If DDR4 memory is so good....

[Daddyjaxx]

Why did GPU companies totally skip it and go to 5?

 

[Bobnova]

GDDRx and DDRx are not related, they're totally different technologies.
GDDR4 did exist, just not for very long. It was a small evolution of GDDR3.

 

[EarthDog]

Apples, meet oranges.

 

[Witchdoctor]

If throw out gddr as it is not applicable and look at what ddr4 will bring to the tabel

http://www.simmtester.com/page/news/showpubnews.asp?num=179

Thinking we will see factory speeds over 4000 before it is all said and done.

I will be screaming fast, question is how will latancies hold up. and how rare and expensive will top end stix be ?

 

[Evilsizer]

well if you look at some of the calcs done in the past using DDR2 speed vs timings. in actual time depending on timings and speed, could be as low as 9ns to 15ns. this might be old but it still is going to show what the ram will do latency wise, not timings.
http://www.thetechrepository.com/showthread.php?t=160
as speed increase so does the timings used, so that everything can line up sort to speak. some ram will be binned for tighter timings and will be on the lower side of actual latency ie lower timings. since from what i recall DDR4 is still going to be 32bit per channel, so they just increase speed. im not sure how much harder it would be just to make the jump to 64bit per channel, then with dual channel, you kinda get the idea. guess thats why intels higher end socket/chipsets went quad channel, for the wider path.

 

[wingman99]

Single channel is 64bit dual channel is 128bit. When the DDR memory moves from DDR2 to 3 to 4 the thing that changes is the internal prefetch to I/O buffer in the ram, DDR has 2 bit prefetch DDR2 has 4 bit prefetch DDR3 has a 8 bit prefetch and DDR4 has 16 bit prefetch so with all that it takes more time to internally prefetch the bit's and that causes a increase in timing.

http://www.hardwaresecrets.com/arti...o-Know-About-DDR-DDR2-and-DDR3-Memories/167/5

 

[Bobnova]

Going to 128bit DIMMs would mean a tremendous number of pins. Current DIMMs (DDR2, DDR3) have 240 pins, DDR4 is still 64bit/DIMM and has 284 pins. 128bit would likely be >400. That's a lot.
What it would also mean is a lot more PCB traces (both on the DIMM and on the motherboard) to length match. That means more layers on the motherboard PCB, and layers are quite expensive.

 

[Daddyjaxx]

I read an article eons ago when GPUs went to GDDR5 their opinion was that regular RAM would also skip DDR4. I bet it will be very expensive at first. Shoot some DDR3 still is. I think mine was 500 for 32 Gb of 2666. I can't run it at 2666 or it down clocks my CPU using a CPU strap of 125 using XMP. If I manually set it to 266 it won't boot so I'm sticking to 2400 for now. I dropped a pair of pliers on my RAM slot and now D1 is dead.

 

[Silver_Pharaoh]

I read an article eons ago when GPUs went to GDDR5 their opinion was that regular RAM would also skip DDR4. I bet it will be very expensive at first. Shoot some DDR3 still is. I think mine was 500 for 32 Gb of 2666. I can't run it at 2666 or it down clocks my CPU using a CPU strap of 125 using XMP. If I manually set it to 266 it won't boot so I'm sticking to 2400 for now. I dropped a pair of pliers on my RAM slot and now D1 is dead.

I read & researched that DDR4 will skip consumers for the first year, and be available to servers only

 

[Bobnova]

Most of the benefits of ddr4 are server things anyway. No great loss there.

 

[Silver_Pharaoh]

Most of the benefits of ddr4 are server things anyway. No great loss there.

Very true.

The videos I saw about DDR4 had it running at CAS 14 @ 21something Mhz

 

[Bobnova]

Yeah that's more or less what crucial said when I talked to them about it last year. Great for sustained transfers, not so much for random access.

 

[EarthDog]

.... At 1.2v. Big difference.from the 1.5 we see at that speed.

 

[ATMINSIDE]

Very true.

The videos I saw about DDR4 had it running at CAS 14 @ 21something Mhz

And at first we only saw DDR3 in 1066/1333 CL10. Now look where we are.
The same thing will happen again, missing the early portion won't be a bad thing.

 

[Evilsizer]

.... At 1.2v. Big difference.from the 1.5 we see at that speed.

well could you imagine 1.5v though DDR4, what it might do! thats assuming the ic's can stand up to that kind of voltage long term. its going to come down to internal shielding/manufacturing process, if that high a voltage might even be possible.

** internal shielding might be the wrong term here. only one i could think of that would work/fit. not long ago though we could see the effects of something going on with cruical manufacture ic's for DDR2 even at a low or stock 2.1v. i dont recall ever hearing why/what was causing all those IC's to fail.

 

[Silver_Pharaoh]

And at first we only saw DDR3 in 1066/1333 CL10. Now look where we are.
The same thing will happen again, missing the early portion won't be a bad thing.

Indeed.
People would just stick with DDR3 unless DDR4 is super fast and has lower latency. Just a matter of time

.... At 1.2v. Big difference.from the 1.5 we see at that speed.

True, but does .3 volts really amount to a lot of power savings?
I mean I'm all for using less power for sure..

 

[Bobnova]

It's not 0.3v so much as it's ~20% lower voltage.
Given a resistive load (which RAM is not, but can sometimes be modeled as), that means a ~36% drop in power consumption.

When you figure Google will be switching over eventually, it's a pretty big power savings. In your PC? Not so much.

 

[EarthDog]

'dat box... way to get outside of it BobN!

 

[Bobnova]

It's thinking in the box that I find difficult

 

[Evilsizer]

It's thinking in the box that I find difficult

we dont have time for that nonsense here....