Will there be faster (>10% performance) processors within 3 years?

[TidusZ]

From January 2011 to now we've had 2700k -> 4770k with 5-10% performance improvements after typical overclocks. Is this trend indicative of the next 3 years? From May 1997 to June 2000 desktop processors went from pentium II 233 mhz to athlon 1 ghz but I'd be satisfied with perhaps 2x the performance after 3 years, not 5-10% though. It seems the only components worth upgrading post 2011 are videocard and SSD?

 

[scannall]

From January 2011 to now we've had 2700k -> 4770k with 5-10% performance improvements after typical overclocks. Is this trend indicative of the next 3 years? From May 1997 to June 2000 desktop processors went from pentium II 233 mhz to athlon 1 ghz but I'd be satisfied with perhaps 2x the performance after 3 years, not 5-10% though. It seems the only components worth upgrading post 2011 are videocard and SSD?

The low hanging fruit was picked some time ago. At this point, it's software that needs to catch up and use multiple cores better. That's getting better, but it has a long ways to go.

 

[NostaSeronx]

There are architectures in the works that will come with 200% increases in performance.

 

[CHADBOGA]

There are architectures in the works that will come with 200% increases in performance.

Are you talking about the next 3 years or in 30 years?

Will this 200% increase apply to general code, or specific programs only. D:

 

[NostaSeronx]

Are you talking about the next 3 years or in 30 years?

One year to four years from now, hopefully. Delays might strike, economies might fall, who knows.

Will this 200% increase apply to general code, or specific programs only. D:

General code, specifically, existing singlethreaded code. Mostly, thought not being able to be parallelized or speed up. Solve the singlethreaded wall, thus increasing the performance of multithread.

 

[Ken g6]

There are architectures in the works that will come with 200% increases in performance.

General code, specifically, existing singlethreaded code.

[citation needed]

Seriously, if you can point to any articles suggesting such a performance increase, I'd be very interested to see them! Edit: Yes, even if they're from Fudzilla!

P.S. Did you mean 200% increase (3x) or 200% of existing performance (2x)?

 

[Soulkeeper]

I share the general frustration, but I suppose this pace is the new norm.
We were spoiled in the previous decade.

 

[shady28]

From January 2011 to now we've had 2700k -> 4770k with 5-10% performance improvements after typical overclocks. Is this trend indicative of the next 3 years? From May 1997 to June 2000 desktop processors went from pentium II 233 mhz to athlon 1 ghz but I'd be satisfied with perhaps 2x the performance after 3 years, not 5-10% though. It seems the only components worth upgrading post 2011 are videocard and SSD?

Unlikely that we will see much usable increase in the desktop space.

In the mobile space - esp laptops - there have been big increases in performance over the past few years. This is in both mobile cpu and gpu.

To illustrate, the i7-4710MQ 2.5Ghz (mobile quad core which you can get in sub-$1000 laptops if you look) gets a passmark score of 7998. The i5-2500 desktop 3.3Ghz desktop CPU from 2011 by comparison gets 6221 and the i7-3770 in your sig scores 9391.

On the GPU front, a GTX 860M scores a median of 4915 on 3dMark11. Compare to a GTX 560M from 2011 which scored 1820.

Maybe more to the point, the newer and higher end GTX 970M scores 11176 on the same benchmark vs a desktop 660Ti which scores 8415.

So you can basically have a laptop that performs as good as a fairly high end gaming rig from 1-2 years ago in both cpu and gpu.

On the desktop front though, it's pretty boring.

 

[NostaSeronx]

From Soft Machines;

From AMD's Research;
http://www.cs.cmu.edu/~seth/wild-and-crazy09/koreysewell.pdf

Additionally, XVP’s virtualization methods can be used to optimize single-threaded processors by providing the illusion of more pipeline resources than is traditionally available on a single-threaded processor

Everything points to small cores that provide the illusion of a huge core.

 

[szatkus]

From Soft Machines;

You are aware that was 400MHz CPU?

 

[Idontcare]

Will there be faster (>10% performance) processors within 3 years?

No.

There will be special cases, isolated examples that will exhibit such speedups.

And those isolated cases will be spun by respective marketing teams to ensure they give rise to the impression of representing the general case rather than the special corner case they merely represent.

But the general case, single-threaded existing apps, are going to continue to see relatively paltry performance gains year over year from here out.

 

[NostaSeronx]

You are aware that was 400MHz CPU?

The clock speed doesn't matter as we are talking about IPC and EPI.

There will be designs that can exploit 1T IPC by running it across multiple core/threads/clusters. This leads to a rather large boost in performance with minimal increase in complexity. The reason no one wanted to do this before now is egos.

Why step backwards to move forwards. When semiconductors can endlessly run with more futile complexity after each shrink.

 

[szatkus]

The clock speed doesn't matter as we are talking about IPC and EPI.

Clock matters.

1. IPC doesn't scale linear with clock (mainly because of memory bandwidth and latencies).
2. As that chip was designed with very low clock in mind, it probably can't reach 3 or even 2GHz.

Netburst was a very good example of design sacrificing IPC for higher clocks.

 

[witeken]

From January 2011 to now we've had 2700k -> 4770k with 5-10% performance improvements after typical overclocks. Is this trend indicative of the next 3 years? From May 1997 to June 2000 desktop processors went from pentium II 233 mhz to athlon 1 ghz but I'd be satisfied with perhaps 2x the performance after 3 years, not 5-10% though. It seems the only components worth upgrading post 2011 are videocard and SSD?

The improvements must come from better process technologies that enable cheap increases in clock speed.

 

[Enigmoid]

The clock speed doesn't matter as we are talking about IPC and EPI.

There will be designs that can exploit 1T IPC by running it across multiple core/threads/clusters. This leads to a rather large boost in performance with minimal increase in complexity. The reason no one wanted to do this before now is egos.

Why step backwards to move forwards. When semiconductors can endlessly run with more futile complexity after each shrink.

OP's statement was explicitly about performance, NOT IPC. Unless absolute performance compared to high end 2600k/3770k/4770k increases then this point is moot.

 

[Grooveriding]

Maybe if as a large trend software starts being written to be more multithreaded.

 

[AtenRa]

Im expecting more GPGPU gains than general perpuse CPU performance the coming years.

 

[hawtdawg]

edit: plz delete

 

[gqlhonourbv]

We were spoiled in the previous decade.

 

[Mopetar]

It seems unlikely as companies are chasing after energy efficiency more than anything else as the world moves increasing towards mobile devices with limited battery capacities.

I suspect that in another five years, things may have changed a bit and we might see more of a push towards increasing performance at the expense of battery life, but that really depends on some serious improvements in terms of battery capacity.

 

[Fjodor2001]

We were spoiled in the previous decade.

We were even more spoiled in the nineties. And we've been spoiled ever until ~2008 or so. But now the fun is over. And no, I'm not bitter.

 

[ninaholic37]

We were even more spoiled in the nineties. And we've been spoiled ever until ~2008 or so. But now the fun is over. And no, I'm not bitter.

Ah yes... having to go to the library to do school research, trying to navigate a new town with actual maps, having to actually pick up a vacuum cleaner instead of pressing "Clean" on the Roomba... those spoiled 90s brats indeed.

 

[witeken]

Ah yes... having to go to the library to do school research, trying to navigate a new town with actual maps, having to actually pick up a vacuum cleaner instead of pressing "Clean" on the Roomba... those spoiled 90s brats indeed.

That's a strawman.

 

[chrisjames61]

I share the general frustration, but I suppose this pace is the new norm.
We were spoiled in the previous decade.

Why are people frustrated? What can't an i7, i5, or even any i3 or AMD FX cpu do on their computer that today's processors don't accomplish? I can answer that- nothing.

 

[BonzaiDuck]

From January 2011 to now we've had 2700k -> 4770k with 5-10% performance improvements after typical overclocks. Is this trend indicative of the next 3 years? From May 1997 to June 2000 desktop processors went from pentium II 233 mhz to athlon 1 ghz but I'd be satisfied with perhaps 2x the performance after 3 years, not 5-10% though. It seems the only components worth upgrading post 2011 are videocard and SSD?

Just going back to the original post, having scanned through some of the responses.

In some benchmarks, I can match a 4770K @ 4.4Ghz -- with a 2600K @ 4.6 or 4.7.

I have been watching the forums on the Ivy Bridge and Haswell releases including Devils Canyon and now the "E" triad. I met an engineer who worked at Intel early in the last decade -- in processor design, I think. He had quit by the next time I saw him! His remark: "It's getting down the the MOL-E-CULAR level!!"

Another friend, whom I would describe as a disabled and retired physicist, raises the vestiges of his knowledge of quantum physics to conclude that the die-shrinks may evidence "strange behaviors" -- more specifically through the electrical leakage and heat despite the lower and lower TDP targets.

And certainly, Intel -- with Apple and AMD or others -- is trying to address mobile computing.

Meanwhile, with all the discussions about multi-threaded gaming and "future" games, I'm sitting here with my sig-rig and beginning to ask myself if I would ever really NEED more than this. Even with hesitation, I'd always inclined toward the "latest-greatest" chip and chipset releases, but I'd tried to time my upgrades in a way that made money sense. With more recent contact and mutual assistance with an electronics tech retiree back east who limits his spending to a severe issue of need versus want, I'm beginning to spend more and more time looking at someone else's last-gen cast-offs at the Bay or the Amazon as the owner/seller catapults toward Devils Canyon and dumps last year's system for -- well -- more than a song but "priced to sell."