CPU Technology - Ahead? Behind? Or right on track?

[ShintaiDK]

Before, I only had Comcast in my area. It was great, I had highspeed internet. However, I had constant dips in my connection due to Comcast not having enough throughput for everyone. Then, this magical thing called competition came to my area. Verizon came with a FIOS line.

Now?

I have amazing throughput with Verizon, and Comcast has also upgraded their lines in the area MULTIPLE times with Verizon also upping their speeds to provide a "competitive" rate/speed to the "competition".

I can't believe you'd seriously advocate against competition.....

You confuse competition with normal infrastructure upgrades/rollout. Also again, competition doesnt magically solve it.

Let me give you an example. In Denmark we now had fiber rollout the last ~15 years. However only around 40% of the population got access to fiber today. And thats with HEAVY competition and bankrupt companies due to the harsh competition and power companies putting countless billions into fiber as well. Not to mention the access is very uneven distribution. Its really bad economics to put fiber in the ground as the only thing. You do it instead when other infrastructure needs update. Sewers, district heating/cooling, power etc.

Also just because you got fiber doesnt mean you get a better product. Fiber here even with the 40% is quite unused. Only around 30% of those with fiber access use it. The rest is still on pstn copper or cable. Hell, I even have fiber in my home but I use cable instead.

Let me show you the map of how it looks with the coverage degree in %.

People simply forget the economics behind. And just sit back and demand x product because they feel entitled for it. The same thing applies to CPUs in this thread. Because people look back on old days and then demand the same thing today. Because how hard can it be?

In the case of economics. Your ISP also need to upgrade the rest of the backbone and equipment. And while you may demand x faster product. If the rest doesnt do it in your area to make the business case, then you have to wait for more natural replacement cycles. In the US its also different since you have no access to other networks. So the infrastructure cost is much higher than other countries since each company needs its own network. And the payback on these investments are usually in the 20-30 years range. So if you want fiberrollout to everyone, then competition with multiple networks is a really bad thing.

 

[NTMBK]

The P4 was a speedracer design that didnt work. AMD repeated the exact same mistake with Bulldozer.

Netburst had its ALUs cranking away at >7GHz, when AMD were on <3GHz. Bulldozer is nowhere near the "exact same".

 

[Nothingness]

As a hardware design guy, I definitely share that sentiment. You have no idea how many performance problems starts with the sentence "Well... so there's this old FORTRAN library..."

As a software/hardware guy I can't tell how I feel when I hear pure HW people say "they'll recompile their code and get huge speedup". Legacy is here to stay, and your job as a HW guy is to make sure you don't break performance of legacy code. That's frustrating, but that's reality

OTOH the inability of software guys to properly use multiple cores is definitely a SW issue (unless you have a buggy HW implementation) and a very frustrating one too...

That being said, I'm very impressed by how processors have evolved in the last 30 years, both from a process and a micro-architecture point of view. That was predicted by Moores's Law, so one shouldn't be surprised, but that doesn't mean it isn't impressive

 

[Omar F1]

Sure they'd be stunned at how much transistors, cores, advancement of IPC has gotten into the same die area.

As long as they didnt see what these chips were running, they'd be impressed. But as soon as they saw a version of windows that took even longer to boot than win 95, and then web browsers full of terabytes worth of fail videos, they'd want to shoot themselves and throw their corpses into the machines.

Made me laugh
I truly don't know how MS has got away through all the years with their quality control.

 

[SAAA]

I still need a citation that is in actuality still possible to extract a meaningful amount of IPC out of a CPU (without a hugely disproportional increase in power). If it was so easy to still get big gains, why hasn't AMD released such a chip, or why hasn't Apple doubled CPU performance again with the A8(X)?

I'm 100% sure it's possible in many scenarios, not all definitely, especially if it is branch heavy code or with too unpredicatble behaviour.

Why so? Look at FPGAs and how much they can increase performance once tuned for a particular application, also while running far slower than today processors...
Now imagine the same thing but done in hardware by the CPU itself: not anymore a fixed architecture with the phisical limits of how it was designed (it's just a piece of silicon in the end) but a processor that changes depending on the code it's executing, within a limit of course.

