Tsavo
Platinum Member
- Sep 29, 2009
- 2,645
- 37
- 91
I want a cpu made by GM, that way I can have transistors falling off as I drive down the information super highway.
I want a cpu made by GM, that way I can have transistors falling off as I drive down the information super highway.
for the majority of computer users with the typical everyday usage (ms office, web surfing, multi-media, lite gaming) - a G1610 ($35 from MC) (equivalent to an e8400 from Q1 2008 $185) is more than sufficient in handling these tasks.
Last edited:
Moore's law will slow down even this decade. lithography will be a limiting factor. EUV looks like it will take 2020 to be available for high volume manufacturing. Immersion litho with double patterning can handle 20 nm / 14nm. For 10nm quadruple patterning is going to be needed. 7nm is going to be very difficult without EUV. Maybe the semiconductor companies can eke out that node with immersion litho, maybe not. This decade we will not see a new node every 2 years even at Intel, the king of the semiconductor industry.
Yes, Moore's Law can continue below 5nm. A multi-state transistor (e.g. memristor) would allow exponential computing growth from a fixed area/power budget.
How many transistors does it take to write your name? How about letters of the Latin alphabet?
How many transistors does it take to write your name? How about letters of the Latin alphabet?
Last edited:
They will hit a wall just the same if we can't continue to double density at the same mm2.
For the majority of daily commuters, a Ford Focus or Dodge Neon is more than sufficient in handling the task...and yet plenty of commuters do expect more and are willing to pay to get more.
How many bathrooms does your house have? If it has more than one then you likely have more bathrooms than are necessary.
How big is your TV? Probably too big, a smaller screen would surely be more than sufficient in handling the task of viewing television.
sm625
Diamond Member
- May 6, 2011
- 8,172
- 137
- 106
Picture a 5nm space as a football field. The types of traces that can be put on that field are the size of the individual yard lines. It just cant be done with current technology. Hell the flagellar motor isnt much bigger, and we're talking about a mechanical device with 20+ unique moving parts. 20 years ago we were using a 600nm process. 20 years from now it is not unreasonable to assume we'll be at 1.0-2.0 nm.
Phynaz
Lifer
- Mar 13, 2006
- 10,140
- 819
- 126
What 'I' think he's saying is that there will be a significant flattening of that curve for the next ~30 years in terms of xtors/sq. mm. At some point one of the two or three promising technologies that are now being worked on in University and government labs is going to gain traction an significantly improve that curve (economically) for the production of high density ICs. Whatever new process wins, it probably won't experience the same success as CMOS xtors did; but it will likely yield gains much better than the extension of CMOS like structures would. Maybe we could even see a return to doubling ever two years for a while.
The next revolution will not come from materials engineering like the scaling of the xtor has delivered over the past 60yrs.
The next revolution MUST come from the field of computer science itself. How the actual computing (logic) is done.
Computer science must transcend the era of binary logic if we are to once again experience anything like what we have experienced since 1945.
Materials science engineering can continue to carry the bulk of the legwork for the near future (next 20-30 yrs) but only with a complicated and arduous transition from traditional semiconductor materials (the stuff we've been scaling, and stuff you hear people talk of like CNT, graphene, III-V, MoS2, etc) to scaling the electronic properties of meta-materials (reciprocal lattice, Brillioun zones, etc).
There is a whole realm of untouched opportunity in the field of electronic property engineering with meta-materials once you start leveraging the reciprocal lattice, and the virtual scaling (in k-space) is infinite because you actually scale the materials in the other direction (things get bigger, not smaller, as the electronic properties scale smaller).
But we really need computer science to step up its game and evolve past the same binary logic constructs that were developed hundreds of years ago (if not thousands depending on where you see the abacus fall in the computing topology).
I'd agree, but I just don't know we are going to be able to pull off any of the technologies that you are mentioned above. They seem to be pretty solidly in the R&D phase like FinFet was 7 years ago. If everything goes OK and EUV can ramp up to 1 KW by the end of the decade, and we can get higher performance M-beam expitaxy to build the right structures with graphene and perhaps III-V materials; then the post 2020 future may be much better than it looks right now. The question is when will these technologies transition out of R&D into economical production - and right now, it seems to be and open question to me (although MB and EUV at least seem to be showing linear progress ATM).
