Building The Linux Kernel In 60 Seconds | Hacker News Reader

Building The Linux Kernel In 60 Seconds

phoronix.com

64 pointsllambda21 comments

tl;dr version - Intel's latest chip is pretty fast.

Having 'grown up' around computers I have found that this sort of 'damn!' moment is not uncommon. At some point in a person's experience there is something they do with computers a lot and its 'hard' for the computer to do (aka slow) and then suddenly one day its 'easy.' It calls attention to the growth that is the industry. My particular moment was a 1 GB disk drive, at USC that was how much space was hooked up to all three DEC-10 machines (330MB per) and it took up a room.

That being said, the new Sandy bridge is a beast of a machine. One can wonder if it is near the peak of what you might use as a desktop in the post-PC era. Sort of like the muscle cars in the 70's which were replaced in the 80's and 90's with cars that went for different targets (mileage, handling, etc).

"use as a desktop in the post-PC era". Isn't a desktop computer pretty much a definition of a PC?

I doubt that this is the peak. With muscle cars, there was no need for more muscle. With desktop computers you can easily see use cases for more "muscle" and that need will keep the growth going on.

It would be interesting for you to say more about 'easily see use cases for more "muscle"'. I was under the impression that one of the challenges that PC vendors were facing is that laptops and PCs are 'good enough' for more and more of the computer marketplace and so they were unable to depend on 'upgrade' revenue as the components got more capable.

I don't doubt that there are folks that can use infinite amounts of computer power, but if they become so specialized perhaps it will shift the market around again.

One common scenario are the browsers. Imagine websites with real 3D model e.g. nike.com with a modeled shoe. Browsers are becoming an application platform and to deliver rich client style UI, it takes juice.

Video editing is becoming more and more common place with cheap digital video cameras and the ones on camera phones.

I agree that the upgrade cycle may become longer, but to me it seems like there is a market for more CPU power on the desktops. I think there is some normal fluctuation where you improve one or more corners of the industry at a time.

I also don't think it's the peak. Perhaps Ivy Bridge (the next iteration due in 1H2012) since it's rumored to have more or less the same TDPs but with 20-50% more performance. Although the TDP is going to be programmable so OEMs could decide they "underclock" the chip.

I know this is not what the article talks about, but ... Bellard did this in 15 seconds, in 2004 with a 2.4Ghz Pentium 4. In the 7 years since, I except a 4-fold increase in speed, so on modern hardware it will probably be 4 seconds, maybe 1 or 2 on the beast described in this article.

> Bellard did this in 15 seconds, in 2004 with a 2.4Ghz Pentium 4

But isn't Linux kernel a lot bigger nowadays?

He did it with his own C compiler (tcc).

See: http://en.wikipedia.org/wiki/Tiny_C_Compiler#Compiled_progra...

There are tradeoffs for compiling very fast: Slower program execution.

The base kernel is not a lot bigger. There are about four times as many drivers. (But bellard's compilation only included the ones needed, IIRC. google tccboot for more info)

Ah, compiling C is so pleasantly fast.

Wondering how fast it could build Android 4.0. (http://source.android.com/source/building.html) Bit of a pun intended. I have recently realized that a 4-core smarthpone, one of those coming in mid-2012, on a 1.5 GHz (e.g. the rumored HTC Edge) theoretically packs similar power to that of my dual-core Macbook Pro 2.8 GHz (4x1.5=6 vs. 2.8x2=5.6). One probably cannot compare OMAP with x86 too much, but still, the CPU developments in mobile space excite me more than the desktop ones, which frankly, disappoint me a bit. Somehow, sometime in 2000s I expected to have massively parallel CPU sometime around now, not a mere 6-core.

A desktop CPU probably has about an order of magnitude more power per cycle than a mobile chip.

They have bigger caches (less time spent waiting for memory), more logic units, deeper pipelines, and can handle far more 'in-flight' instructions.

That said, the smart phones of today probably do have more power than desktop computers of about 10 years ago.

edit: now that I think about it, a order of magnitude is probably generous, but I would still guess that it's at least a factor > 2.

I don't have time to get better data, i.e. I couldn't find any MFLOPS data for a mobile CPU (not GPU), so just quickly unscientifically:

Texas Instruments OMAP4460 (in Galaxy Nexus at 1,2 GHz): 2047.7 BogoMIPS

1.7 GHz Xeon from 2006 (the small EC2 instance, single-core): 5203.55 BogoMIPS

Would be interesting to see some real benchmarks though..

You dont want flops figures - no floating point is used to compile.

Bogomips lives up to its name.

I suspect that compiling has fairly poor locality and is quite cache dependent, so you might be better off just benchmarking on a mobile cpu.

Not a great test either but better then the above:

  Geekbench 2 on iPhone 4S: 617
  Geekbench 2 32bit on iMac i7 @ 2.80GHz: 8557

Honest question, because I'm actually curious about this. Are these extreme edition chips used almost exclusively by enthusiasts, or are they actually used in certain software engineering branches?

Most enthusiasts don't buy top of the line. It's way cheaper to buy the "second best" and overclock your way to the Extreme Edition's performance (which is really easy with the new chips).

The second best in this case is i7-3930K, same 6 cores and architecture (just 100MHz difference) but with less cache (12MB vs 15MB) at half the price.

Thanks for the response. That makes sense, obviously you can pay a lot less and get more by overclocking the CPU. Then my question is, who actually buys these chips, and for what purpose?

Well, isn't this basic differential pricing? If there are customers who are willing to "pay as much as needed" for "the fastest chip they have", why wouldn't Intel offer it? It shouldn't be too expensive to "brand" yet another model, and there are enough enthusiasts to shell the required cash on this thing. $1K isn't that much, really.

the ads on phoronix are getting pretty over-the-top

Reply on news.ycombinator.com