For those interested.

I’d like to propose an open think-tank on future computer architectures. I suppose you could call it a wish list. You know, if the industry was more willing to adapt and change. Now I’m thinking more along the lines of simple specs that seem possible given current technologies in development, preferably being producible in the next 10 years (hypothetically), mainly for those (like me) who don't know enough about the nitty-gritty of architecture... but if you feel like getting more in-depth go right ahead. I personally would be fascinated by any contribution by someone who could really go in-depth into a design.

Here is an example;


Cheap:

Two 3 THz Cores with 64 Mbs of dedicated L2 cache and 12 Mbs of shared L3 cache. (BDT processors, ZRAM cache)

16 GB XDR2 @ 10 GHz

128 GB PRAM Flash Drive


Ideal:

Four 3 THz Cores with 12 Mbs of dedicated L2 cache and 6 Mbs of shared L3 cache. (BDT processors, BDT type memory cache)

BDT RAM @ 800 GHz?

MRAM 64 GB drive (this could be more far off than 10 years )


I'm sure there are some silly items in those hypothetical specs, so please feel free to point them out as I consider this a learning experience. I tried my best to be true to the potential of said technologies. Also I figured sound and video processing would take place without separate cards, maybe cores dedicated to different task sets. Though in that case you might need to separate the cores into individual cpus as I’m not sure if you’d need distinct designs to efficiently process those different types of data, unless the arch lends itself well to all sorts of tasks. Then it would be entirely up to the programmers.

oh... you were serious.. haha

Well I suppose not... just for some reason reevaluating an existing tech doesn't seem as enticing as pushing on to something new, but the best situation will always be a combination. Lets say we do find a way to make the performance ramp up in perfect multiples, add BDT tech and that ghz becomes a thz, replace silicone with diamond and it becomes 80+ thz per core? But with stacked performance from multiple cores that becomes well over 100 thz worth of performance.... but I guess we're just talking wet dream at this point.

Here's what I predict for there to be in ten years:

1. 16 nm transistors. With the current 2-3 year gap between shrinks, 2008 will see 45 nm, then 2010-2011 will be 32 nm (and immersion EUV, which might delay it) , 2013-2014 will be 22 nm, then 2016-2017 should be 16 nm.

2. 16 nm transistors will result in an average 150 mm^2 die yielding roughly 2.4 billion transistors, as opposed to the ~300M on a 150 nm 65 nm die. This is three doublings. I do not see die sizes going up much as die size is tied to yields and die size hasn't gone up *that* much over the years.

3. CPUs with 32 to 64 cores. 4 cores has happened on 65 nm, 8 will on 45 nm, 16 on 32, 32 on 22, and 16 nm will see 64 cores fit in a ~150 mm^2 die. My guess is that the chips with 8 or more cores will start to be heterogeneous like the Cell is, with some specialized cores in place of general-purpose ones. There will be stream processors, GPU cores and such on there as well as a fair number of general-purpose execution cores.

4. Clock speeds will likely not be nearly as high as you predict. Assuming that we don't see a return of a low-IPC, high-clock arch like NetBurst or a radical change to a high-IPC, low-speed one like the UltraSPARC T1/T2 or Itanium, clock speeds will slowly raise with each generation. I predict something in the range of 5 GHz range. It could be much, much higher if number of cores does not double with each generation, but I think that the cores will keep multiplying.

5. TDPs will remain largely unchanged, except for perhaps a line of ultra-low-voltage desktop CPUs like AMD's hard-to-find 35 W X2 3800+ EE SFF. The manufacturers will simply modify clock speeds to remain in the current thermal envelopes as 40-80 W seems reasonable for a desktop and has been the average range since the late 1990s.

6. We will likely be running 128-bit RAM offering quad-pumped data accesses per clock. Comparing 2007 DDR2-1066 in dual channel to 1997 PC66, we saw a 32-fold increase in bandwidth. I predict a similar increase in ten years and as before, it will be mostly due to many accesses per clock cycle rather than running memory modules at extremely high speeds. I doubt they'll touch 1 GHz actual module speed. I predict that we'll see a similar increase in RAM capacity as 32 MB or so in 1997 versus 2048 MB today will suggest that we'll run 128 GB RAM in our machines in ten years. Maybe more if Windows is widely used

7. Hard drives will get larger but not all that much faster. If they'll pull more than 200-300 MB/sec, I'll be floored. A 1997-era drive pulled about 10 MB/sec on about 4-6 GBs while a modern HDD will pull 70 MB/sec on 500 GBs. Speed is slacking off relative to capacity, so we'll have 50 TB HDDs that aren't all that fast. I do see 10,000 rpm drives becoming more popular in that time as it will be possible to make 500 GB, 1 TB drives out of the little platters.

