Is the era of cheap GPUs over?

sdlvx

Gawd
Joined
Dec 9, 2007
Messages
567
I stumbled upon this article today. It has slides from NV basically saying that 28nm is more expensive than 45nm. In the past, we've always had smaller nodes cost less. Now, we're seeing a reversal in trend and it looks like if things keep going the way they are, GPUs will always cost more and more.

But, knowing this, everything makes sense. It explains why big Kepler isn't here and why it might never be feasible at a reasonable consumer price. It explains why NV has made the sudden shift from big, powerful dies to smaller, more efficient ones. It also explains why AMD never plans on releasing a big die. Most important of all, it explains why 7970 is $550 and GTX 680 is $499 when the both have die sizes which would put them in the mid-range market segment.

http://www.extremetech.com/computin...y-with-tsmc-claims-22nm-essentially-worthless

What do you guys think? I think it means that the era of cheap 300mm GPUs is done. Also that we're done pushing for lower nodes. Why would anyone want to move to a lower node if it's going to cost more? And do you think this is a TSMC thing or a laws of physics thing? It doesn't seem like Intel is having problems, but a CPU has far fewer transistors.
 
I think no single GPU reference card should cost more than $350 or $400 max, being so expensive will only make people lean more towards consoles, which as a result will lead to more big titles being developed mainly for the consoles while the PC get bad parts and never fully utilize the powerful and expensive hardware available for 'em.
 
I don't see GPU's, to date, as more expensive then they've been last 10 years, considering some inflation. Also - something to note - if you don't need AA and all the sampling stuff, a $150 - $250 card will handle 1080p and/or 1900x1200. That's reasonable... I remember paying $400-$500 to get the same performance at 1024 resolution. X1900XT anyone?
 
I don't think there was an "era" of cheap GPUs. It tends to fluctuate from generation to generation and a lot of it depends on the level of competition.
 
I don't think we've ever been in a cheap GPU era. When I purchased my NVidia TNT2 Ultra about 10 years ago or so, I thought it was so expensive at $299. That was their flagship model at the time. The mainstream cards were a lot less. We're talking $50-$100 here. Today, those would be considered low end. (Also, it was a one slot card, which is something I miss. I don't have nearly enough PCI-E slots for what I want to do with the PC).

Prices rise, technology increases. Sooner or later they'll drop again.
 
Let 'em have their $500 halo GPUs. The $200-$300 midrange cards are where the majority of the business is from. As for single slot cards, they need to introduce a single slot GPU that uses an integrated waterblock (similar to the Corsair WCs). They need to bring back high performance single slot cards!
 
Didn't nvidia try to go up to around $650 for the 280 gpu before getting hammered back down by AMD?

If anything, prices are pretty stable.
 
Didn't nvidia try to go up to around $650 for the 280 gpu before getting hammered back down by AMD?

If anything, prices are pretty stable.

Yeah. AMD tried to creep up with the $550 7970 but with the 680 being priced at $499 I think we will see that come back down. ~$499 has been the norm for the top-end cards at release for awhile now.
 
It doesn't seem like Intel is having problems, but a CPU has far fewer transistors.

People with their hands on an Ivy Bridge have been reporting that it runs hotter and does not overclock as well as Sandy Bridge. That might be due to the tri-gate transistors, but I believe that is directly related into their new process.

Another thing about the increased prices on new cards is that the old cards continue to be amazing deals. You can pick up a used 5970 and get nearly 7970 performance. A 6870x2 sells for the same price as a 7870 (~$350) but offers performance that often surpasses a 7970. Most people looking for a mid-range card are still going with the 6870 over the 7770.
 
I just see some inevitable slowdowns as it gets more difficult to process shrink at each step as we approach get closer to physical limits.

You simply can't count on 2X transistor count every 1.5 years anymore. Cost reductions will be slower.
 
My point is that 6970 was a 389mm chip and MSRPed at under $400. GTX 680 is nearly 100mm smaller and MSRPs for $499.

The problem is that GTX 580 was a very big chip and it cost as much as the ~300mm chip. That's the point I'm trying to make. If NV and AMD really aren't price fixing and the problem is the cost of a TSMC wafer, that means a big Kepler should realistically cost $700 to $800, and that price won't go down.

NV-Pres4.jpg


That's the graph that concerns me the most.
 
Nvidia opted to aggressively optimize GK104 precisely because the old strategy of bolting on more cores and ratcheting up transistor counts isn’t sustainable.

Gee, that's why we don't have a GK110 that consumes 300W and generates boatloads of heat. It's not that nVidia wanted to save power, they had to.
 
My point is that 6970 was a 389mm chip and MSRPed at under $400. GTX 680 is nearly 100mm smaller and MSRPs for $499.

[...]

NV-Pres4.jpg


That's the graph that concerns me the most.

