Why don't AMD and nVidia make monster-like GPUs and charge embarressingly high sums?

[-Slacker-]

Like why don't they make GPUs with huge dies (1000+ mm^2) for the rich enthusiasts and charge thousands of $$$ for them?

I know that, normally, the R&D costs for developing a new architecture/ series- what with the careful designing, rigorous testing and bug fixing, maximizing the efficiency of the cip etc - would outweigh the profits, but I'm not talking about that; I'm thinking more between the lines of throwing up a design in a hurry, like double up everything on the chip that won't be completely redundant and then slap a huge ass cooler+heatsink and 3 8 pin connectors.

I'm sure that there are enough rich nutters out there that would pay $5k~10k for something like this, so there would be good profit margins to be made.

 

[busydude]

Like why don't they make GPUs with huge dies (1000+ mm^2) for the rich enthusiasts and charge thousands of $$$ for them?

They are fabless companies.

 

[frowertr]

Like why don't they make GPUs with huge dies (1000+ mm^2) for the rich enthusiasts and charge thousands of $$$ for them?

I know that, normally, the R&D costs for developing a new architecture/ series- what with the careful designing, rigorous testing and bug fixing, maximizing the efficiency of the cip etc - would outweigh the profits, but I'm not talking about that; I'm thinking more between the lines of throwing up a design in a hurry, like double up everything on the chip that won't be completely redundant and then slap a huge ass cooler+heatsink and 3 8 pin connectors.

I'm sure that there are enough rich nutters out there that would pay $5k~10k for something like this, so there would be good profit margins to be made.

No market here or it would already have been done. Secondly, how would one run such a thing? Power demands, heat, space, size???

Logistically nightmare if you ask me. And who is going to buy it? I can't imagine anything of the like being made without a purchasing contract already penned and signed to deliver x amounts of units. No one is going to stick their neck out to build something like this with literally ZERO demand warranted.

My .02¢

 

[aigomorla]

Like why don't they make GPUs with huge dies (1000+ mm^2) for the rich enthusiasts and charge thousands of $$$ for them?

Did u miss the shift in industry?

Efficiency typcially means smaller die.. which also means smaller node process.
Which also means not making them that big of a monolith.

Here apple is thinking of ditching intel in lue of arm, and u want the gpu makers to go larger dies?

 

[happy medium]

profit? Not many people could afford them.

I much rather have 10 tegra 3 type chips running together using a whole 6 watts and costing 180$.

 

[-Slacker-]

Ah, so it's something along the lines of TSMC having to approve the design on the basis of weather or not the product will be commercially successful.

In that case, why can't amd/nvidia just ask them to make the chips and tell them "look, we'll pay you for them off hand, weather we make a profit or not, you just worry about manufacturing them"?


edit: to the happy and aigomorla: but I'm only talking about a small, enthusiast consumer base, like a few thousand people.

 

[dma0991]

I can only imagine that the GPU would require a secondary 1kW+ PSU for the GPU alone and phase change for cooling that beast.

 

[Arkadrel]

Like why don't they make GPUs with huge dies (1000+ mm^2) for the rich enthusiasts and charge thousands of $$$ for them?

I know that, normally, the R&D costs for developing a new architecture/ series- what with the careful designing, rigorous testing and bug fixing, maximizing the efficiency of the cip etc - would outweigh the profits, but I'm not talking about that; I'm thinking more between the lines of throwing up a design in a hurry, like double up everything on the chip that won't be completely redundant and then slap a huge ass cooler+heatsink and 3 8 pin connectors.

I'm sure that there are enough rich nutters out there that would pay $5k~10k for something like this, so there would be good profit margins to be made.

Because the fabs like GF and TSMC, only make chips upwards of 600mm^2 or something.

Because the bigger the chip, the easier the flaws, the more problems, the bigger the costs pr chips (it raises exponentially). Nvidia with its 529mm^2 chips are close to the edge.


Also you know how much power a 580 uses.....
Avg load of 200watts used, and upwards of 320watts at TPD maximum.

Now imagine a 1000mm^2 chip.... that would probably be like 600+ watts.



Also are you forgetting the 590? or the 6990?
The fact that you can SLI or Crossfire them (smaller chips), means you dont need big Single chips, its wastefull, too little profit in a product like it.

 

[Lonbjerg]

Also are you forgetting the 590? or the 6990?
The fact that you can SLI or Crossfire them, means you dont need big Single chips, its wastefull, too little profit in a product like it.

That is a statement with modification.
Multi-GPU brings along it's own set of added probblems/issues...it's not the "golden path".

 

[aigomorla]

edit: to the happy and aigomorla: but I'm only talking about a small, enthusiast consumer base, like a few thousand people.

lol...

3 580GTX + 1 Gulftown on a Classified already pushes the 1kw limit on a psu.

imagine the die's on the gpu your talking about.

We dont have a single psu powerful enough for that.

 

[smackababy]

I would say that if they are significant advances in GPGPU technology, we might see something like this for large companies. Besides, I'm sure anyone who really "needs" one has the cash to entice someone to make a custom, one-off model for them.

 

[cusideabelincoln]

They wouldn't make a profit off such extreme extremes. The market size just isn't big enough. It's wiser for them to invest their resources in the more profitable sectors, because it would take a lot of resources (time, money, personnel) just to do what you are proposing.

 

[-Slacker-]

That is a statement with modification.
Multi-GPU brings along it's own set of added probblems/issues...it's not the "golden path".

