1 Billion transistors and beyond...future of ati & nvidia

[Junkstyle]

Look at any gpu chart and see the number of transistors climb. With ATI already having 380 million transistors on there current chip and Nvidia rumored to have 700 million transistors on the g80, 1 billion transisitors is just around the corner. How will nvidia cope with this? Do they expect TSMC to be able to fab these insane counts? ATI has a way out. They have AMD which could start using its soon to be old 65nm fabs to do 1 billion transistor count chips. Just not sure about nvidia.

If transistor counts hold steady (which i doubt) and clocking goes up instead, we have even an harder problem. Neither ATI and Nvidia have never been able to make super high clocked chips such as CPU's. From my understanding going from 500Mhz to 2Ghz required lots of engineers and exotic custom gate desgin. ATI again has a way out. AMD engineers could be recruited on ATI's gpu design to get its clock higher.

I'm just not seeing good things for Nvidia. I've been a fan of nvidia and have made tons of money off its stock, but right now unless Intel buys them (which i highly doubt) I'm very worried about Nvidia's long term survival (btw I do not own any nvda stock at this time). Intel also has been rumored to have 2000 engineers working on graphics technologies http://www.smartmoney.com/Techsmart/index.cfm?story=20061020&afl=yahoo
Whereas Nvida is to be rumored to have an x86 design team working on an integrated cpu/gpu chip. I doubt Nvidia's ability to get a cpu into the 2Ghz range. (theinquirer..take it with salt)

ATI's survival will depend on an AMD executive decision on how they want to pursue the high end graphics market. If AMD cuts off ATI's resources for high end chips its over for them.

I hope we have Nvidia vs. ATI for many years to come. Consumers have really benefited from the competition. But right now we are entering uncharted waters.

 

[Skrying]

You're completely wrong if you think ATi has a way out. AMD is not just going to start giving them massive amounts of fab space. hell AMD doesnt have any space to give. As AMD has scored a big contract with with Dell and then just in general increase in orders from their already their OEMs. Sure, they have a few fabs coming in, but that's going to be needed to keep up with Intel.

Frankly, I dont imagine any time soon, if ever, that we'll see high end ATi graphics chips being made at an AMD fab.

 

[Junkstyle]

You're completely wrong if you think ATi has a way out. AMD is not just going to start giving them massive amounts of fab space. hell AMD doesnt have any space to give. As AMD has scored a big contract with with Dell and then just in general increase in orders from their already their OEMs. Sure, they have a few fabs coming in, but that's going to be needed to keep up with Intel.

Frankly, I dont imagine any time soon, if ever, that we'll see high end ATi graphics chips being made at an AMD fab.

I doubt I'm wrong about that part. However, I think you misunderstand the timeframe. I am thinking a billion transistors will come toward the end of 2008 or later. AMD cant keep using 65nm fab process when intel is on 45nm. So some part of some old fab is going to make gpu's..2008? 2009? AMD is eventually going to make gpu's--and thats not anything to even argue about.

 

[Skrying]

I doubt I'm wrong about that part. However, I think you misunderstand the timeframe. I am thinking a billion transistors will come toward the end of 2008 or later. AMD cant keep using 65nm fab process when intel is on 45nm. So some part of some old fab is going to make gpu's..2008? 2009? AMD is eventually going to make gpu's--and thats not anything to even argue about.

Uhh, at the time frame we're on right now AMD will be using 65nm while Intel is on 45nm. Just like AMD is right now on 90nm while Intel is on 65nm. AMD is behind Intel by a decent amount when it comes to process they are using.

Yes, AMD will make GPUs. The ones that are inside their chipsets....

 

[Junkstyle]

So you feel AMD will never make high end GPU's in their fabs. Interesting take.

 

[BByte]

Do they expect TSMC to be able to fab these insane counts?

The short answer: yes. Why not?

There's a reason GPU clocks are lower than CPU clocks, they have very different performance characteristics. If getting a GPU to 2GHz was so easy why hasn't AMD or Intel done it yet? Curiously enough Intel's integrated GPUs have core clocks similar to the discrete GPUs.

Sure, ATI will probably (hopefully) be able to leverage some of AMDs expertise in GPU designs. Fab space is more tricky, at least in short term and TSMC & co. aren't that far behind (if at all). But is it enough to predict doom of Nvidia, before we've even seen combined AMD+ATI in action? I don't think so.

 

[Megadrive1988]

ATI has a 500M+ transistor GPU (R600) for late 2006 or early 2007

Nvidia supposedly has a 700M transistor G80 card, split over two chips, for late 2006. if the 700M transistor 2 chip GPU is false, then its probably a ~500M transistor single chip.

the 1 billion+ transistor ATI GPU won't hit until 2008 or 2009

 

[Junkstyle]

I have to agree with you about that 700million. If that is 1 chip I'm really going to be surprised.

 

[Geo]

Orton said they already have a 1B design in-house, last week.

So is that 2008? Could be, I suppose. Even if NV's rumored 700M is in the ballpark, that was probably forward-looking to having 65nm in their pocket, for instance. 1B is probably a 45nm target rather than a 65nm target.

Edit: Why do you say that geo? Well, in part because the rumor mill has also been saying that the steep power requirements for 1st gen DX10 are a one time deal, and that second gen will be much more reasonable. I interpret this to mean that the IHVs won't push 65nm to the ragged edge the way they might be with 90/80nm.

 

[psurge]

They are going to be stuck with 65nm for a while, so maybe the first gen on 65nm won't be pushing things too hard, but I fully expect the second gen to be as completely ridiculous as 500+million transistors on 90/80nm... If they were really planning on coming back to their senses, why are they revising the PCI-E spec to include a new power connector and a section on 200W+ cards?