I also agree on the compu-sci side. There have been some very interesting developments - like encoding data as algebraic statements (MIT, IIRC) which could open a whole new field (it's only being used in communications right now AFAIK). The reality is, so long as we keep getting relatively cheap performance increases - progress in computer science will likely continue to be slow. Oddly enough, if semiconductor companies were only able to deliver linear, instead of exponential gains; I think computer scientists would recieve more funding and progress their field much faster. But right now it's cheaper to work on things like scaling algorithms for HPC clusters than fundamental changes in programming paradigms, computation and data handling. There would need to be architectural changes to support some of the more radical changes like moving away from binary and that means breaking the golden handcuffs of x86 - so there is a very substantial heavy momentum impediment there.
That analogy is so wrong I dont even know where to start.
I think there are easier ways to make your point than making such ridiculous statements.
And in the car one: those 2 mentioned are already overkill.
ClockHound
Golden Member
- Nov 27, 2007
- 1,111
- 219
- 106
@IDC & Ajay: Well he did say "I don't think the world's going to give us a lot of extra money for 10 percent [annual] benefit increases,".
I think the answers have been going in several directions. Are you saying:
a) That we're not going to see only 10 percent annual benefit increases (i.e. that going forward, performance is going to start increasing much more than that again)? Based on the improvements the last couple of years I don't see us returning to the performance improvements we saw during 1990-2005. We would need some disruptive technology for that to become reality and I'm not sure that will happen. Or are you that certain the existing research projects you mention actually will bring such improvements?
b) That we'll actually only see 10 percent annual benefit increases, but that the semiconductor industry will thrive anyway due to improvements in other areas? I.e. similar to how the automotive industry has improved not by providing more powerful engines, but by providing infotainment systems, better safety, and lower fuel consumption? In that case what such alternative improvements do you see for the semiconductor industry?
I think the answers have been going in several directions. Are you saying:
a) That we're not going to see only 10 percent annual benefit increases (i.e. that going forward, performance is going to start increasing much more than that again)? Based on the improvements the last couple of years I don't see us returning to the performance improvements we saw during 1990-2005. We would need some disruptive technology for that to become reality and I'm not sure that will happen. Or are you that certain the existing research projects you mention actually will bring such improvements?
b) That we'll actually only see 10 percent annual benefit increases, but that the semiconductor industry will thrive anyway due to improvements in other areas? I.e. similar to how the automotive industry has improved not by providing more powerful engines, but by providing infotainment systems, better safety, and lower fuel consumption? In that case what such alternative improvements do you see for the semiconductor industry?
Last edited:
Based on research I know to be underway, well in the pipeline, shrinking definitely has another decade of "known solutions" identified...what remains, the reason it will take a decade to iterate through the known solutions and bring them to market, is to develop the cost-effective means of implementing them in mass production. (same reason finfets didn't come to market in 2001, but did come to market 10 years later)
Likewise the transition from 300mm to 450mm will bring cost reductions that are necessary for the continuation of Moore's law (reducing the minimum in the cost per xtor curve by roughtly 50% every 2 years).
IMO the next 10yrs is fine, but the work still has to be done. For the people who don't know how to get the work done it will seem like impossible magic, for the people who do it every day as a matter of life it is already spelled out and is crystal clear.
The question on those people's minds is "what is next, after the next 10yrs?"...and that question has always been there, so nothing is new really, including the fact that there is no shortage of folks whose career paths have vaulted them to the position of being an "industry veteran" but they have been out of the trenches for so long that they lack the imagination and know-how to see themselves and the industry through to the next 10 yrs.
In other words, to the people who are still in the trenches, applying physics and engineering on a daily basis to solve real-world scaling problems, these guys who grab the headlines are really just dating themselves and outing themselves as being out to pasture and having lost touch with what the industry is doing on a day-to-day basis. Like a UN official contemplating the daily challenges of a Syrian citizen, some things cannot be comprehended if you aren't walking in the same shoes.
I am saying that even if this came to pass, that the worse-case scenario was that the digital CMOS portion of the logic MPU semiconductor segment were to see annual cost reductions fall from the current 30% to a "paltry" 10%, the reality is the business segment is as critical to modern life as gasoline and automobiles and the industry itself is going to do just fine.