This. That's why I brought it up. I would think that rich people are too rich for micro stutter etc

lol...

3 580GTX + 1 Gulftown on a Classified already pushes the 1kw limit on a psu.

imagine the die's on the gpu your talking about.

We dont have a single psu powerful enough for that.

I realize that TDP doesn't increase linearly with increasing the size of the die, but would something like a 1500 watt psu really be stumped by a card like this?

They wouldn't make a profit off such extreme extremes. The market size just isn't big enough. It's wiser for them to invest their resources in the more profitable sectors, because it would take a lot of resources (time, money, personnel) just to do what you are proposing.

I'm beginning to see your and the others' point... I guess the main thing I was curious about was how much could they charge for a gaming video card and get away with it, but I suppose there's a point where even rich hardware nuts loose interest.

 

[Idontcare]

Like why don't they make GPUs with huge dies (1000+ mm^2) for the rich enthusiasts and charge thousands of $$$ for them?

I know that, normally, the R&D costs for developing a new architecture/ series- what with the careful designing, rigorous testing and bug fixing, maximizing the efficiency of the cip etc - would outweigh the profits, but I'm not talking about that; I'm thinking more between the lines of throwing up a design in a hurry, like double up everything on the chip that won't be completely redundant and then slap a huge ass cooler+heatsink and 3 8 pin connectors.

I'm sure that there are enough rich nutters out there that would pay $5k~10k for something like this, so there would be good profit margins to be made.

The physical limit to die-sizes are around 850-900mm^2. The exposure field of the litho equipment is around 30mm x 30mm.

Ah, so it's something along the lines of TSMC having to approve the design on the basis of weather or not the product will be commercially successful.

In that case, why can't amd/nvidia just ask them to make the chips and tell them "look, we'll pay you for them off hand, weather we make a profit or not, you just worry about manufacturing them"?


edit: to the happy and aigomorla: but I'm only talking about a small, enthusiast consumer base, like a few thousand people.

TSMC does not "approve designs" on the basis of their expectations of market viability. They'll build anything you are willing to pay for. You are their customer when you order the wafers, whether or not you can sell them for a profit is your problem (provided your check clears the bank ).

The short (self-evident) answer to your proposition is that there is clearly not enough money to be made in that part of the supply/demand curve. If this were not the case then to be sure Nvidia or AMD would produce a product to fill that niche.

Same reason Intel doesn't release a $10k uber-extreme CPU, or LCD manufactureres don't release a monolithic 10240x6400 50" LCD for $50k.

 

[SlowSpyder]

The desktop is dead, or at least dying. Chips that are made for the desktop have to be able to be used in the mobile space or in the professional market. I would imagine a chip that needs 800 watts (just pulling a number out of my butt) would cause some issues there in all but the most specialized markets.

Not to mention, how many games really push a GTX580 on a single monitor now? How much GPU horsepower do we need to push an XBox 360 port at 1080P?

 

[bunnyfubbles]

some people just never bothered to take economics 101 it seems

 

[smackababy]

LCD manufactureres don't release a monolithic 10240x6400 50" LCD for $50k.

Actually, you can buy some monolithic screens. I know there has been a 92" plasma screen for like $35k and there have been a few others as of late. Look at Charlotte Motor Speedway and Cowboys Stadium. If you have the money, someone will build it for you.

 

[Kenmitch]

I'd think it's just a matter of time till we see quadfire on a single card. Not sure if it's doable at 28nm but I'd think 22nm would be possible. Guess it would be $1000 plus card.

 

[-Slacker-]

The desktop is dead, or at least dying.

Depressing as hell...

 

[dma0991]

I'd think it's just a matter of time till we see quadfire on a single card. Not sure if it's doable at 28nm but I'd think 22nm would be possible. Guess it would be $1000 plus card.

I can't even imagine how would a Quadfire card fit into any case at all, not to mention the flex of the PCB from that length.

 

[happy medium]

Depressing as hell...

I agree, but I'm afraid Slowspider is right.

The personal computer is becomming even more personal. Your gonna carry it around in your pocket in a few short years.

If they could only make a program to encode movies on a smart phone and burn them with a external dvd drive. Thats about the only thing missing.

Smart phones allready have big screen hdmi output, mouse/keyboard ability, web browsing, and gamming all rapped up.

The first pocket device with usb 3.0 output wins.
Then you would just need a USB hub to connect to all your usb devices, including your printer.

 

[happy medium]

I can't even imagine how would a Quadfire card fit into any case at all, not to mention the flex of the PCB from that length

Dual pcb would work, just like the gtx295.

 

[Lonbjerg]

The physical limit to die-sizes are around 850-900mm^2..

And if designing such a large (or larger) die, one must also factor in the physics of electrons.
They move at ~65% of c in a chip, meaning that at a certain size a signal will take a large chunk of time comming from one end og the die to the other side, relative to the clock frequency (we are talking several "ticks"...)

A "cube" would in that regrad be better than a large sqaure die...but that added dimension bring forth a hole new subset of issues from heat dissapation to logic design.

I really think we should enjoy the current pace of node advancement, we might not be so lucky in 20 years...

 

[gevorg]

Quad SLI is not enough?

 

[Lonbjerg]

Quad SLI is not enough?

You think that quad-SLI is "zenith" of performance?

I hope in 10 years time that we can look back at the perfomance of quad-SLI today and think:
"Dang...and I thought that was fast..how time has moved on"