This would be a fantastic step for computation and informatics in general: the problem now is how hardware is just that, parts that don't change, while software and code in memory do.
Just think of your brain: not only you acquire new memories but also adapt and evolve, you can get smarter!

 

[ShintaiDK]

The FPGA approach is cute. But I really doubt it would work in a generalized workload. So you basicly have to reconfigure the FPGA all the time and after what app you have running. Assuming the app itself runs a very limited type of code.

Its the exact same issue with GPU vs CPU. Whenever you run the 99.x% code, the GPU just falls flat on its face. But in the rest its very fast.

 

[Fjodor2001]

Coming from 1994 to 2014 I think they would be amazed.

But if they took a second closer look and saw the progress made only during the last 5 years I think they would be very disappointed and worried about what the future would hold going forward. Because most of the progress made in the 20 years from 1994-2014 was done during the first 15 years. At least if we're looking at actual CPU performance increases.

 

[sm625]

I don't really understand this comment. I can barely walk over to my chair after pushing the power button before my system is fully ready which is hugely different from even 10 years ago.

It was a sort of proxy remark or euphamism intended to encapsulate the total absurdity of microsoft software in one quick statement. Windows is just plain bad. And I mean bad. I captured this screenshot today:

Half the time the Store doesnt work. It will say the same thing "try again later". REALLY? TRY AGAIN LATER? Or sometimes it will suggest a reinstall? REALLY? Reinstall windows because a stupid store doesnt work? lol. Why is there even a "store"? Whose dumb retarded idea was that? And why cant they make such a simple thing work? That is why I say anyone from a 1994 fab looking 20 years into the future would totally freak out to know the kind of garbage that is running on such high tech chips. It truly is garbage.

 

[witeken]

Just think of your brain: not only you acquire new memories but also adapt and evolve, you can get smarter!

But your brain isn't a Von Neumann chip.

 

[VirtualLarry]

Coming from 1994 to 2014 I think they would be amazed.

But if they took a second closer look and saw the progress made only during the last 5 years I think they would be very disappointed and worried about what the future would hold going forward. Because most of the progress made in the 20 years from 1994-2014 was done during the first 15 years. At least if we're looking at actual CPU performance increases.

What do you think they would think about about 2014 Kabini E1-2500 CPU? Considering how slow it is.

 

[witeken]

Coming from 1994 to 2014 I think they would be amazed.

But if they took a second closer look and saw the progress made only during the last 5 years I think they would be very disappointed and worried about what the future would hold going forward. Because most of the progress made in the 20 years from 1994-2014 was done during the first 15 years. At least if we're looking at actual CPU performance increases.

I guess Intel would disagree. The point of an exponential is that most of what's changed always happens in the last few years (if power dropped 20X, then 10X of that would be because of 14nm, e.g.).

 

[kimmel]

lol. Why is there even a "store"? Whose dumb retarded idea was that?

Because millions of customers and developers love the Apple and Google stores. Not having one now that everyone expects an official store, was hurting Microsoft. It certainly wasn't Microsoft's idea.

 

[Fjodor2001]

I guess Intel would disagree. The point of an exponential is that most of what's changed always happens in the last few years (if power dropped 20X, then 10X of that would be because of 14nm, e.g.).

I was talking about desktop CPU performance increases.

 

[kimmel]

I was talking about desktop CPU performance increases.

Ignoring all of the progress over the last 5 years, we've had no real progress over the last 5 years. You know how silly and myopic that sounds?

 

[Fjodor2001]

Ignoring all of the progress over the last 5 years, we've had no real progress over the last 5 years. You know how silly and myopic that sounds?

How much progress has been made in max car speed over the last 10 years?

Plenty!

Really?

Yes, the fuel consumption per mile has gone down.

Huh...?

