DailyTech - Hynix Introduces World's First 7Gbps 40nm 2Gb GDDR5 DRAM

Submit News

Hardware Hynix Introduces World's First 7Gbps 40nm 2Gb GDDR5 DRAM
Jansen Ng (Blog) - December 21, 2009 4:56 PM

18 comment(s) - last by haukionkannel.. on Dec 23 at 3:34 AM

Coming next year

Most video cards shipping today come equipped with 1GB of graphics memory. This enables the GPU to access large amounts of fast memory on the video card without having to go through system memory. The amount of memory used in video cards has been growing rapidly in size in order to accommodate the capabilities of today's modern DirectX 11 GPUs.

GDDR5 DRAM has become the dominant standard due to its high bandwidth capabilities. First introduced in 2007, it has since been demonstrated by Samsung to be capable of 7Gbps, ensuring a bright future ahead.

Hynix was the first to show off 65nm 1Gb GDDR5 chips in 2007, and has since made significant advances. The company has now announced the world's first 40nm-class 2Gb GDDR5 DRAM capable of 7Gbps of bandwidth.

The new chip was designed to minimize power consumption with a 1.35V operating voltage. This is a 20% reduction over the company's 50nm class GDDR5 chips. Hynix plans to start mass production of 2Gb GDDR5 in the second half of 2010 in order to meet the increasing demand for high performance graphics DRAM.

Increased bit densities are enabled by moving to the smaller, more efficient process. This will allow video card manufacturers to either use fewer chips for the same amount of memory, or double the size of the frame buffer.

Future cards similar to ATI's Radeon HD 5970 are prime candidates for the new chips. The world's most powerful video card is currently equipped with 2GB of GDDR5.

Comments

Threshold

Username
Password
remember me

This article is over a month old, voting and posting comments is disabled

It seems like the author doesn't know what a frame buffer is for

By jdietz on 12/21/2009 5:52:11 PM , Rating: 5

A frame buffer is just that - a piece of memory that stores the picture that is on the screen. Nowdays for add-in video boards, the frame buffer is located within the card's on-board memory (called video RAM).

A technique called "triple buffering" uses three frame buffers to do the screen drawing. A long story short, first there was one frame buffer, then "double buffering," and then triple buffering. Triple buffering uses three times the amount of memory as single buffering because it uses three frame buffers instead of one.

How much RAM does it take to do triple buffering? It depends on your screen resolution and color depth. A 1280x1024 screen at 24 bit color depth needs about 4MB of video ram for one frame buffer.

The frame buffer is not the main use of the video ram. The main use is storage of texture data which is used to render the frame. Speed is important when considering texture data - the faster texture data can be accessed, the faster the frame can be drawn. Speed is not an issue for the frame buffer. The actual frame is many times smaller than the texture data used to create it - therefore high speed is not required.

RE: It seems like the author doesn't know what a frame buffer is for

By SSDMaster on 12/21/2009 6:11:42 PM , Rating: 4

I don't care about the accuracy of the article. As long as there's no spelling errors...

Parent

RE: It seems like the author doesn't know what a frame buffer is for

By MarkLuvsCS on 12/21/2009 6:15:02 PM , Rating: 5

I don't care about minor mistakes, such as spelling, as long as there are no factual errors.

Parent

RE: It seems like the author doesn't know what a frame buffer is for

By salimbest83 on 12/21/2009 6:24:01 PM , Rating: 2

+10

Parent

RE: It seems like the author doesn't know what a frame buffer is for

By AnnihilatorX on 12/21/2009 6:23:55 PM , Rating: 5

lol they need to do a poll on DT main page on this

Parent

RE: It seems like the author doesn't know what a frame buffer is for

By Barfo on 12/21/2009 6:23:45 PM , Rating: 5

If I cared that much about spelling or factual errors I wouldn't be reading DT.

Parent

RE: It seems like the author doesn't know what a frame buffer is for

By JonnyDough on 12/22/2009 3:34:31 AM , Rating: 2

Since we can only rank you up to five (what the hell kind of rating system is this anyway) I would like to add the following to your post:

+6.

