Why don't AMD and Intel seem to get that the process is dead?

Like the topic says, both companies are apparently having problems with the 90nm deisgn process. This is apparent with the problems Intel is having with Prescott, and the non-linear way it reacts to votlage. Obviously, making CPU's at 90nm is hard. From what I've read this is due most likely to current leakage, and the loss of signal integrity.

Now, the other day we get an article on the front page that says AMD is expecting to have to same problems as Intel when they move to 90nm processors. This really isn't all that big of a surprise. I mean, Intel went with Strained Silicon, and AMD went with SOI. Besides these, the processors are basically made the same way, with the same transistors, and the same size wires for current and what not. Apparenly SOI worked a bit better than Strained Silicon did, but it's no miracle solution.

Anyways, let me get to my point... If the 90nm proccess is as bad as it is, why are both AMD and Intel sinking tons of cash into building 65nm fabs? I mean, do they think that 90 is just a bad number, and that 65 will be that much better? I would think with all of the apparent problems with further shrinking the design proccess they might be sinking all of that money into R&D instead of fabs to build chips on an even smaller process.


Anyone got any insights they want to share on this?

It's all about progress. They have to move on to smaller and smaller processes, or they will be left behind. I am sure they are spending more than a few dollars to perfect the 90nm process, but they also have to think about the future. R&D for the 65nm process will also be helpfull with the 90nm process. They are building the Fabs now, so that when they actually move to a 65nm process, they already have the facilitys to make it, and won't have to wait while they are built.

if they're having trouble with 90nm, theyll have even more trouble at 65

I took an introductory course to Nanotechnology, and man when you get below around 100nm or so things do get dicy. The transistor I/V curves change significantly and so the basic makeup of the switches must be carefully altered. It may take some delays, but don't think its impossible. 65nm though, man that may be very hard to do.

As far as I think, this eventual inability to keep shirinking (65nm is probably is pretty good wall) will hurt intel more though since they seem to need the higher clocks and lower power more than AMD.

wall? what wall? it will trip them up and slow them down, but the wall is gona be when the transistors are made of only a couple atoms. Technology is going to progress, no matter how difficult it is.

First off, the title of this thread is "Why don't AMD and Intel seem to get that the process is dead?" This is utter nonsense! Nothing could be farther from the truth from the Intel point of view. Let's look at facts: Intel has already released and is marketing CPUs that are 90nm. These Prescotts are working fine for the typical consumer at their rated speeds.

Yes, there are a few bugs and heat issues that are being worked out. Sure, us overclockers are disappointed that we can't overclock the snot out of them and hit 4 gig on a regular basis without suffering a meltdown, but us overclockers are also a small minority in the great scheme of things. Intel has a long history of improving on a design and a new Prescott stepping will be released in a few weeks that will address some of the problems.

The fact is Intel is way far ahead of AMD with the 90nm process. It was a smart move on Intel's part to be the first to 90nm. Better cooling will be required, but us overclockers know that can be accomplished. By the end of the year, Intel will have it all figured out while AMD flounders around and will get left behind in the dust.

Batboy typicle response.....

From what i heard AMD is having way less problems then Intel did with the 90nm process which to me shows that AMD wont need much to be catching up. Amd at the time didnt need to go to the 90nm process. Intel cpu's were the ones that couldnt be pushed anymore thats why they went a diff route. Just because one company has problems @ doing something doesnt mean another will.

Intel is the one thats is left in the dust, thier Prescott cpu's are junk and everything ahead of it was Canned that shows u alot.

Hmm... Its a moot point anyways... Both companies are moving to dual cores and Intel is going to use its notebook processor ( Dothan?) for future desktops...

Batboy and CandymanCan... You guys need to chill... The prescott IS pretty ****ty (compared to what intel had been making) and nothing AMD is doing is impressive either... I think both companies are running into the same problems and are being subsequently slowed by them...

The immediate future is pretty bleak... but who cares? The current generation can run anything smoothly and can overclock like mad. Enjoy it while it lasts...

CandymanCan said:

Prescott cpu's are junk

Click to expand...

Hahaha, now that's funny. I recently sold my 2.8E Prescott and currently am waiting on the new Precott stepping so I can try a 3.0E. I was able to run a stable 3.8 gig with my old Prescott CPU. I have lots of benchmarks that I can post if you want to post your AMD system's benchmarks to compare to see how badly it gets whipped. Look at this screenshot... does that look like junk?

Yes, because you must take more extreme measures to get it and it gives u less increase power per increased clock speed than a northwood...

Compared to the Northwood, Prescotts are not as nice... But thats without SSE3, I dunno when software is gonna utilize that and how it will effect performance... It may be really really fast and i will eat my words LOL ...

Yes, you need to do much more cooling on a Prescott compared to the Northwood. Just like you need to do much more cooling with an AMD system compared to a Northwood.

