Why doesn't AMD start producing/ramp up old generations again?

[wand3r3r]

I've been wondering why they don't ramp up (some at first) production on old GPUs again. I would bet that 58xx series, as well as 69xx series would fly off the shelves, at say the $200 range.

A couple things immediately come to mind, die size being one, but you'd think with all the craze on the new Fabs (and rushing to switch over) that an old one would stay in business since they have all the existing infrastructure.

Couldn't they also (basically) start producing the old dies on the new processes with a little reworking, or is it too difficult for $200 GPUs.

The mining fad has been going pretty strong for a few years and has outlived all expectations. Obviously it's impossible to predict the future, but it looks good. Yeah there could be a flood of old cards on the market, but one thing is that they've already successfully raised the high end from $380 -> $550-1k.

I suppose the only people qualified to actually comment on this situation are under NDAs.

Thoughts?

 

[Ancalagon44]

Well, why would they bother? Nobody is going to buy a 5xxx GPU instead of a 7xxx GPU, or Rxxx whatever they call it these days.

Moving an old die to a new process requires a significant amount of engineering work, so they wont bother. Look how few GPUs have ever done that - off the top of my head, I think only Nvidia has ever die shrunk a GPU. I dont think AMD ever has. That should tell you that its not cost effective.

At the moment, mining is too much of a fad for AMD to dedicate that much resources to it. If they sell only a few high end GPUs, imagine how few high end mining GPUs they will sell.

 

[Galatian]

I'm puzzled...why would they even consider releasing slower GPU again?

Also I think AMD knows very well how volatile the whole mining rush is right now. If it comes to a crash for whatever number of reasons, current gen AMD cards will be flooding Ebay and thus severely lowering profit on their newer models if and when they come out. This is a speculation but from that point of few it makes sense for AMD to actually not produce more chips, because they are speculating that cryptocoins will crash eventually.

 

[Red Hawk]

Exactly what Ancalagon44 said. Why bother? In order to make it competitive with AMD's current GCN-based product stack, you'd have to shrink the old chips down to 28nm, an expensive proposition, taking away resources that should be going into Volcanic Islands and Steamroller. There's still a lot of used 6000/5000 series cards currently on the market, which are much cheaper than the price AMD would be willing to sell a rerelease of those chips at. Even if AMD did put them back on the market, what's the appeal? GCN cards are better with GPU compute, tessellation, software feature support, etc. There's a reason that AMD moved on to GCN in the first place.

If the main motivation behind this would be to target the crytocoin mining community, it's misguided. GCN based cards already have plenty of appeal to that community, and the older cards won't remedy the competition that ASIC boxes represent any more than the GCN cards do.

 

[wand3r3r]

Yeah, it just made me wonder if it could be done fairly easily and without any major costs (utilize existing designs, fabs etc.). Apparently you guys think it would be not only difficult, but expensive.

I am sure they would sell at the right price, especially when the new cards are selling out instantly (anywhere near retail).

 

[Shmee]

I would think they would rather just up production of current GPUs, and are having a hard enough time with that.

 

[SPBHM]

it's best for them to have all their card from the same gen, architecture, in terms of optimization and support (think things like mantle, a new card with no support for it is... strange),

realistically a 5870 makes no sense, the card uses a lot of power and it's a complex thing to make (needs a good pcb/power circuits, big die with 256bit GDDR5), when you consider a simple 260x is superior for newer games (low power, half the die size, half the memory bus).

you could say the same about 6900 vs r9 270, 6850 against R7 250x (7770) and so on... the cost is higher with the older tech, profit margin would be much slower I think.

 

[ShintaiDK]

It makes no sense in any way to produce older generations. Not to mention you would have to extend support as well, both hardware and software. And deal with multiple masks with lower volumes. It would be an economic disaster. AMD doesnt exactly make much money on GPUs already.

 

[Arkadrel]

I've been wondering why they don't ramp up (some at first) production on old GPUs again. I would bet that 58xx series, as well as 69xx series would fly off the shelves, at say the $200 range.

A couple things immediately come to mind, die size being one, but you'd think with all the craze on the new Fabs (and rushing to switch over) that an old one would stay in business since they have all the existing infrastructure.

Couldn't they also (basically) start producing the old dies on the new processes with a little reworking, or is it too difficult for $200 GPUs.

The mining fad has been going pretty strong for a few years and has outlived all expectations. Obviously it's impossible to predict the future, but it looks good. Yeah there could be a flood of old cards on the market, but one thing is that they've already successfully raised the high end from $380 -> $550-1k.

I suppose the only people qualified to actually comment on this situation are under NDAs.

Thoughts?

260x die size = 160mm^2 (on 28nm)
6970 die size = 389mm^2 (on 40nm)


Something like the 6970 would probably compete against a 260x.

I dont see whats to be gained by going backwards, even if they use a cheaper 40nm fab than the 28nm ones.

The die size difference probably eats up the differnce in saveings from useing 40nm.

 

[Red Hawk]

Just look at the benches.

http://www.anandtech.com/bench/product/1043?vs=1062

http://www.anandtech.com/bench/product/1043?vs=1061

Pitcairn, currently AMD's third tier chip, trounces both Cypress and Cayman.

