Double vs triple channel: Running multiple stuff at once?

[Uraotoko]

Hello guys. I want to upgrade my mobo/cpu/memory kit and im looking for something that can last for many years without feeling "slow"

I once was looking at the 1366 socket intel i7-950 but with the new release of the sandy bridge cpu i am very tempted to buy the new intel 1155 socket cpu i7-2600.

My question is: Does triple channel help to run more applications at once? For instance, encoding while playing games (in order to stream).

I havent found a review that tests running multiple applications at once. They only focus on 1 at a time and syntetics tests.

-Is double-channel enough for my needs?
-Will triple-channel become a factor in the future?
-Do you think that the upcoming 1356 can exploit the triple channel bandwidth better?
-Should i rather wait for the 1356 socket cpu or the bulldozer from AMD?

Thanks.

Well i am not getting much response on this thread... lets discuss the theory of this.
Having a higher memory bandwith, in theory, will make more memory bandwith available for the CPU to use.
This will let the CPU make more read/write operations in less time.
Now, i am assuming that the CPU that supports triple channel (eg. lga 1366 socket) will be capable of using the whole bandwith, correct?

+So, in theory, the more software you have running in paralell the more read/writes you need, correct?
+And finally, in theory, triple channel should help me to run more stuff at once more smoothly, right?

Now, i am curious if the CPU can actually use more than dual-channel bandwidth while doing other operations rather than read/write. Most syntetic benchmarks that favor triple-channel do single operations like just read or just write.

Will i ever (near or far future) be capable of using the whole bandwidth in real life situations? like the one i wrote above.

 

[Emulex]

yeah on the x58 chipset - huge difference in triple channel mode with 6 dimms versus dual channel mode with 8 dimms. nehalem/westmere love extra memory bandwidth. remember MEMORY bandwidth is 90% of what made the speed increase from Core2 to I5/I7/Nehalem/Westmere real. if you choke memory - you choke processor big time.

Very real. If i add another dimm to the X58 board, it goes into dual channel mode and you lose 10-20% bandwidth str8 up (i have dual xeons on my x58).

 

[Blain]

 

[poohbear]

yeah on the x58 chipset - huge difference in triple channel mode with 6 dimms versus dual channel mode with 8 dimms. nehalem/westmere love extra memory bandwidth. remember MEMORY bandwidth is 90% of what made the speed increase from Core2 to I5/I7/Nehalem/Westmere real. if you choke memory - you choke processor big time.

Very real. If i add another dimm to the X58 board, it goes into dual channel mode and you lose 10-20% bandwidth str8 up (i have dual xeons on my x58).

how does this apply to SB which only has Dual Channel? Intel saw fit for it NOT to have triple channel for a reason, it simply doesnt need it i imagine.

But im curious about the OPs question as well, do some of you think SB would benefit from triple channel?

 

[Emulex]

oh so sandy bridge is too slow to handle triple channel ram i guess. lol. just stick to the x58 and old school i7 and you'll go faster.

seriously the amount of speed drop when i go to dual channel is enormous.

you know if you have the x58 (or equivalent sandy bridge) you get the option of running dual or triple channel.

if you don't overclock (i don't) the bandwidth to the processors is enormous when you switch from triple to dual. gimps out your X5670/X5680 big time. (or W single socket equivalent).

 

[IGemini]

Sandy Bridge can handle triple-channel memory, it's just that they haven't released the version that supports it yet. They target TC memory to the enthusiast market, it's the exact same concept as what they did with Nehalem and why you only see X58 boards support it, nothing's changed here.

 

[corkyg]

Keep in mind, also, that one of the main enhancers of multitasking is the 64-bit OS.

 

[Troll Trolling]

oh so sandy bridge is too slow to handle triple channel ram i guess. lol. just stick to the x58 and old school i7 and you'll go faster.

seriously the amount of speed drop when i go to dual channel is enormous.

you know if you have the x58 (or equivalent sandy bridge) you get the option of running dual or triple channel.

if you don't overclock (i don't) the bandwidth to the processors is enormous when you switch from triple to dual. gimps out your X5670/X5680 big time. (or W single socket equivalent).

Excuse me sir, but you are wrong.

Sandy Bridge is faster than Nehalem/Westmere. It has already been proved that triple channel doesn't yield any significant performance increase over dual channel.
Most of the performance increase of Nehalem was because of architecture enhancements, not because of more memory bandwidth. Proof is Lynnfield

-Is double-channel enough for my needs? Not sure, but I think it is enough.
-Will triple-channel become a factor in the future? Maybe. Depends on how powerful the processors are going to be and if the newer memory technology will be able to feed them with enough bandwidth. I would bet not.
-Do you think that the upcoming 1356 can exploit the triple channel bandwidth better? Don't know.
-Should i rather wait for the 1356 socket cpu or the bulldozer from AMD? I would wait Bulldozer.

 

[Emulex]

i was yanking your chain dude

but triple channel is here to stay. and guess what? Now the 4-way procs have quad channel. It's a total flip upside down. The 16GB quad rank are faster than the single/dual rank. and running in hemisphere mode outweighs everything else - go read up on the new intel hemisphere mode for 4-way. totally breaks all the rules we learned with numa/suma intel architecture before.

What board you got? Maybe your board doesn't optimize for ram the same. typically you get interleaving for maximum speed with 6 dimms per cpu of the same time.

ie 8gb * 6 @ 1333 = perfect, with two cpu's that 8gb * 12.

going up one next level to 8gb * 9 drops to 800mhz (23-25% bandwidth loss).

Perhaps your chipset doesn't optimize the same way. The X58 and its server brother very much do optimize for 3 across and 2 deep for maximum interleave performacne PER socket.

But do google hemisphere mode. Quad-memory controller even faster.