It is the customer who will be getting less for their money, not the industry.
OverVolt
Lifer
- Aug 31, 2002
- 14,278
- 89
- 91
With a whole new technology I think we would be back to square one with software and hardware development. That just is not going to happen there is 50+ years invested into x86. Just like we still use a qwerty keyboard because it kept typewriters from jamming I don't see x86 going anywhere no matter what physical limitations they will encounter. Like I said computers used to get smaller fast and cheaper. In the near future its going to be pick 2 of the 3. Small and fast but not cheap. Fast and cheap but not small. Cheap and small but certainly not fast, etc.
You already see this in the computer market with apple (small and fast, not cheap). Net books (small and cheap, not very fast) and desktop (fast and cheap, not small). etc.
Convergence is an idiots game. Convergence ends the same time moore's law ends IMO. Which is within less generations of fabs than you can count to on one hand.
You already see this in the computer market with apple (small and fast, not cheap). Net books (small and cheap, not very fast) and desktop (fast and cheap, not small). etc.
Convergence is an idiots game. Convergence ends the same time moore's law ends IMO. Which is within less generations of fabs than you can count to on one hand.
Last edited:
Nope, but using awful (as in: context doesnt applies) analogies arent helping you make that point.
In CPUs we have performance levels as one of the most important deciding factors that make it "good enough" or not. Bathroom and car analogies dont apply since both elements in they very most basic itinerations are already "good enough", this isnt the case with CPUs.
So in my old house I used to have 2 bathrooms, guess I was a bathroom enthusiast back then :hmm:
I may well have jumped the gun on Colwell's comments since I don't know his resume. I assumed that he was like a USAF captain, who I went to university with, who was a rocket scientist both in degree and in intelligence and who's job was very similar to Colwell's except that he worked for the USAF vs DARPA. Ed (my friend) had great visibility into the technologies he was tasked to assess by the USAF, since he actually traveled to the facilities doing the work and interacted with those on the frontlines developing these new technologies.
Colwell, may just be the top guy reading summarized research reports and, as you point out, may not be in the trenches and may not be in touch with how the engineering and science a progessing the field day to day.
I agree with you that a variety of technical advancements like 450mm wafers, EUV (in the longer term, not the next couple of years) and MBE which is being heavily pursued as well can certainly have the potential to continue Moore's law for the next decade (or there about).
You've brought up reciprocal lattices before and I spent some time doing some basic research on the theory. Unfortunately, I came away with very little info on their practical use in building future semiconductors, so I have to defer to you on this if you say this is an area of ongoing research with the potential to continue extending Moore's law at a more aggressive, and traditional rate of ~30%/yr.
I already fear that the slow down in annual performance gains is one symptom that the commoditization of CPUs has already begun as happened in DRAM markets years ago. The fact that desktop CPU prices are not escalating, even with AMD virtual exit from that market is, to me, another leading indicator that xtor/mm2 scaling alone will not provide the price support (or more pointedly, high margins) that companies like Intel depend on.
On that basis, something more radical than the standard trajectory of semiconductor R&D progress will need to happen to "possibly" change the direction of "high performance CPUs". But the fact that so many are reasonably happy with the performance available to them in ultra mobile compute platforms (tablets/smart phones) is probably the final nail in the coffin that lead me to think that our beloved "CPUs and Overclocking" forum could become a shadow of it's former self in the not so distant future.
I realize I didn't address your question directly Fjodor2001. But this is my line reasoning. IDC has much better visibility into this issue than I do based on his degree/experience and the knowledge of his friends still working in the industry.
Keep in mind that is is just a forum post and I am relying for the most part on publicly available information. It's not like I'm writing a thesis on the "Future Trajectory of High Performance Process Technologies" or anything like that - so there's allot of hand waving going on here.
Last edited:
Given that having just one bathroom, complete with functioning internal plumbing, is a first-world luxury; yes, the fact that you had two internal bathrooms in your house very much puts you into the "enthusiast" category from the perspective of more than 3B people on this planet.
Need versus want.
Do you really need two bathrooms, or was it just kinda nice to have two bathrooms...
Getting back to the analogies, the point of all of them was to highlight the fact that as consumers there is no such thing as "need" when it comes to the 21st century of a first-world consumer.
