The CPU is dead! Whats next?

[perzy]

With the heatwall at 3-4 GHz and the power-envelope of maybe maximum 200 Watts, the bell rings for the CPU. The future is the discrete processors. AMD has seen this and bought ATI.
But what will the CPU do in the future? Which processing will be on separate (discrete) chips?

 

[sonoran]

Actually, I suspect your thinking is completely backwards. We're steadily headed towards SOC (System On a Chip). Soon CPU's will be able to do real time raytracing, eliminating the need for add-on graphics cards/chips. Beyond that, expect to see even more system functionality incorporated into the chip, reducing overall system cost and complexity.

 

[falloutman]

OP, the future is not discrete chips but chaotic chips. A chaotic chip is one chip (SOC) that can do it all.

 

[WR]

There may be a frequency wall with some architectures, but over time, there hasn't been a wall for instructions completed in one cycle, even with one core.

I think CPUs and GPUs are well alive, but so are SoCs. They serve different markets in volume. SoCs are great for cell phones, PDAs, even laptops if someone would make them. Discrete chips are great for enthusiast systems, workstations, and servers, where you need performance more than low integration cost.

 

[zenmaster]

I think you are missing the boat Perzy..............

The truth lies in writing multi-threaded software.
Also, there is not even a speed wall.

Expect to see 8-Core Intel Chips in about a year.
This will in essence double performance for multi-threaded apps.
It will also allow for more processing of more things on the CPU, not less.

Also, Intel is hitting 4.0Ghz very easily on the new 45nm architecture.
Expect to see 5.0Ghz w/o much effort as new steppings are released.
Intel will also be moving to the 32nm architecture in the not too distant future.

While things may change in the future, that future is many years away at this point.

 

[perzy]

Well isen't SOC just what both chip-gigants want us to belive?
For example, no GPU-core on the CPU can compete with the graphicscard with it's own dedicated memory.
If Intel/AMD admitted that the CPU has met its end, their budget would suffer.
Who can manufacture a 'mega-chip' with 100 cores, gpu,ppu etc with economy?
Should future computers be a mega-chip with 'all-you-can-use' on and everybody must pay for all that but only a few uses it all?
If your a non-gamer, do you want to pay for the GPU and the PPU for example?
The most common desktop pc is an office pc, who runs word, antivirus and some other slow apps. I sit a such a machine and I noticed no difference at all when it was upgraded from a P4 to a dualcore.

 

[Harrisson]

CPUs will stay with us for a long time, but indeed integration will gain speed. SoC will be more and more used IMO.

Just because high-end cpu's hit the "wall" now and then, doesnt mean there wont come new tech which will help to increase the performance and still wont go over board with power. Like AMD previous gen or current C2x (not counting high-end...).

Example of nice future tech: http://news.zdnet.com/2100-9584_22-5928541.html

Now thats something I'd like to see

 

[cnumartyr]

Well isen't SOC just what both chip-gigants want us to belive?
For example, no GPU-core on the CPU can compete with the graphicscard with it's own dedicated memory.
If Intel/AMD admitted that the CPU has met its end, their budget would suffer.
Who can manufacture a 'mega-chip' with 100 cores, gpu,ppu etc with economy?
Should future computers be a mega-chip with 'all-you-can-use' on and everybody must pay for all that but only a few uses it all?
If your a non-gamer, do you want to pay for the GPU and the PPU for example?
The most common desktop pc is an office pc, who runs word, antivirus and some other slow apps. I sit a such a machine and I noticed no difference at all when it was upgraded from a P4 to a dualcore.

Since we're seeing 4.0 GHz on air at 45nm and we have yet to reach 22nm.. I think we'll be ok.

I can't remember what it is, but the theortetical max of the X86 was something like 10 GHz. The future isn't too far off.

And yes, chaotic cores *ARE* next.

 

[Shadow703793]

^Agreed.

 

[michiganteddybear]

believe it or not, our biggest bottleneck in the current speed of systems is the I/O (and I am talking about everything except video). Thats why you can go from a crappy p4 to a c2d and not see much difference, if any in the 'everyday' applications (again, nothing that needs the video).

what we need to work on more than cpu/gpu speed is someway to remove the bottleneck in the i/o department.

