Nvidia 7950GTX 2 Coming Soon

Mathew Everett 13 Apr, 2006 at 4:02pm ET Article published in Tech

Nvidia just told its partners that another quad SLI chip will be rolling out of the Nvidia Hangers by the end of this month, the difference being that these Dual GPU cards are going to be shorter than the current 7900GTX quad SLI cards and will sport a new name dubbed 7950GTX 2.

The retail only cards are branded as Geforce 7950 GTX 2. We remember that Nvidia used this 50 suffix with its Geforce FX chips and had cards such as Geforce 5950 and 5750 but this suffix actually meant that the card is bridged to PCIe, as the cards without suffix, Geforce FX 5900 was AGP only.

This card might be a bit more manageable being shorter in length making a quad SLI setup easier to obtain without having to get a new PC case. With a price tag of $800 to $900 per card I can see ton of these in systems around the world, as the short supplied 7900GTX already are selling for $600+. What is another $200 for twice the card?

Source: The Inquirer

Comments

13 Apr 2006 ~ 6:32pm Leonardo Why haven't we seen a dual-core GPU? Or a dual-GPU graphics card? Does it have to do with a data transfer bottleneck to the motherboard?
13 Apr 2006 ~ 9:14pm Khaos It may not be practical from an engineering standpoint. GPU cores tend to be large and complex compared to their CPU bretheren. Manufacturers probably find it more practical to keep them separated as it is easier to layout the PCB/memory and easier to control heat build-up that way.

In the same way that this was needed before we had true dual core CPU's, I think that we will have to see significant die shrinks before we see true dual core GPU's that support all the cutting edge technologies. Otherwise, you're looking at one truly MASSIVE chip, which is crappy for yields. Single GPU's already sport hundreds of millions of transistors in a single chip. Imagine a dual core GPU with over 500 million transistors. That could easily turn into a manufacturing nightmare.

Just guessing!
13 Apr 2006 ~ 9:19pm Leonardo For the sake of argument, let's say that ATI, Nvidia, or Intel were producing GPU cores at the 65nm specification. Do you then think dual-core GPU would be possible? I know you stated that you were approaching the question from an engineering standpoint, but do you have knowledge or opinion concerning data input/out to the motherboard?
13 Apr 2006 ~ 10:11pm Why wouldn't it be possible? We shall see it sooner or later. BUt the price of $800 to $900 is just rediculous.
14 Apr 2006 ~ 6:57pm Sledgehammer70 Here are a few images of the original Quad cards. These bad boys are HUGE!
17 Apr 2006 ~ 12:51pm skywalker45 Jeez those boards are huge. They remind me of PC (AT) Vista graphics cards made by Pixel Innovations. Not something you'd use with the typical home PC though.
18 Apr 2006 ~ 11:25am profdlp

Sledgehammer70 wrote:

Here are a few images of the original Quad cards. These bad boys are HUGE!

Bitchin' Fast!

I have an identical keyboard - that card is darn near as big.
18 Apr 2006 ~ 1:35pm

For the sake of argument, let's say that ATI, Nvidia, or Intel were producing GPU cores at the 65nm specification. Do you then think dual-core GPU would be possible? I know you stated that you were approaching the question from an engineering standpoint, but do you have knowledge or opinion concerning data input/out to the motherboard?

Honestly, I think it has more to do with die size and yields than PCB engineering. The latter can always be worked around by using an extra layer of PCB (6 instead of 5 or 4, etc), whereas the yield problem simply hurts the bottom line.

Think about this: The latest GPUs from both companies sport over 300 million transistors. Assuming some of that architecture would be redundant in a true dual-core chip, you're still pushing 500, maybe even 600 million transistors. On our current manufacturing processes, that results in a simply massive chip. Yield would be terrible, and profit, if there was any to be had, would be less than what it is with the current implementations.

With a dual-chip rather than dual-core design, they can keep the transistor counts high and cram extra power on a board without sacrificing anything in terms of yield.

As for my background, you know I'm no engineer... Just a skeptic like most everyone else here. What I know about circuit board layout and construction as well as processor architecture is what I've picked up over the past six or so years as an enthusiast.

Dual-core GPUs possible? Of course. Financially practical? Not by a long shot.

Like I said in the first post, we'll see true dual-core GPU's when the dies shrink to the point where twice as many transistors are fitting in the same amount of real estate. In other words, 45-65nm (Nvidia is still at 130nm, ATI is at 90nm currently). This is my ballsy prediction based on industry trends that can be observed in other chip manufacturers, such as AMD and Intel. It all has to make dollars and cents in the end! =]

Of course, they could always go in the other direction and add more to single core chips when the die shrinks occur. That's basically what ATI did when they shrunk the die to 90nm and released the R580. All this extra wafer space... And we need to compete with Nvidia. What do we do? Add a bazillion pixel shader units! Yeah!
18 Apr 2006 ~ 1:35pm Oops... That was me. =]
18 Apr 2006 ~ 1:36pm Sigh... I put my name in that time! This is Rob aka Khaos. Maybe I should just log in?!?! wtfffzzxiuisj
18 Apr 2006 ~ 4:34pm Sledgehammer70

wrote:

In other words, 45-65nm (Nvidia is still at 130nm, ATI is at 90nm currently).

Nvidia is making 90nm cards as we speak the 7900 series are all 90nm as ATI already will have a 80nm card in a few short weeks... if all goes to plan.

As to dual core GPU's. I can see it in the future and also see it not happening until 45nm and 65nm tech is created. The reason I can see it is because physics seems to be a bigger rage. now plopping 2 cores 1 GPU the other Physics in one chip could be possible, and increase performance depending on if the market and Dev's follow this new Physic's crave. Now if more power is needed die shrinks will continue to go down and transistors to go up. Which could yield a dual core GPU... but I highly doubt it.

Why do you think Nvidia has focused more on stackable boards in SLI rather than die shrinks? I mean a ton of OEM's have tried dual GPU's with limited success following with no sells to back the research. Look at ASUS and Gigabyte, more issues with dual chips than anything. Now if they can do something like a Core Duo and run a dual GPU at the same wattage and heat ratings as a single GPU than we are talking but one again I doubt this will happen anytime soon or at all. As current CPU’s still can’t feed the need for SLI at 100% you will always become CPU bound unless we change the way all programs run and how certain instructions are dealt with.
19 Apr 2006 ~ 3:22pm Khaos Fair enough, I was basing most of my argument on the 7800 series (Which was all 130nm) since I really haven't kept on top of things since. Been real busy with work.
19 Apr 2006 ~ 3:31pm Sledgehammer70 Also the 7800 GO series is based on the 90Nm tech if I recall correctly