 

[hawtdawg]

But as soon as they saw a version of windows that took even longer to boot than win 95

What? computer boot times have gotten so fast that its almost not even an inconvenience anymore. I had time to take a piss while windows 95 booted. Now I barely have time to walk to the door.

 

[III-V]

I was talking about desktop CPU performance increases.

The speedups you're looking for are there... They just require a recompile. So new software runs very much in line with your expectations. AMD also fares much better in this regard.

TuxDave pointed this out nicely.

 

[Fjodor2001]

The speedups you're looking for are there... They just require a recompile. So new software runs very much in line with your expectations. AMD also fares much better in this regard.

TuxDave pointed this out nicely.

Hmmm.... will that really apply to all general SW, or only certain specific tasks like encryption or multimedia encoding/decoding?

 

[NTMBK]

The speedups you're looking for are there... They just require a recompile. So new software runs very much in line with your expectations. AMD also fares much better in this regard.

TuxDave pointed this out nicely.

Good luck with that. Non trivial code is a nightmare for vectorisation, and massive chunks just don't map to vector instructions at all.

 

[III-V]

Good luck with that. Non trivial code is a nightmare for vectorisation, and massive chunks just don't map to vector instructions at all.

Getting software developers to optimize is a nightmare in and of itself... but what you're saying has no relevance to the point I was making. There is only so much you can do for legacy apps aside from ramping up clock speed.

Hmmm.... will that really apply to all general SW, or only certain specific tasks like encryption or multimedia encoding/decoding?

There are improvements made virtually every generation that can benefit a wide range of code, and need a recompile to take advantage of.

 

[Fjodor2001]

There are improvements made virtually every generation that can benefit a wide range of code, and need a recompile to take advantage of.

Sure, but I just don't see it compensating enough for the slow performance increase we're seeing in desktop CPUs nowadays. Certainly for certain specific tasks like media encoding/decoding, but not for general code. If anything code should be made more multi threaded though.

Also, this "recompile trick" applies to CPUs from 1995-2006 too since we got new instructions during that period as well, so it's not unique for the last 5 years.

 

[Nothingness]

Getting software developers to optimize is a nightmare in and of itself... but what you're saying has no relevance to the point I was making.

You wrote a recompilation would speed things up. How much speedup will you get if no vectorization happens? Do you know OoOE has basically made code scheduling specific optimizations almost useless? So how do you think recompilation will speed up things?

There are improvements made virtually every generation that can benefit a wide range of code, and need a recompile to take advantage of.

Do you have examples apart from new instructions?

That being said I certainly agree software developers are lazy and most of them don't even know what a cache is and even less how to adapt data structures to use it efficiently. But hardware developers also are lazy when they say recompilation should be done to benefit from a new micro-arch.

 

[OVerLoRDI]

I think what we have seen in the last 20 years is amazing. However, we are starting to slow down quite as far as making things smaller. Technology always advances, but at this current stage the advancement is getting incredible complex and expensive. Maybe things will pick up in a couple years when people find a creative solution, but for now progress is very incremental.

Also raw performance is mattering less and less now. As is evident by the people still rocking C2Q/C2Ds happily.

 

[ehume]

And in general, with every jump in performance of the CPU, our OS becomes more complex. My 4.3 GHz 4770k with Win7-64 on an SSD is not much faster than my old 4.0 GHz 875k with Win 7-64 on an SSD. My 486 with Win95 on a spinning disk seemed as fast. It ran like a striped ape -- 15 years ago.

OTOH, in my basement I have a 4.5 GHz 4790k with Win 8.1-64 on an SSD. That flies. So I guess software may be advancing. About time.

 

[Carfax83]

OTOH, in my basement I have a 4.5 GHz 4790k with Win 8.1-64 on an SSD. That flies. So I guess software may be advancing. About time.

Yeah I agree with this. Windows 8.1 is a very good operating system, which handles and uses modern hardware very well.. Windows 7 on the other hand, is a bloated pig in comparison.

And Microsoft is going in the right direction, because Windows 10 or whatever it will be called, will be more optimized for hardware at a lower level courtesy of DirectX 12.