Nearly everything we purchase is based simply on want, not on need. And that includes our FX-8320's and our i7-4670k's...and yes it also includes the FX-9590 and i7-4960X.
They are ALL in the same spectrum of "want", nothing about any of those products speaks to our "need". So trying to argue that any one of them is somehow an egregious affront to the delicate sensibilities of the judiciously selective consumer is rather silly in my view.
About half of the world's population would look at that debate and collectively mutter a "meh :| yet another first-world problem
"Good opinion, the vehicles are made for more than 50 years with not much of an max speed/torque increase, yet they are still used and safety and fuel consumption and comfort are improved instead. Think that something like this will go with computers as well. There is still so much to improve in computing beside raw computation power. Reliability, compatibility, power consumption and more are still awaiting to be dealt with.Not at all. Well, yes, he is quoted as saying something that can easily be interpreted as that but I really doubt that was the context for his statement.
Look at space travel versus automobile.
Both went through their meteoric rise in popularity and CAGR. Space race climaxed and most certainly crashed after the Apollo program was cancelled.
Automobiles don't give us 10% per year improvements and yet the industry is thriving. (we just don't treat them like disposable assets
Now some people want us to view the ending of Moore's Law as being akin to the last chapter of the space race to the moon, after which everything is supposedly going to implode into a singularity and the semiconductor industry will shrivel into a shadow of its former self as the money stops flowing, consumers stop spending, and innovation stops happening.
That was, to a large part, how the space race fizzled. No moon bases, no trip to Mars or Jupiter's moons, no space hotels, etc.
But that was because people were not dependent on that aspect of space travel. We became critically dependent on satellites though, and the commercial satellite business has steadily grown year after year, albeit off the front page of any major newspaper.
And look at the auto industry, we are critically dependent on our vehicles. So despite being a mature industry, and despite the rather abysmal rate of improvements in terms of energy efficiency and safety, the industry continues to grow larger every year (recession induced dips withstanding).
So which do you really think the semiconductor industry is going to follow? The space race or the auto industry?
I think we are too dependent on our computers, for both work and play, to decide we aren't going to pay for a 10% increase every year. I think we will continue to pay for it, but we won't be replacing stuff every 12 months anymore.
Not to be confused, smaller nodes are economy issue rather than performance issue, nodes are becoming smaller because that causes more CPUs(chip dies) to fit in area of one wafer, thus lowering operating costs and increasing revenues for manufacturers. It ofc pushed some performance up but it's not the reason moore's law exist. Moore's law is formula for economists not technicians.
Agree with this too, there is probably 2 billion of people who live in areas with built infrastructure. Rest live under the trees living from nature gifts.Given that having just one bathroom, complete with functioning internal plumbing, is a first-world luxury; yes, the fact that you had two internal bathrooms in your house very much puts you into the "enthusiast" category from the perspective of more than 3B people on this planet.
Need versus want.
Do you really need two bathrooms, or was it just kinda nice to have two bathrooms...
Getting back to the analogies, the point of all of them was to highlight the fact that as consumers there is no such thing as "need" when it comes to the 21st century of a first-world consumer.
Nearly everything we purchase is based simply on want, not on need. And that includes our FX-8320's and our i7-4670k's...and yes it also includes the FX-9590 and i7-4960X.
They are ALL in the same spectrum of "want", nothing about any of those products speaks to our "need". So trying to argue that any one of them is somehow an egregious affront to the delicate sensibilities of the judiciously selective consumer is rather silly in my view.
About half of the world's population would look at that debate and collectively mutter a "meh :| yet another first-world problem
Last edited:
Gryz
Golden Member
- Aug 28, 2010
- 1,551
- 204
- 106
And then there are every year 760000 (760k) children under 5 years, who die because of Diarrhoeal disease.
Are you sure that having proper sanitation is just a matter of want ? I agree of course that having multiple toilets in your house is a matter of want.
I think the car industry could be a good example of what could happen to the computer industry. What's the main goal of a car ? To get from point A to point B. Fifty years ago, it would take 5-7 days to drive from the US East coast to the West coast. Today it basically still takes exactly as long to drive that same distance. Cars have stagnated in their main aspect for half a decade now. But yet, at the same time, the industry has continued, and cars have technically evolved.
Last edited:
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 23K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter