Scaling dead at 130 nm (I'ts a Joke?)

[harl]

http://www.eetimes.com/semi/news/showArticle.jhtml;jsessionid=EHHZDWX2DW4S2QSNDBGCK HY?articleID=19502091

That's cannont be (prescott it's 90 nm? Is not?)

[Pafet]

now thats scary :eek:

There are various reports on the web about lower than 90nm being researched and tested to be available in a couple of years. 90nm is still new so no hurry.

There are also some new ideas (http://www.geek.com/news/geeknews/2004Mar/bch20040412024702.htm).
There will be some point at which the problems will be too great to deal with but untill then we'll just have to wait

[EMunEeE]

I know here at NC State in partnership with Motorola, trying to better the 65nm process.

[bloodbob]

http://www.eetimes.com/semi/news/showArticle.jhtml;jsessionid=EHHZDWX2DW4S2QSNDBGCK HY?articleID=19502091

That's cannont be (prescott it's 90 nm? Is not?)

Its not quite a brick wall but yes why do you think their is SOI and Low-K and stuff. Their will be a brick wall around 1nm when we hit quatum tunneling though.

[Geforce4ti4200]

no wonder intels prescotts are having such heat problems at 90nm, that process does not work well. could be why amd wont be going 90nm till early 2005 when we figure how to make 90nm work

[Pafet]

Is it true most of the "90nm" prescott is not 90nm but 130?

[retsam]

well that artical also says that we need to me smarter with our chip designs not jsut smaller..... i think we are going to see more and more time between process releases (like going from 130 to 90 and so forth) i think this is why amd is going dual core.... they know they cant keep up so there being smarter with there designs ......

oh well

[harl]

well that artical also says that we need to me smarter with our chip designs not jsut smaller..... i think we are going to see more and more time between process releases (like going from 130 to 90 and so forth) i think this is why amd is going dual core.... they know they cant keep up so there being smarter with there designs ......

oh well


Well I belive that smaller integration it's necesary in order to
make multicore CPU's
Or can we spect a 200+ mill transistors in 130 nm
100W+ 70 Cº+ :retard:

[a688]

Is it true most of the "90nm" prescott is not 90nm but 130?


I don't know about that but all 90nm or 130nm means is that the average component is created at that size, some smaller, some larger.

[nutball]

http://www.eetimes.com/semi/news/showArticle.jhtml;jsessionid=EHHZDWX2DW4S2QSNDBGCK HY?articleID=19502091

That's cannont be (prescott it's 90 nm? Is not?)

I don't think the point of the IBM guys statement was that it's not possible to go smaller than 130nm. I think the point was that a good fraction of the performance gains that have been achieved over the past, say, five years have been purely down to process shrinks allowing clock speeds to be wound up.

His point, if I interpret it correctly, is that this era is now over, and that future performance gains will have to come from other avenues, in particular architectual changes.

So, for example, a shrink from 130 -> 90 -> 65nm will give you more transistors, but not necessarily (much) faster transistors. So you have to think of ways in which extra transistors can give you extra performance, rather than just relying on the transistors you already have getting faster.

The recent fads seem to be dual-core processors, and stonkingly large on-die caches, both of which are means to get extra performance without necessarily increasing clock-speed.

[Galvin]

Just to add. Intel is going dual core next year. So instead of trying to make smaller and smaller dual core will be much faster. I think IBM is doing the same too.

[nutball]

The problem with dual-core is that it's not immediately obvious you will get a significant boost in a typical desktop environment. Servers and HPC, yes, dual-core is good. As things stang at the moment however most PCs spend most of their life doing only one thing at once, and in that case you end up with a dual-core processor with one core doing nothing.

If you're folding, MP3-encoding, playing UT2004, etc., all simultaneously you'll see a boost in overall system performance, but each task might not run very much faster than it would have done running stand-alone on a single core processor.

Until CPU intenstive desktop applications (ie. games) become pervasively multi-threaded then dual-core isn't going to have a massive impact. Fortunately however I think games will be reasonably easy to multi-thread-ify, and once there's millions of multi-core desktops out there I'm sure the games companies will use the extra horsepower. It just might take a while (eg. a year or two).