8. Flash memory WILL be popular for storage, particularly in notebooks. This will take nowhere near 10 years to see, my guess is about three years before a fair-sized (64-128 GB) Flash drive has enough r/w cycle ability to withstand being a HDD and cost a hundred bucks. I doubt that Flash will be as popular in desktops as quickly as it will take Flash drives or Flas drive arrays of hundreds of GBs to a few TBs to be seen as a replacement for mechanical drives. In ten years, they might start to be common, especially in SFF machines, but likely not in desktops.

9. Desktops will become less and less popular as people will buy more and more laptops. Most people will have a large desktop-replacement laptop that sits on the desk and never moves instead of a desktop. We're starting to see that already. Screen sizes of 17", 19", and 20-21" will be good sellers, as will usable ones with 14.1", 12.1", and 10" screens. The remaining desktops will either be really cheap low-end units, gamer boxes, or workstations.

10. Internet will get faster, but not terribly so, especially in areas with one ISP. Cable has a limit of about 56 Mbps IIRC, so this will limit cable users to roughly 25-30 Mbps. DSL will hit maybe 10 Mbps, so expect limited fiber rollouts a la Verizon FiOS with speeds probably getting up to 100 Mbps. (They don't have to go much higher since nothing else can.) Long-range wireless will allow some competition in suburban and rural areas as urban areas will exhibit too much interference from both buildings and the embedded telcos/cablecos. Prices likely will not get really cheap, but the wireless will will allow for some competition.

11. DRM on files will not die- mark my words. It's too lucrative not to. It will peak when some company thinks that they have a good implementation and it will ebb when zero-day cracks shred it and people find the wares a PITA to use legitimately so they bitch about it and don't buy or download warez. The companies reduce the amount of DRM for a while to let it blow over, then repeat. The "repeat" cycle will bring even more draconian schemes, which will be facilitated by computers incorporating more closed-loop DRM systems like TCPA.

12. The patent office will remain stupid and there will always be a ton of patent-as-a-weapon suits filed. We'll never see the amount of stupid patents and stupid lawsuits stop because patent attorneys are attorneys and so are legislators. They profit from the mess, so they won't stop it.

13. Laws regarding copyright and such will probably get modified to allow for backups (the DMCA) but the RIAA, MPAA, BSA will just get more and more powers due to effective and sleazy lobbying. They will see their *effective* impact be reduced as the Internet (out of the RIAA, MPAA, and BSA distribution channels) becomes a bigger medium and physical disks become a smaller one. Illegal copying will perpetually remain as strong as ever.

14. Microsoft will diminish significantly as a company. They will have the most market share in OS and office suite programs, but it will be less than today. The current trend is towards open file formats, which will gradually reduce lock-in to Office. More things will be Web-based and OS-agnostic, so it matters less what's on your computer. There is and will continue to be a marked trend of people not upgrading to the latest versions of MS Office and Windows since new versions offer fewer reasons to upgrade than in the past. As a result, MS will see ever-decreasing revenues as they sell less and less new stuff. MS will largely rewrite Windows to make it modern and competitive, but this will largely break compatibility with old legacy apps. This will provide yet another break down of the lock-in. I don't predict that MacOS will become stupendously more popular, nor do I predict that Linux will be terribly popular at home. They will both see modest gains, but most people will still use MS OSes as that's what's being sold. But I do see Linux making a big inroad in the corporate sphere after the MS lock-ins start to go away. That will not happen until the end of the 10-year period, though, maybe not for a little bit later yet.

15. Most computers will become more and more closed, appliance-like devices with less user configurability. This will be done in the name of increasing the ability of computers to become "multimedia devices" and the anal-retentive demands of "content providers." Also, this will sit very well with people wanting to not have to do any maintenance or configuration a computer- just turn it on and go. Enthusiasts will be still able to make their machines and tweak to their heart's content, but like today, DRM will be a big PITA. Probably a bigger PITA. But cracks will be alive as ever and widely used by enthusiasts.

I think the accuracy of your predictions depend entirely on whether or not companies like intel will move beyond just die shrinks and multiple cores. If they did move to a transistor type like BDTs it would decrease power requirements by a factor of 10 and increase performance by a factor of about 100, but it would also require being more flexible than they're use to. I would hope that hard drives are fazed out for the new types of flash with supposedly infinite write cycles. Sadly though there is a very good chance the hard drive companies will try everything they can to discourage that.

I do agree with verndewd that the need for modifications and number of components will drop off... i'm just afraid the industry won't allow that to happen for quite some time. It would eliminate a lot of companies.