You negated your own point with that graph. 6970 was a 40nm part. 680 is a 28nm part. It's not the same feature size. And like that graph shows, as the feature size gets smaller, the price goes up. Eventually it levels out a bit, but it seems like overall the prices are still rising.
 
Why are there no cheap sports cars? The top end is a bad place to look for something cheap. But if you wait for the generation to be over you can usually pick up a good deal on "old" hardware.
 
My point is that 6970 was a 389mm chip and MSRPed at under $400. GTX 680 is nearly 100mm smaller and MSRPs for $499.

The problem is that GTX 580 was a very big chip and it cost as much as the ~300mm chip. That's the point I'm trying to make. If NV and AMD really aren't price fixing and the problem is the cost of a TSMC wafer, that means a big Kepler should realistically cost $700 to $800, and that price won't go down.


That's the graph that concerns me the most.

The price for a graphics card is heavily market driven. Both are priced at what they think they can sell for.

Beyond that look up Moore's second law. It states that going forward, the price for semiconductor fabs increase exponentially.

These factors are coming to head this decade, so you can expect a slowdown in easy gains. But there will still be gains and progress.
 
I paid $665 for an 8800GTX on launch day with overnight shipping and it was worth every penny.

On the other hand I'm now rocking a used GTX 570 that I got right here on [H] 3.5 months ago for the spectacular price of $265 delivered, and I am very happy with it.

Shrug.
 
Without fail, one of these threads (almost with the exact same title) pops up as soon as a flagship model is released.

Prices for top end graphics cards have been about the same for ages. Do you remember how much a Voodoo setup was? Do you remember how much a 5900 was, or a 9700pro was at release? Should we go on? X850 and 6800 ultra? 7800/7900gt's? 8800 Ultra?.. They were all released around the same price-point.
 
You negated your own point with that graph. 6970 was a 40nm part. 680 is a 28nm part. It's not the same feature size. And like that graph shows, as the feature size gets smaller, the price goes up. Eventually it levels out a bit, but it seems like overall the prices are still rising.

Prices take longer to drop down to beneath previous-node-levels, meaning higher prices for newer gen products will be likely to stick around longer. Frankly, I think he's right about GPUs being more expensive. My guess is that the 450mm transition would be the only wild card here as far as bringing prices back down, but we're already quite close to hitting the proverbial wall

NV-Pres3.jpg


We're already doing FinFETs at 14nm (maybe even 20nm) and past that nobody's really sure just what the hell to do. Well, at least publicly. Privately they probably have their own workarounds but 1) it will be expensive and 2) it'll take a while.
 
Last edited:
I think the GTX680 has a really solid price point. 499 for a brand new 28nm chip that goes toe to toe with the 7970 while besting it in price, power, and heat.
 
I think my Radeon 9800 pro cost $600+ when I bought it at release.

That was the last flagship card I bought actually. I found it didn't last that much longer than the midrange and I could upgrade twice as often for the same price.
 
I think my Radeon 9800 pro cost $600+ when I bought it at release.

That was the last flagship card I bought actually. I found it didn't last that much longer than the midrange and I could upgrade twice as often for the same price.

I'm pretty sure I paid $399 or so on release for the AIW 9800 Pro...

Maybe $499. Not sure. Wasn't any $600+ though.
 
To all of you saying you paid a lot for your 8800 GTXs and such. My point is that those were big chips. G80 was 480mm. GK104 is 294mm. That's a huge difference in die size, and die size affects cost. If you're paying $500 for a chip that's a lot smaller than a $600 chip, something is boned. Everyone who is comparing this generations high end to previous ones are completely neglecting die size.

Show me a generation when Nvidia had a $500 part that was their high end and had a die size under 300mm. There is none. That's the point I'm trying to make. Of course a ~500mm chip is going to cost a lot of money on a cheap process, it's a huge chip. We've reached a point where it's not worth it to make a big chip and the smaller ones are replacing the spot which would have been taken by the high end chips when they should be mid range size.
 
i think flagship single gpu cards around 500 is reasonable. however mid range cards could be cheaper.
 
I think the prices will come down. As others mention every time you shrink the die and go through a new ramp there is a lot of overhead cost associated with that. Once the ramp is up and the wafers are flowing the costs come down. I would predict that around 15-20nm there is no more benefit to squeeze out of shrinking beyond that. Things will still get faster, but it will be from another technology nano-processors, organics etc...
 
There was a brief period of 'cheaper' GPUs when ATi/AMD decided they had to get more market share and dropped their prices a fair bit. They did that because they had to, not because they wanted to.

What most people here seem to be forgetting, is that the price of the card isn't just the manufacturing, it's the R&D that goes into all the nice little bells and whistles that all the rabid fanboys scream for every time a new card gets released. Do you think these people work out of the goodness of their hearts?

