1TB DDR3 vs 512GB DDR4

Vusa360

Reputable
Aug 20, 2014
13
0
4,510
Which one of these options would perform better.

Option 1
32 x 32GB 1866MHz DDR3

Option 2
16 x 32GB 2133MHz DDR4
 
Solution
must be a server , no consumer motherboard holds that much , i think 128gb is max.

it all depends, quad channel? you can duplicate the memory setup for redundancy, ddr4 does use a bit less power but those speed differences wont make any huge difference, so the 1tb would probably be better, unless you run a huge server you will never use anywhere near those two numbers, that would be for multiple vm workloads
The first rule of RAM configuration is to have enough RAM for the intended application since nothing ruins performance faster than having to use the swapfile or reloading data from the storage system. Once you run out of RAM, it does not matter how fast the CPU or RAM is, all they will do is wait for the HDD/SSD.

Once you are certain you have more than enough RAM for your application, only then can you worry about putting in faster RAM and faster CPUs.
 
must be a server , no consumer motherboard holds that much , i think 128gb is max.

it all depends, quad channel? you can duplicate the memory setup for redundancy, ddr4 does use a bit less power but those speed differences wont make any huge difference, so the 1tb would probably be better, unless you run a huge server you will never use anywhere near those two numbers, that would be for multiple vm workloads
 
Solution
Was that you swinging this InvalidError 😉

maxresdefault.jpg
 



The answer is "it depends"

With DDR3, as you add extra banks of ram into the system beyond one per CPU (a bank can be 2 or 3 units depending on the architrecture), the bus speed will usually slow by one step per extra bank.

Usually... this changes if you are using LR dimms as the buffering means they don't slow down - but they weren't that fast to start with. It also varies on whether you're using buffered(aka registered) or unbuffered ram (which is different to ecc/non ecc)

With DDR4 there isn't that same slowdown.

BUT, that speed is the absolute maximum throughput. There's a delay of between 12-20ns between asking memory for data at an address and actually getting it. Memory controllers try to predict what's going to be asked for but they're not entirely accurate. Think of it like random vs io access on a mechanical hard drive.

In general, the DDR4 will be cheaper and consume less power per GiB. In practice you probably won't see much difference in actual throughput unless you're running specialised loads. At these kinds of sizes you're foolish not to use ECC/Registered ram so you're losing a few clock cycles to the buffers anyway.

For a new build: Take the DDR4 every time. For re-engining an old system, think about whether this is a good idea in the first place. The power savings of a new system alone can make up the difference in purchase cost of a new system over putting more memory in an old one

One example from the old days: Putting 32GB of FBDIMMs in one server I have increased power consumption by more than 300W over the 8GB it was originally fitted with. Changes between contemporary systems are much lower but it all adds up across system boards, CPUs, PSUs and even ethernet controllers.