Does triple-channel memory really make much of a difference in a 12-core system (i.e. installing 6 RAM modules instead of the full 8)? The only articles I've been able to find are from a couple years back, and claim about a 2-3% performance increase. Curious if those claims still stand . . .
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Triple-channel memory
- Thread starter 5050
- Start date
- Sort by reaction score
That 2 or 3% claim would be an average overall increase. If you're doing something where the data moves at the speed of memory then it's closer to 300%. Just remember that if that operation takes only 5 seconds to complete and the next few operations done are not ram intensive and require 90 or so seconds then the total speed increase will only be about an overall 8 or 10%. With the triple channel saved you only 10 or 12 seconds in total.
Here you can see that the more the operation(s) you're doing need to operate at the speed of RAM the more the speed increase will matter. For example in a 3D rendering engine rendering an animation the geometry needs to be moved with every new frame before the light and shading can be calculated. If the geometry were comprised of a billion or so vertices and the shading and lighting were simplistic then most of the time spent by the rendering engine would be moving and sorting this vertices which is something mostly limited only by "the speed of RAM" and the difference between triple channel and single channel configurations would probably be in the realm of a 200 to 250% speed increase.
Video editing is another area where the benefits of triple to dual channel RAM configurations is or can be, quite significant. etc. Another is displaying a few thousand cached file icons. And so on and so forth.
Last edited:
Hi Tesselator,
What type of work do you do?
We seem to use our workstations in the same capacity (or at least your breadth of knowledge on the subject suggests). I work in motion graphics, compositing, and visual effects. My last job was the first time working with Element 3D, and that pushed my previous system configuration to the max. After the job, I re-evaluated my setup, which set into motion an aggressive round of hardware upgrades.
Anyways, back to the RAM question, does triple-channel bandwidth really matter when you have "enough" RAM? It sounds like the bottleneck for non-triple-channel configurations happens when the demand for RAM is greater than the installed amount. Is this a correct assessment?
I'm a jack of trades I suppose: Page Layout (Artist/Manager), CGFX for film and television (Producer/director/artist-[modeler/animator/matte/comp]), Programmer (Application/Game/DB coder/designer), Music (composer/artist), Hardware (reseller/importer/designer-manufacturer), VOD systems (maintenance/design - SE), Camera operator (video/film/stills), Optical systems (maintenance/repair/sales) etc. it goes on... and even gets into manual labor like cabinet maker, framer (houses), roofer, boat builder, and so on. I've even recently been thinking about getting into custom leather carving. If you don't mind me bragging a little here's my very 1st project finished just a couple of months ago:
VFX/Comp can be demanding so I sure get that.
No. Those are two separate issues; quantity and speed. The only time quantity affects speed is when the OS needs to cache to disk and that will happen at the same point whether single, dual, or triple channel is configured. It really can be thought of as RAID0 - as opposed to the same drives being only concatenated. The volume size stays the same but I/O speeds are faster one way over the other.
In general we can say that if you're using applications which R&W large amounts of RAM then you're a candidate for dual or triple channel configs. One should always generally try to achieve it anyway. I mean why not? It's faster after all and RAM is pretty cheap these days. But if currently not configured that way and you don't have such a need then there's no real reason to go buy more just in order for that.
Last edited:
you can run 96GB and stay triple channel. If thats not enough maybe it's time to evaluate staying with OSX or treat you DP as a temporary machine until a new one is/not announced
I'm a jack of all trades too, but I must admit I have a few less tools in my utility belt than you ("production" wise, music is probably the only thing I don't do). The custom leather work looks great! Thanks for sharing! I guess it's clear why you're an excellent Mac Pro candidate!
The RAID 0 analogy helps. Curious why the data rate does not increase as the multiple of the number of channels increases (i.e. it performs best at 3), but I assume this is an inherent limitation of the memory being "triple-channel" memory? And I assume "quad-channel" memory would push data even faster (and is there anything faster than quad memory available today)?
Gullibility? Or just a Microsoft hater?
Music is pretty easy tho! Becoming accomplished at playing an instrument is a little different but composition only requires an analytical mind. I can play classical guitar a little bit and of course anyone can play keyboard but I'm pretty out of practice these days.
There probably is I dunno. But I think it's a diminishing returns thing with RAM just like it is with RAID0. For example with RAID0 two drives will be like 1.9x (almost two times the speed), with three drives it's about 2.5x, and again with 4-drive RAID0 you only get about 3.1x or something (barely over three times the speed). I do think I remember seeing Intel datasheets that mentioned quad-channel tho so I guess it exists somewhere.
I guess the top part of your paragraph is asking why the increase isn't linear in ratio to the number of channels? I think it has to do with latency and controller switching (also part of latency). But also I don't know what the actual theoretical throughput and bandwidth is for the different configurations so I dunno just how linear it is nor what the curve looks like with RAM.
"Triple channel" is a bit like having 3 doors on a store or 3 rivers coming out of a lake. You can support more traffic going in and out with 3 doors than just 2 or 1.
It is really 3 memory controllers, each controlling the DIMM(s) on they are connected to. On current Mac Pros two of them have assigment of a single DIMM. The 3rd has two DIMMs. If fill both of those two then the controller has to "split time" between the two it has. Timing issues with the other two means the whole set-up takes a slight hit on well distributed memory accesses.
Four memory controllers doesn't mean the memory is faster it means that 4 read/write requests can be in flight at the same time. The system throughput/bandwidth is higher. Higher bandwidth is important as crank up the number of cores. Think of those as people going in/out of a store. If just two people then one door may suffice. If there 10 people trying to enter/exit at the same time then one door is likely to cause a queue to build up. If there were four doors there will be little to no queue (as long as everyone doesn't target one specific door).
The total bandwidth to memory does increase as add more controllers under most normal conditions. One 16 GB DIMM and one 8 GB DIMM aren't going to have the same bandwidth as three 8 GB DIMMs even though present as the amount of total memory.
RAID 0 is a bit off because real memory locations aren't proactively distributed over the DIMMs at fine grained levels. The speed required to respond to read/write requests pragmatically limits things to a simple mapping.
Does this mean that you will realize maximum performance with 3 equal sized RAMs installed, but that adding a 4th one will cause a slight performance hit (which may be offset by having the additional memory)?