Yes, chaotic cores (or SoC's) are comming, and will help some. but you still need to comunicate with the outside world at some time or another

 

[perzy]

Well I remember the 3,8 GHz P4. So hot that Intel had to ship it with real expensive coolingpaste, without it just throttled... That was ..eh 2003? 2004? Intel had to scrap everything and redesign their whole line. Maybe you havent noticed?
Bur shure, improvments will be made and maybe 5 GHz is possible, but a dead end is still a dead end even if you can extend it a bit further.

As for I/O I dont understand what you mean? You have to explain some more. Sure the HDD is slow but it's no real problem as long as there is plenty of RAM. And solid state drives is a future maybe.

 

[cnumartyr]

Well I remember the 3,8 GHz P4. So hot that Intel had to ship it with real expensive coolingpaste, without it just throttled... That was ..eh 2003? 2004? Intel had to scrap everything and redesign their whole line. Maybe you havent noticed?
Bur shure, improvments will be made and maybe 5 GHz is possible, but a dead end is still a dead end even if you can extend it a bit further.

As for I/O I dont understand what you mean? You have to explain some more. Sure the HDD is slow but it's no real problem as long as there is plenty of RAM. And solid state drives is a future maybe.

P4 was also Netburst, which was entirely different than Core2. Nehalem will be entirely different than Core 2.

Your argument there was pretty weak. You also contradicted what you initially said saying that 3-4 GHz is a heat wall. The envelope isn't 200W, it's like 110-130 normally. Heck the Phenom 2.6 is supposed to have 140W and you see how much that freaked people out.

Maybe you missed a couple of articles about replacing the electrical circuits in the CPU with fiber optic circuits or layering gas in between the 3D design.

Nothing is impossible in the world we live in.

 

[azfj60]

Well isen't SOC just what both chip-gigants want us to belive?
For example, no GPU-core on the CPU can compete with the graphicscard with it's own dedicated memory.
If Intel/AMD admitted that the CPU has met its end, their budget would suffer.
Who can manufacture a 'mega-chip' with 100 cores, gpu,ppu etc with economy?
Should future computers be a mega-chip with 'all-you-can-use' on and everybody must pay for all that but only a few uses it all?
If your a non-gamer, do you want to pay for the GPU and the PPU for example?
The most common desktop pc is an office pc, who runs word, antivirus and some other slow apps. I sit a such a machine and I noticed no difference at all when it was upgraded from a P4 to a dualcore.

A Model T will get you from point A to point B, but you don't see anyone making and selling them anymore do you? Have a little faith in R&D, too. With a little research, you might be amazed at what people can achieve. Also remember, computer companies aren't around to make your life easier. They're here to make money. They will find a way to do just that.

 

[bitrate]

Disk subsystems are still the major bottleneck for all PC systems even with RAID 0/1/5 installed. SSD technology looks promising but only if they significantly improve random-access transfers and keep costs down. This won't happen for a while yet.

 

[ZOldDude]

"A Model T will get you from point A to point B"
Still might be the fastest production car ever built in the world for the first 15 feet.

 

[apache_lives]

Since we're seeing 4.0 GHz on air at 45nm and we have yet to reach 22nm.. I think we'll be ok.

I can't remember what it is, but the theortetical max of the X86 was something like 10 GHz. The future isn't too far off.

And yes, chaotic cores *ARE* next.

Considering a Pentium D 965 is hotter then a Quad Core Q6600 at the same manufacturing process (65nm) yet offering more then 4x the performance, theres not too much to worry about for the next couple of years - multi core is the way, and the next step from Intel is an IMC and CSI (on nehalem), which will eliminate one chip on the motherboard which atm puts out ~30+ w of heat so there goes some of that to start with.

AMD is hitting the bump intel hit with its Prescott, and nvidias fx series - it was only time before they woke up to realise it.

I remember a time when AMD fan boys praised there SOI tech, now it seems to be holding them back.

 

[perzy]

Well the power envelope maybe is'nt 200 Watts, but somewhere around that. It's unnessesary to mark words like that when we obvisioly is talking about trends!
And yes multicore seems like the best thing since sliced bread if you listen to the marketing hype now. But nobody wanted it a few years ago!
Programmers hate it! It makes programs more expensive!
Everybody wanted the MHz-race to go on for ever, getting more and more for our money.
And yes there maybe some 'futuretech' that will make everything good again but Intel sure dident find it during the 5 years (!) they spent turning their architecture around!

But I give up this discussion now, I belived that some real ideas could be diskussed here but when I just meet...the burped up 'marketing' i wont bother typing any more.