Samsung Mass Producing 8 Gb GDDR5 Memory

Yet my 970 has hynix memory. Oh my God I will never get over this.

 

Good m soon will start seeing 16 - 32 GB and evb 64 GB Graphic cards .. Deal with that console lovers

 

a single lil 8gb ram chip will help massively with power /cooling.
gtx 980 direct cu mini 8gb waw.

 

It is painfully clear that not everyone understands what king of capacity this chip has.
Nor they understand bandwidth and why one need more memory chips.

And there is final kick in balls called capacity versus bandwidth. So, dream on.

 

christ, so the next gen of gpus will prolly start to come with 8gb as the bare minimum. Lets hope the costs are not stupid (er)

 

Not likely - nvidia wants to keep the vram increase pr generation at a minimum, so people are forced to upgrade more often.

 

Um actually the '8Gb' means 8 Gigabit, which refers to its operating speed rather than capacity.

 

Thank you for pointing that out. I would have felt like a smartass if I'd have posted it before, was reading HH's article twice just to be sure I understood it correctly

 

What does that mean?

 

You are correct sir.

 

so this memory is running at 8GHZ then? (to get the 64GB/s with 2 chips) hopefully those numbers (or better) can be realized on graphics cards.


Edit: To Pill, the article states 2GB (gigabytes) of memory can be created with just 2 chips, meaning each is 8gb (gigabit) since 2 chips would be 16gb divided by 8 = 2GB.

 

Will any of this matter once stacked memory starts being used?

 

Stacked memory or HBM use "GDDR5" technology ( Not sure we can still call it GDDR5 this way, because we are no more really speaking about the same "package" ) , its just the way they are implemented on chips and connected with the GPU / CPU who is different ( ofc )..

all innovation on the GDDR side can be usefull on the HBM side. ( The "ram transistors" stay the same but are stacked and you have an interposer )

Its a little bit strange that Samsung dont annunce allready stacked GDDR as finally they allready do it with their Nand ( SSD ).

If rumors point that AMD could implement it on their next line up, Nvidia will not before 2016-2017 Line up.

HBM should cost a lot to implement, 4096bit memory controller, the interposer, the developpement etc.. ) i dont think we will see it on a top to bottom lineup.. standard GDDR5 have still some good years in front of them.

 

The title is a bit confusing, I also assumed chip density even though it states Gb not GB.

 

It is correct there. Capacity of each chip is 8Gb = 1GB. Each has standard 32bit interface with 256Gbps = 32GB/s data transfer rate.
it is at edge of usability for graphics memory purpose too. 16Gb = 2GB per chip would not provide sufficient data transfer speeds.

 

Yeah I immediately (and incorrectly) assumed it was density since atm chips come in 1GB, 2GB and 4GB flavors....but then I read the article.

 

Article is little bit wishy washy. It is indeed on about density of the chip been 8Gb. Confusion a little here, think article means 8Gb capacity @ 8Gbps speeds, but misses off the Gbps an only refers to Gb. Article states would need 8 of these chips to match PS4 memory capacity.

 

You aren't kidding....

Like what does this mean?

I doubt Hilbert wrote any of it (I hope).




8x 1GB chips = 8GB total vram. Each 1Gb chip can manage 8Ghz, up from the 7Ghz currently produced by Hynix.
Does that sound right? I guess so...

 

Yes, this is the best definition so far.

so remember, 1 Byte (B) = 8 bits (b). In this case, if you take 8 chips, each with 8 Gigabits (64 Gigabits (Gb) total capacity), you can divide 64Gb by 8 to get 8 GigaBytes (GB) capacity.

So in other words, 8 of these chips will give you 8GB GDDR5 because they are 1GB each. Using 4Gb chips you previously needed 16 chips to get that capacity.


This works for transmission speeds as well, so 64Mbps is the same as 8MB/s