When you look at what a modern GPU can do, what you're paying is really bugger all. Stop whinging that everything should be cheaper and realise that these are businesses who need to make money to pay workers and shareholders. Without these, you wouldn't have anything.
 
adjusted for inflation the 9700 pro which launched in 2002 at $399 would now cost $504 todays dollars.
 
This topic has been beaten to death.
There are supply and demand issues as well as "perceived value". As long as the product is perceived as a $500 part, it will sell at $500. The card itself could cost $200 to make, but perform as well as the competition's product, it will be priced accordingly.
 
Show me a generation when Nvidia had a $500 part that was their high end and had a die size under 300mm. There is none. That's the point I'm trying to make. Of course a ~500mm chip is going to cost a lot of money on a cheap process, it's a huge chip. We've reached a point where it's not worth it to make a big chip and the smaller ones are replacing the spot which would have been taken by the high end chips when they should be mid range size.

I agree completely.

Welcome to the ERA of new Mid range GPU's! Look mama they save 20 watts!
 
I spent quite a bit for my geforce 2 Ultra about a decade ago. Over $500.

http://www.guru3d.com/review/guillemot/3dprophet2ultra/

I guess if you consider dual GPU cards, then yeah cards have gone up a bit.

Awwww... That link brought back some memories... I remember drooling over that bad boy when I was in the 7th grade.

For all of you who want to walk down memory lane with me, here is my video card upgrade path over the years:

1. Voodoo 2 8MB
2. Voodoo 5 5500 (ugh...)
3. Geforce 3 (regular)
4. Geforce 4 4600ti
5. Geforce 7800gt
6. Geforce 8800gt
7. Geforce GTX 460 768mb, which is what I am running now. It's been a fun ride. Maybe the GTX 680 is in my future
 
I remember picking up my first GPU and installing it myself. A Voodoo 3 AGP at Best Buy for 179.99 and it maxed out Half Life, Counter Strike, and Team Fortress as well as Unreal Tournament and Quake 2. The Voodoo 3 AGP basically had the same performance as Voodoo 2 SLI in one chip. Those Golden days of PC gaming aren't coming back at that kind of price to performance ever. Glad I was there. :)
 
I've never paid more than $300 for a gpu
Granted I never bought top of the line.

-I had a Diamond Monster Stealth64, which I got for 170ish.
-Then I paired it with a Diamond Moster3D 3DFX which was around the same price
-From there I went to a 3DFX Voodoo2 for something like $200
-Then the Voodoo 3 3000 replaced both for $249 (seeing a pattern here?)
-From there I went to a Geforce 2 MX, then Geforce 2 GTS, then Geforce 2 Pro(Free upgrades)
-Then a Geforce 3 ti 200 for awhile, again $200
-Then a Radeon 9700 pro for under $200, I bought it when the 9800pro came out.
-From there I skipped a few generations and went with dual Nvidia 6600gt's, I bought one for $230ish, won one during an AMD tech tour
-After that a friend sold me his dual 7800gtx setup for $250 as the 9xxx series was about to be released.
-Replaced those with a single 9800GTX that my brother gave me(He upgraded to the newly released Nvidia 280 line)

And thats where I stand. Looking to upgrade but like hell am I paying $500 for a new card. My mobo+CPU+RAM don't even cost that combined. 200-250 is my sweet spot.
 
I usually buy the best single gpu card I can afford and skip a generation.

At the beginning of the century I was using a Voodoo3-2000 and upgraded to a 9800 Non Pro. From that I went to a X1900XT, then a 3850 (garbage card) followed by a 4980 and then finally the current 6970.

Saves more money in the long run while keeping the gaming experience consistently good.

edit: but cheap gpu eras? not really, but fast gpus do get cheaper over time
 
PC gamer and next console I'm done with pcs for awile. It's too expensive too keep up. Guys been playing years for only $300. If it wernt for starcraft which I play every day I'd be done already.

If they bring out Planescape Torment 2 I'll come back though.
 
Last edited:
sdlvx, you have my sympathy. There's so much reading comprehension fail in this thread that it's ridiculous.

The argument being made is not that either the 7970 or the 680 are $500+ cards, but that their GPU die size is virtually the same as the mid range ones of previous generations, yet they're being sold as the upper end due to fab costs. The worry here is that larger chips such as those 480mm in size or so are so expensive that they can't fit into the market.

As for the original question, I'm not overly surprised it's starting to come this. I've been somewhat concerned as to when Moore's Law will begin to hit a brick wall, even if it's temporary, towards price and performance. I think it's somewhat to be expected, sadly.
 
In any case these prices are showing a reinvigoration of PC gaming, which is good.
Prices are a little outrageus though I will admit that, I remember back in they day when Geforce 3 was first realeased (first DX 8 card) it was only like 300 bucks, and it was top of the line.
Obviously what were seeing now is a war between the red and the greens, and they both know perfectly well they can charge these insane prices cause PCs are currently a superior gaming platform when compared to 6 year old tech consoles.
 
Back
Top