[Riptide]

Check this out:
http://episteme.arstechnica.com/eve/ubb.x?a=tpc&s=50009562&f=77909774&m=602009952631

"My computer systems seminar had as a speaker today Bob Colwell, who was one of the main architects for Pentium Pro through the first P4.

It was an interesting talk, and not terribly technical, so it's not too hard to follow even for those not terribly familiar with comp. architecture. And some of the anecdotes from the chip design world were good, as well.

The video is available at:
http://www.stanford.edu/class/ee380/
(Along with the video for all the other presentations this term, some, like the AMD presentation, were also CPU-related)

He seems to be of the opinion that we'll soon run into a combined power/complexity barrier, and that we've likely also almost reached the level of CPU performance that's satisfactory for most users. Therefore, he thinks that chip companies will need to look into new approaches to what to do with chips and the billion transistors that we can soon stick on them."

[nutball]

I guess a key question to ask is why do we need more performance than we already have? What applications require all the extra horsepower?

Well, there's games of course. And media encoding that people seem to be very fond of. But what else? In both of those examples there's still much more scope for other pieces of hardware within the PC becoming more powerful to off-load some work from the CPU (those being the GPU and the APU/GPU for media encoding).

So if your GPU handles all your scene-graph stuff or whatever it is that makes games slow these days, the VGA engine handles MPEG-2/4 encoding and decoding, and MP3 encoding/decoding is handled by your sound card, what do you actually need anything faster than a 3GHz Athlon 64 for anyway? (Apart from posing of course). I'm not sure I know the answer. Anyone?

There are pretty obvious reasons why Intel and AMD haven't been pushing this approach, but from the point-of-view of system performance overall maybe it's time to look beyond pure CPU horsepower and start thinking about some of the more neglected bits of silicon inside our boxes.

[Arrghman]

I don't know about that but all 90nm or 130nm means is that the average component is created at that size, some smaller, some larger.

Not quite, the size of the process specifies the grid that everything in the chip is aligned to. You can't make anything smaller than the grid on a chip... there's nowhere to put it! There's also minimum sizes of structures and minimum spacing and the whole thing becomes a rather large mess to deal with. But, in short, 90nm specifies the minimum size anything can be, and everything is aligned to that.

Also, someone said the brick wall would be at 1nm... everything I've read/been told points to somewhere around 25-35nm

[Galvin]

Well you're forgetting about servers. cpu's need to be made faster for servers. Servers are never fast enough. Look at Nvidia. They are always looking to faster servers. I'm sure research firms that do weather models need faster servers. The faster the cpu the more things that can be done in the professional market. And for home use games is the biggest reason of all for faster CPU. Newer games like half life 2 require cpu power for the physics engine. Things like making a certain sound when two different objects hit eachother. And AI all that is done by the cpu. For now i'm sticking with my 3ghz P4. I can't afford huge upgrades this year.

From what intel is doing, is that dual core will be the standard across all platforms next year. Intel knows they can't just make faster cpu's with die shrinking. things are getting too hot. And dual core will be nice for servers too. It will be interesting to see what kind of performance they give in apps designed for 2 cores.

[nutball]

Yes, but do servers need faster processors, or more processors? A great deal of server workload is embarrasingly parallel, just throw more processors at the problem. The individual CPUs don't necessarily need to get faster.

There are of course packing density and power/cooling constraints, but this is where dual- and multi-core chips are a big win -- they allow you to fit more computing power in a given space and power budget.

But what I'm getting at is that that is subtly different from same number of processors with more gigahertz.

As for AI in games, the CPU budget for AI in games is currently something like 10% I think. The same for the physics. Most of the workload is still in the graphics engine, managing the scene-graph, etc. I'm wondering how much of this will get off-loaded onto ever increasingly complex and specialised graphics sub-systems. There are already people working on using GPUs to do game physics (I'm one of them!). IMO it's only a matter of time before GPUs become multi-chip solutions, and some of these chips start to provide hardware acceleration for the higher levels of the graphics engine (scene-graph management for example), and assistance for physics and AI.