why do companies offer 1gb, 2gb, 4gb, but not 3gb RAM?

hi. why do many companies offer customizations of 1gb, 2gb, and 4gb of ram? can't they just put a 2gb and 1gb in and make it 3gb? thanks! is there like a severe shortage of the 2gb ram stick?


edit: i called one of the companies. they said that 3gb would not be optimal for my notebook.

sager 5791

 

they tend to try and promote dual channel which can only be used if you have two sticks of the same size installed. it also helps get more money, their sticks are already expensive so 4GB is more expensive than 3GB. the majority of buyers will not know what dual channel does and that it gives minimal performance improvements, they will also not think of 3GB and assume you can only get 1, 2 and 4GB

 

dual channel. most laptap have 2 slots.

 

When you get larger amounts of RAM, you're going for performance. You get the best performance out of matched pairs of RAM, thus the reason the notebook company told you that 3 GB would not be optimal - the RAM couldn't take advantage of a pairing.

 

Dell does offer 3gb in some models, one 2gb stick, other 1gb stick. Don't know if dual channel would treat that as dual channel or not though.

 

Cheaper to buy, sell and assemble in bulk.

Less work = less labor = less wages = cheaper prices = more notebooks sold = win!

 

Honestly, if you want 3 GB and don't care about the slight performance hit from not using dual channel, just order 2 GB then buy a 1GB SODIMM separately.

 

thanks guys. all your answers are great.

so basically if i buy 4 gbs (2 x 2gb sticks), i can take advantage of dual channel even though my 32 bit vista will take advantage of 3gbs of it at most, correct?

 

yeah probably but dual channel really isnt as big a deal as most ppl are trying to make it out as. although it will give you a boost, the boost is tiny. go for 3GB and save yourself money cuz 4GB is very expensive right now even if you buy it from a store.

 

Remember that dual channel will also work with mismatched pairs aka 3GB (1x1GB + 1x2GB)

 

Vista can take advantage of more than 3GB.

It's not that it doesn't support 4GB RAM, Vista is perfectly capable of dealing with systems that have 4GB installed.

It's just that it can't, for other reasons, allow the last ~500MB to be *used*. But it's perfectly aware that they're there, so it doesn't affect single- vs dualchannel at all.

 

Why is Vista unable to write to the last 500MB, Jalf?

 

It's something with 32 and 64 bit Windows, isn't it?

 

For an explanation of why 32-bit operating systems can't use more than a maximum of ~3.5 GB, check out:
http://en.allexperts.com/q/PC-hardware-CPU-1023/Windows-XP-memory-limit.htm

The gist of it, as I recall, is that the 32-bit OS is limited to 4 GB total, but there is more memory being used than just the RAM, and this memory also draws against that 4 GB total (this includes video memory).

 

I think a better question is: Why don't they sell a 3.5 gig memory since that's all 32-bit Windows can handle?

 

Well, it's not Vista specifically, but all 32-bit OS'es, and it's not exactly 500MB.

But the reason is that a 32-bit OS pretty much by definition operates with 32-bit memory addresses. That is, each byte of memory is identified by a 32-bit address, and that places a strict upper limit on the number of available addresses.
32 bit gives us 2^32 different addresses, and as each address points to one byte, we can address a total of combined 2^32 byte = 4GB.

But please keep in mind that we're talking about memory addresses here, not actual RAM. To use a byte of RAM, the CPU has to be able to identify it with an address. The CPU operates with 32-bit memory addresses (So the numbers 0 through 4294967295 are valid addresses, but there's no way it can ask for the contents of byte #4294967296, because that simply won't fit in 32 bits)
It doesn't matter how much RAM is actually installed. (hypothetically, you could install 8GB in a 32-bit system), it just limits how many addresses the CPU can deal with, so if you have 8GB, half the memory couldn't be referenced by the CPU.

So a 32-bit OS has enough addresses to cover a memory space of 4GB. So far so good, but the problem is that this memory space is used for more than just RAM.

One common way to communicate with external hardware (harddrives or graphics cards are common examples) is to reroute a range of addresses so anything that's written to those addresses is intercepted by the memory controller, and instead of being written to RAM, is directed to that piece of hardware. And conversely, when the CPU tries to read from those addresses, what it gets is not the contents of those RAM cells, but the contents of the attached hardware.
That's called memory-mapping.

With harddrives this is used to map individual files to memory. Got a 500MB file you need to access often? Easiest way is to set aside 500 million memory addresses and make them point to this file instead.
Same story with graphics cards. We need to upload 200MB of textures for our game? Simply map a bit of GPU memory to a range of system memory addresses. Then the CPU just has to write the textures to these local addresses, and the data is magically transported to the graphics card.
The BIOS also uses a small chunk of addresses to make itself available to the CPU.
Very convenient (and efficient) technique.

Problem is, of course, all this requires memory addresses. We need to set aside a range of addresses to be rerouted, which means they can't also point to actual RAM.

So if we have 4 billion addresses available, and we have to reserve some of them to memory-mapped hardware, we're left with.... something less than 4 billion to address the actual RAM.
Exactly how much is left depends on the OS, your hardware, your drivers and other factors.

So if you install 4GB RAM in a 32-bit Vista system, Vista recognizes that you've got 2x2GB ram sticks installed, and you'll run dual-channel just fine. it's just that something like 500MB of memory will never be used, because their addresses are mapped to some piece of hardware instead.

On a 64-bit system, this isn't a problem, because if you've got a total address space of 256 terabyte, and only have 4GB RAM installed, there are tons of unused addresses that can be used to memory-map bits of hardware.

 

1,2,4,8,16,32,64,128,256,512,1024,2048,4096

see a pattern? basic computer logic circuitry doesn't make it readily feasible.

 

I wasn't serious, just fooling around.