 

[parvadomus]

260x die size = 160mm^2 (on 28nm)
6970 die size = 389mm^2 (on 40nm)


Something like the 6970 would probably compete against a 260x.

I dont see whats to be gained by going backwards, even if they use a cheaper 40nm fab than the 28nm ones.

The die size difference probably eats up the differnce in saveings from useing 40nm.

And whats the mining rate of the 260X compared to 6970. Also what is the difference in transistor count?.

 

[Red Hawk]

The mining rate of a 6970 is likely higher than that of a 260X, but that advantage is probably negated by the higher power consumption of the 6970.

 

[Ancalagon44]

Maybe a better question is why doesnt AMD design and build a GPU without the graphics rendering components? Ie no texture pipelines or raster output units?

Lacking those functional units, it would be more efficient in terms of cost to produce than a GPU part that can also function as a GPGPU unit. And would be very good for mining, amongst other things.

Perhaps AMD considered the possibility but decided that designing an entire GPU just for GPGPU was too risky/costly? It should be noted that as far as I remember, most of Nvidia's revenue comes from GPGPU products, such as Tesla.

 

[Red Hawk]

Maybe a better question is why doesnt AMD design and build a GPU without the graphics rendering components? Ie no texture pipelines or raster output units?

Lacking those functional units, it would be more efficient in terms of cost to produce than a GPU part that can also function as a GPGPU unit. And would be very good for mining, amongst other things.

Perhaps AMD considered the possibility but decided that designing an entire GPU just for GPGPU was too risky/costly? It should be noted that as far as I remember, most of Nvidia's revenue comes from GPGPU products, such as Tesla.

Wouldn't that put it in competition with not only Tesla, but also Intel's Xeon Phi?

 

[JBT]

They already do 270, 270x, 280x are all last gen. I'm sure there are a few cards in the 240 range that are too. The thing is these are cheaper to produce due to die size, why make expensive cards that don't perform well?

 

[f1sherman]

don't forget that shrinks are neither cheap nor straightforward

and AMD already has shortages plaguing it's supply line

 

[monkeydelmagico]

I've wondered the same thing. Many of their older CPU and GPU technologies are still relevent.

For example, VW produced the Beetle from 1938 until 2003. They just kept moving production for a variety of reasons to chase the profits.

I think the difference is AMD doesn't have a buyer for the old nodes/processes/machinery or cannot find an even less expensive place to produce the older designs. If there was a buck to be made you can be sure they would try.

 

[Red Hawk]

I've wondered the same thing. Many of their older CPU and GPU technologies are still relevent.

For example, VW produced the Beetle from 1938 until 2003. They just kept moving production for a variety of reasons to chase the profits.

I think the difference is AMD doesn't have a buyer for the old nodes/processes/machinery or cannot find an even less expensive place to produce the older designs. If there was a buck to be made you can be sure they would try.

That is a horrible, horrible analogy.

 

[SPBHM]

I've wondered the same thing. Many of their older CPU and GPU technologies are still relevent.

For example, VW produced the Beetle from 1938 until 2003. They just kept moving production for a variety of reasons to chase the profits.

OT:
it was never the same, you can go decade by decade and find 100s of parts have changed,

38 and 2003 are hardly the same car, good luck trying to interchange parts

http://allthecars.files.wordpress.com/2009/06/volkswagen-fusca-2003-ultima-edicion-02.jpg
http://image.motortrend.com/f/featu...32179712/1938-volkswagen-beetle-rear-view.jpg

http://allthecars.files.wordpress.com/2009/06/volkswagen-fusca-2003-ultima-edicion-04.jpg
http://preview.netcarshow.com/Volkswagen-Beetle-1938-1600-1a.jpg

also, the basic needs of a car changed less

 

[TeknoBug]

Well Nvidia does it with their GT610, 630, 640, etc. garbage cards.

lol @ comparing a 1938 Beetle vs a 2003...

 

[Subyman]

If they have resources to produce older cards, why not use those resources to produce newer cards?

 

[Maximilian]

Maybe a better question is why doesnt AMD design and build a GPU without the graphics rendering components? Ie no texture pipelines or raster output units?

Lacking those functional units, it would be more efficient in terms of cost to produce than a GPU part that can also function as a GPGPU unit. And would be very good for mining, amongst other things.

Perhaps AMD considered the possibility but decided that designing an entire GPU just for GPGPU was too risky/costly? It should be noted that as far as I remember, most of Nvidia's revenue comes from GPGPU products, such as Tesla.

Miners may avoid something like this because the resale value of them would suck. Depends on the price and performance they could output though.

Part of the reason mining is attractive to some is even if your chosen coin completely tanks and ends up worthless you've still got a bunch of graphics cards to sell so the losses arent that huge

 

[SlowSpyder]

The 5870 holds up pretty well in modern games (compared to it's contemporaries, despite all the doom and gloom about tessellation). But I don't see why they'd produce them when they already have the money spent on the new GPU's. Also, they're pushing GCN pretty a hard.