The early Prescott is immature and needs improvement, there's no doubt about that. I posted comparisons a while back with my 2.8C Northwood vs. my old Prescott 2.8E. At default speed, the Northwood outperformed the Prescott. But, at 3.4 gig, the Prescott had caught up in many benchmarks. The Northwood beat in some and the Prescott beat in some, then in others it was dead even.

The higher the Prescott is overclocked, the more it shines, assuming you can control the extra heat. As the Prescott matures and new steppings arrive, they will keep getting better. By later this year, 4 gig O/C's will be common place. The point is 90nm is here to stay and it ain't dead.

The prescott is very immature at this point. The prescott is made to shine at much higher clockspeeds. When we get into the higher steppings and higher clockspeeds of the prescott we will see a definate improvement in performance.

Nice OC on that prescott BatBoy!

I've hit a wall on my Prescott. Cant seem to get past 3.51ghz on air. Too Hot

Hm... I really could care less about chips hitting 4 gigs... I would much rather have one of those nice notebook cores... Thats the wave of the future, not prescott... IMO, Prescott is the dying gasp of the clock speed race...

The "process" is far from dead. It's just that the rules are changing. Historically, a process shrink meant that you could run the same chip design faster. But due to increased leakage, running chips faster on smaller processes is starting to take more and more power. SOI and strained silicon only help -- without those both Intel and AMD (read IBM) would have significantly more problems with their 90nm process.

While the problems with Prescott are obvious to everyone, few people ever mention Dothan. Intel is also making this chip on its 90nm process. It has nearly twice as many transistors as its predecessor (in about the same die space) and runs ~15% faster. But it does not require more power. While Dothan was delayed due to a bug in the design, I think it's hard to look at that chip and say "90nm is broken and things will only get worse with 65nm."

Certainly there will be challenges moving from 90nm to 65nm to 45nm. That's what engineering is all about.

Batboy,

You cant compare my Barton to a New gen Prescott. I want you to compare youre Prescott to an A64 1mb L2 @ 2.7ghz ok .

The Athlon 64 Crushes the Prescott

You must be blinded by MHZ arent u. My Amd compard to a Northwood which is in its class will be equal to youre P4c @ 3.5ghz when mine is @ 2.5ghz. Equal or better in some areas, and worse in others. Both cpu have thier ups and down's in certain spots. From the way youre typing it seems like youre turning very agressive. You want yo compare my cpu to youre P4 Prescot whe nyou really should be coparing to the A64, but i relize why u need to use my Cpu because youres doesnt compare to the A64.

batboy said:

Hahaha, now that's funny. I recently sold my 2.8E Prescott and currently am waiting on the new Precott stepping so I can try a 3.0E. I was able to run a stable 3.8 gig with my old Prescott CPU. I have lots of benchmarks that I can post if you want to post your AMD system's benchmarks to compare to see how badly it gets whipped. Look at this screenshot... does that look like junk?

Click to expand...

Very poor overclock. Prescott's datasheet says voltage limit for 90nm process is 1.55V. When you put more, transistors will be damaged... and you have much more. Voltage limit for 130nm was 1.75V and Northwoods were dying when using as low as 1.65V for long time. So this is nothing you should be proud of as it can't be used that way for more than few weeks / months.

BTW, Prescott leakes 50Amps ! That is almost 70W of power wasted due to leakage. 90nm process really has big problems.

Hmm i didnt know this, but Batboy seems like a Intel big shot he shoulda known that.


Before you say 2v is to much for me Batboy AMD sheets stat that Certain barton stepping's max safe voltage is between 2 and 2.1v

yes, please boys, keep this thread about the process and not who is better Intel or AMD....

I disagree with the fact that you can just keep getting smaller and smaller indefinitely. As you go much below 50nm, the rules of the game change significantly. The quantum factors that were not very important at all at let's say 130nm, now become very significant if not the driving forces and properties of materials change. (i.e. metals that conducted fine now become hightly resistive, leakage becomes significant due to quantum tunneling etc etc) so once you get down to probably 65nm, you are going to have to have fundamental changes in the designs of transistors, and then are going to have to wait for the automatic design tools and process engineering to catch up to this new fundamental change so it will take longer. Who knows, we may end up using completely new type of computing systems like ones based on light switches or even superconductors. Whatever the future holds, I think that it is clear that we are hitting the limits of process technology and may end up having a revolution instead of evolution in computing.

what we need is Fiber Optics, that would would mean an Infinite Speed in MHZ.

CandymanCan said:

what we need is Fiber Optics, that would would mean an Infinite Speed in MHZ.

Click to expand...

Why? Light has a finite speed. And moves slower through fiber than vacuum. Plus at some point you need to convert the light waves to electrical signals, which has a delay. While light-based processing could be significantly faster than semiconducors, it would not run at infinite speed.