Parent

RE: It seems like the author doesn't know what a frame buffer is for

By thorr2 on 12/21/2009 6:58:01 PM , Rating: 5

"The new chips was designed"
"As long as there's no spelling errors"

Do you care about grammar errors? ;)

Parent

RE: It seems like the author doesn't know what a frame buffer is for

By riottime on 12/21/2009 8:11:48 PM , Rating: 4

DT is pretty bad in the writing department. They need to hire/contract better writers or at least hire an editor (or two) to correct the written articles by these bad writers.

Parent

RE: It seems like the author doesn't know what a frame buffer is for

By cochy on 12/21/2009 11:45:15 PM , Rating: 3

Well they will probably never do that because no matter how many times people complain about those things, the same people return to keep reading other articles...so why bother fixing something that ain't broke?

Parent

RE: It seems like the author doesn't know what a frame buffer is for

By mmcdonalataocdotgov on 12/22/2009 7:10:58 AM , Rating: 2

"...as long as there ARE no spelling errors..." not ".. as long as there is no spelling errors."

Parent

RE: It seems like the author doesn't know what a frame buffer is for

By GourdFreeMan on 12/21/2009 9:01:10 PM , Rating: 5

Good points, however the statement:

"Speed is not an issue for the frame buffer. The actual frame is many times smaller than the texture data used to create it - therefore high speed is not required."

is only true if the programmers have used a depth-first pass or other optimization to avoid overdraw. If you don't do so frambuffer traffic can take ~10% of total bandwidth. Also, it isn't the total size of texture data that matters, but rather the number of samples taken per pixel drawn that results in the ratio of texture vs. framebuffer bandwidth... except, of course, wherein larger textures result in fewer cache hits for texture data.

Parent

RE: It seems like the author doesn't know what a frame buffer is for

By greylica on 12/21/2009 10:31:39 PM , Rating: 1

And with all of this technology of triple buffering to speed things a lot ( pre-render frames, etc ). Microsoft insisted to use Vsync in Vista and Win7 as standard...
Avoid new versions of Windows if you use lot´s of OpenGL, the performance is deceptive, the only way is to use Rivaturner to enable triple buffering back again.
The only hope I am having, is Nvidia and ATI enabling a change in the driver itself to move back to Triple Buffering if the user need (A simple choice in the driver panel). Otherwise, we will continue using Rivaturner to make the right thing...

Parent

RE: It seems like the author doesn't know what a frame buffer is for

By jeromekwok on 12/22/2009 2:08:07 AM , Rating: 2

Think about the 2D VGA cards and original 3dfx Voodoo in the old days. The frame buffer and texture memory are physically on separate cards.

Today shader programs are also running on video memory. Thus performance is important.

Parent

Energy savings

By haukionkannel on 12/22/2009 8:20:35 AM , Rating: 2

Not so bad. ATI 5000 series cards are not bad from energy effience point, so these will actually help redusing the energy usage even more with future cards.
Who wants to bet that Nvidia Ferni will be the first to actually use these chips? When the cost is not the limiting factor :-)
Normally it has been ATI who has gone first to new GPU memory technology (during the last years), but this time Nvia may be so desperade that they need every inch of power they may get.

RE: Energy savings

By jonmcc33 on 12/22/2009 9:06:33 AM , Rating: 2

No, they will probably still be using something old like 512-bit GDDR3 since they can't seem to develop any new memory technology on their own like ATi can.

Parent

RE: Energy savings

By haukionkannel on 12/23/2009 3:34:25 AM , Rating: 2

It's guite possible ;-)

But let's hope that situation is not so desperate... But history has a tendency of repeating itself.

Seriously speaking Nvidia needs a victory even a small one, to clean up it shield, so maybe they use this option even if it would make Ferni cost over $1000...

Parent

I don't get it...

By jp7189 on 12/22/2009 10:26:50 PM , Rating: 2

7Gbps is wicked fast for local video memory, or any memory, but I work on a 10Gbps network, and I see Verizon is now deploying a 100Gbps backbone in Europe. Latency aside, are you telling me I can send data across the building faster than across my video card?