my question is, as we can see from above every single chip so far calculated as official specification has been an 'DDR speed' not DDR2 or DDR3 bandwidth dependent.

As in:
* DDR2 - 800Mhz "Effective Clock" = 400Mhz "Real Clock"
* DDR3 - 1600Mhz "Effective Clock" = 800Mhz "Real Clock"
* DDR3 - 2133Mhz "Effective Clock" = 1066Mhz "Real Clock"
* DDR3 - 2666Mhz "Effective Clock" = 1333Mhz "Real Clock"

Now, does the intel specifications state actual DDR3 speeds in "Effective Clock" or the "Real Clock" ?

(Why am I asking = at Sandy Bridge board while you set DDR3-1066/DDR3-1333 speeds "Effective Clock" (533Mhz/667Mhz "Real Clock") at BIOS this will be exactly half of CPU bandwidth exactly at JEDEC standard. CPU at locked stock speed.)

and IF this is actually understood incorrectly as above presented. Should we actually be aiming DDR3 2666Mhz speeds to actually utilize CPU memory bandwidth in full?

This would be my question... I've been searching answer through overclocking / hardware forums. I was wondering could G.Skill answer this question as understanding memory a bit better.

(I can also provide proof of concept I own and have screenshots of all Intel systems on hand)

Conversation Original post: Guru3D website

[GSKILL TECH]

Intel is ALWAYS effective clock, so their bandwidth values are always based on DDR3-1333. With overclocking, you achieve better results than stated.

Thank you
GSKILL TECH

[genetix]

Thanks for the answer. Opinions/understanding is welcome.

Somehow just the specification ain't that clear. Although, as far I re-looked the spec the speeds are mentioned as 'Max Memory Bandwidth' and not the bandwidth calculated by (6,665 x 100.0 = ~667Mhz x Double Data Rate = 1333Mhz = 21GB/s) as I probably was originally thinking.

Also, the new (""FSB"" or memory controller) clock seems to be hidden really deep. So, it stays sync with the DDR3 speed which is kinda weird perspective to start tweaking as we do not now
know the rate of (""FSB"" or memory controller) against utilizable bandwidth between the actual CPU clock.

or do we?