>>54764923
It is a meme. Just like linux, ssd, jews, gimpworks, staleman sticky etc. this whole board is one dank meme

>>54764923
Always double the amount of RAM /g/ recommends.

There was a time when they laughed at people who had 8GB.

>>54765033
>>54765043

so theres zero truth to how much an application can actually use total based on any 1 dimm's size?

>>54765061
You have to distinguish between memory leak of unoptimised application and actual use but generaly speaking, more ram = smoother performance which of course applies to certain point in standart use of your pc. Also more ram can not degrade your performance so not paying 20€ extra for 16GB is a joke.

>>54765061
Yes, zero.

>>54764923
watch some videos on youtube where they compare running memory in single, dual and quad-channel, and there's barely any difference. biggest difference is around 20% slower file compression in 7zip with single channel iirc.

RAM is so fast these days that it's never a bottleneck for 99% of use cases, the only time RAM presents a problem is when there's not enough of it, forcing the system to page data to disk. In the old days paging slowed your system to a crawl, but with SSDs it's not that much of a problem at all.

>>54765061
how much an application can actually use is based on things like whether or not it's 32 or 64-bit, with the former being limited to about 3.5GB RAM (although there are ways to attempt to force them to use more). another aspect is how efficiently the program is actually designed to use RAM. You can fully expect a professional program like After Effects to literally eat up all the RAM you have, but there is nothing about the hardware itself that dictates how much of it can be used.

The one thing I can think of are multi-CPU systems, they split resources across multiple CPUs, so if the OS suddenly needs to access the RAM or PCI lanes assigned to CPU1, it will have to do so through the QPI link from CPU0. If your program is designed for this kind of access (like AE), then you'll get a performance increase. if not, it'll be slower than a comparable single-chip system.

>>54765106

how much actual speed does a dual channel set have over 2 single dimms

>>54765141
memory doesn't have channels, your motherboard does. If you put two dimms into the proper slots, they automatically become dual-channel. if you put them in the slots of different channels, you might run into boot or performance issues.

>>54764923
I saw a couple of posts about this the last week and I was puzzled by it too. I've done some research and it seems to me it's total BS.

There are no such limits in a 64-bit application. It does not relate to number of dimms or dimm size at all.

If anyone knows better and can correct me with citations, I welcome it.

>>54765627
there are limits but its not like any of us are anywhere near it.

the limit is always on channels and on the motherboard its self becuase wire traces and bus speeds and contorlers

for example most mobos made on the 1151 mobos are ether 32GB~64GB limits but here are like extened 128GB limits on x99 mobos for the broadwell umbrella for proessors the only ones i know that have super large limits are servers and hardware

its mainly because making it without limit makes it even more expenvise to make and so forth in my theroy at least.

>>54766277
Well Intel specifies RAM limits of 64 GB for Skylake and Haswell-E, even though 128 GB does often work anyways on LGA 2011 platforms.

No point officially supporting memory configurations on a Mainboard if the CPU doesn't support them as well.

>>54766546

why exactly wouldnt a CPU handle the memory?

what limits that?

and why does Intel's CPU's almost always only support the 1st or 2nd slowest of each DDR generation and need OC for higher, but AMD usually natively supports all except the highest ends of each DDR generation without OC

>>54766687

Intel jewry in a nutshell

Okay so basically for dual channel motherboards you can only access memory from 2 DIMMs at once per application unless it has been specifically written in order to take use of more RAM (generally pro software AE, Maya, etc) This isn't really much of an issue when using 8GB DIMMs as 2x8GB is more than enough for pretty much any single application outside of the professional world, and if you needed more another 2x8GB would give you 32GB, meaning any single application would have access to 16GB assuming a 64-bit application. And you could have two applications running that use 16GB each without issue.

Where issues CAN come into play (but would still be very rare) are 4GB DIMMs. In a dual channel motherboard 2x4GB is getting to the point some games are using more than this and when trying to access the other 2x4GB most games take a performance hit when you compare to a system with 2x8GB DIMMs and the game is only accessing the 2 DIMMs instead of trying to use 4x4GB on the other system.


So TLDR is basically for 90% of situations it wont matter at all, there are some rare occasions where it will, and to be safe I recommend at least 8GB DIMMs in 2016.


Also this issue still exists with triple and quad channel memory, but programs can access 3 or 4 DIMMs at the same time respectively. So it's much less of an issue.

>>54766687
>why exactly wouldnt a CPU handle the memory?
>what limits that?
>and why does Intel's CPU's almost always only support the 1st or 2nd slowest of each DDR generation and need OC for higher, but AMD usually natively supports all except the highest ends of each DDR generation without OC

A chipset has to address a certain amount of memory. Chipsets are generally locked to one processor type, generally defined by socket. The chipset usually supports as much memory as is expected to be used, based on channel capacities.

I have a 1366 desktop that has 3 channels for a total of 6 DIMMs.I have 4 gig DIMMs in it for 24 gigs of memory. That's the largest unbuffered and unregistered memory it'll accept. If it had support for buffered and registered memory I could probably take it to 8 or 16 gig DIMMs.

I have a server that has a similar 1366 CPU set. Each CPU has three channels but each channel takes 3 DIMMs, for a total of 9 DIMMs. Since they're XEONs it'll take more than single ranked memory. Each memory rank is like another addressable DIMM. You can get single, double or quad ranked DIMMs. Each channel on my chipset can address a maximum of eight ranks so if I want to fill all the DIMM slots I have to use single or double ranked memory.

Most of the time the memory runs at some multiple of the system bus speed. You get the most performance by overclocking your bus speed but you can't go too high that you overheat your northbridge or throw off your PCI-e timing. When you increase bus speed you can use faster memory to match. There's no reason to pay a premium for faster memory if your bus is going to be overclocked.