Pros and cons of linked vs unlinked RAM?

[Nena Trinity]

Okay if you have two RAM sticks with non-matching specs its stated by the most that its better to have them unlinked. So is this true? I kinda want some input on this such as stability and does it actually work is it faster in anyway to have it unlinked? Right now I have it like this...

 

Linked

Channel 1: 1866MHz CL9-10-11-27 (HyperX Savage)

&

Channel 2: 1866MHz CL9-10-11-27 (Patriot Viper III) slower speed

 

Unlinked

Channel 1: 1866MHz CL9-10-11-27 (HyperX Savage)

&

Channel 2: 1866MHz CL9-10-9-27 (Patriot Viper III) default speed

 

I personally see no instabilities or advantages of running them in linked or unlinked so what is going on? 

 

[Nena Trinity]

Turnip has the lead really?

[Nena Trinity]

Nobody knows really? o3o

[Nena Trinity]

I expected more votes and posts by now...

[LukaH]

if you notice zero difference in system speed/instability when linked or unlinked just run it linked mode... i am assuming that linked=dual channel?

[runit3]

linked is an odd term to use. If you mean linked as dual channel operation with matching speeds, that would be the correct choice. You can run them separately in single channel operation with miss-matched latency, but the difference between tRCD 9/11 is so small there's no reason not to run them in dual channel. The benefits of dual channel operation outweigh the very small increase in latency for one primary timing. http://www.gamersnexus.net/guides/1349-ram-how-dual-channel-works-vs-single-channel/Page-3

[Nena Trinity]

1 hour ago, LukaH said:

if you notice zero difference in system speed/instability when linked or unlinked just run it linked mode... i am assuming that linked=dual channel?

In theory the unlinked should be faster!

49 minutes ago, runit3 said:

linked is an odd term to use. If you mean linked as dual channel operation with matching speeds, that would be the correct choice. You can run them separately in single channel operation with miss-matched latency, but the difference between tRCD 9/11 is so small there's no reason not to run them in dual channel. The benefits of dual channel operation outweigh the very small increase in latency for one primary timing. http://www.gamersnexus.net/guides/1349-ram-how-dual-channel-works-vs-single-channel/Page-3

Umm no its still dual channel but...

 

Linked (all four sticks run at same speed)

Channel 1: 1866MHz CL9-10-11-27 (HyperX Savage)

&

Channel 2: 1866MHz CL9-10-11-27 (Patriot Viper III) slower speed

 

Unlinked (two and two run at same speed in pairs)

Channel 1: 1866MHz CL9-10-11-27 (HyperX Savage)

&

Channel 2: 1866MHz CL9-10-9-27 (Patriot Viper III) default speed

[runit3]

1 minute ago, Nena360 said:

In theory the unlinked should be faster!

Umm not its still dual channel but...

The colors, THEY BURN

 

You need to clarify the "linked" thing. Are you manually going in and changing tRCD for the Patriot kit? If they're in dual channel operation the Patriot kit should automatically be matching timings to the slower (Kingston) kit. You would need to confirm any manual adjustments with CPU-Z within the SPD tab. If you are manually setting the Patriot kit to run tRCD at 9, and you're not currently experiencing any hard memory faults, I would expect them eventually. If you're hell bent on defying all logic and reason manually set tRCD to 9 and run memtest overnight, I would be surprised if you weren't throwing errors from the manual miss-match.

[Nena Trinity]

Just now, runit3 said:

The colors, THEY BURN

 

You need to clarify the "linked" thing. Are you manually going in and changing tRCD for the Patriot kit? If they're in dual channel operation the Patriot kit should automatically be matching timings to the slower (Kingston) kit. You would need to confirm any manual adjustments with CPU-Z within the SPD tab. If you are manually setting the Patriot kit to run tRCD at 9, and you're not currently experiencing any hard memory faults, I would expect them eventually. If you're hell bent on defying all logic and reason manually set tRCD to 9 and run memtest overnight, I would be surprised if you weren't throwing errors from the manual miss-match.

I did run Memtest86 and it never failed at all but none of my checking tools shows the memory sticks separate speeds they all show the speeds of the first pair of sticks... (in other words I am not even sure it works even if the BIOS says so...)

[runit3]

Just now, Nena360 said:

I did run Memtest86 and it never failed at all but none of my checking tools shows the memory sticks separate speeds they all show the speeds of the first pair of sticks... (in other words I am not even sure it works even if the BIOS says so...)

So each kit listed is actually a pair of DIMM's within a separate bank? I assumed you had 1 stick of each, both operating within the same operation bank. This changes things a bit.

 

If you're not throwing errors, and you have 4 DIMM's, 2 matched to each other, and you're not receiving any errors in Memtest, then fuck it -run it how you have them labeled in "unlinked'. Again, would be worth checking the implementation with CPU-Z.

[Nena Trinity]

6 minutes ago, runit3 said:

So each kit listed is actually a pair of DIMM's within a separate bank? I assumed you had 1 stick of each, both operating within the same operation bank. This changes things a bit.

 

If you're not throwing errors, and you have 4 DIMM's, 2 matched to each other, and you're not receiving any errors in Memtest, then fuck it -run it how you have them labeled in "unlinked'. Again, would be worth checking the implementation with CPU-Z.

So you say I should check it again? And maybe run more memtest86s? My theory so far is it works and my current checking software cannot see it, or it does not work at all... (I will try CPU-Z!)

 

Slot 1: 1866MHz CL9-10-11-27 (HyperX Savage)

Slot 2: 1866MHz CL9-10-9-27 (Patriot Viper III)

Slot 3: 1866MHz CL9-10-11-27 (HyperX Savage)

Slot 4: 1866MHz CL9-10-9-27 (Patriot Viper III)

 

[runit3]

25 minutes ago, Nena360 said:

So you say I should check it again? And maybe run more memtest86s? My theory so far is it works and my current checking software cannot see it, or it does not work at all... (I will try CPU-Z!)

 

Slot 1: 1866MHz CL9-10-11-27 (HyperX Savage)

Slot 2: 1866MHz CL9-10-9-27 (Patriot Viper III)

Slot 3: 1866MHz CL9-10-11-27 (HyperX Savage)

Slot 4: 1866MHz CL9-10-9-27 (Patriot Viper III)

 

Should throw that in the OP. That configuration makes much more sense in terms of being able to function. I am not learned enough in RAM operation to give a 100% statement, but if logic dictates anything I would assume the RAM calls are bank segregated to the matching kits, meaning there's no cross-addressing of operation between each kit (I would hope).

 

Running AIDA's memory stress test and benchmark may be a good way to test full load saturation

[Nena Trinity]

1 minute ago, runit3 said:

Should throw that in the OP. That configuration makes much more sense in terms of being able to function. I am not learned enough in RAM operation to give a 100% statement, but if logic dictates anything I would assume the RAM calls are bank segregated to the matching kits, meaning there's no cross-addressing of operation between each kit (I would hope).

 

Running AIDA's memory stress test and benchmark may be a good way to test full load saturation

That sounds great! I will try that to!