Intel Developer Macs Outperform G5s

bonch writes "Developers working with the new Intel-based, developer-only Macs are impressed with the performance. The machines take as little as 10 seconds to boot from Apple logo to desktop, and apparently run Windows XP at 'blazing speeds.' Rosetta tests demonstrate the PowerPC-native build of Firefox running just as fast as it does on a high-end G5."

Comments

Boot times disk/network bound

[Toby The Economist]

OS boot times are usually disk and network bound.

I don't see how even an order of magnitude increase in CPU power could shorten boot times to the extent described here.

There must be other factors.

--
Toby

Re:Boot times disk/network bound

[hazee]

I've never understood why it takes so long to boot a computer. It takes far longer than the time necessary to transfer the actual amount of data that ends up in memory after the boot sequence.

Why not use something more like a resume from hibernation, where you just restore the contents of memory directly from the disc in one go and be done with it?

Actually using hibernation rather than booting is no good, becuase it only restores you to the state that your computer was in last time you used it, which might not be "clean". For example, if you had been running something with a memory leak, hibernation won't fix that.

But the solution is simple - instead of writing the hibernate data to disc just before you shut down, instead store it just after you've finished booting, so that you've got a "clean" system ready.

This way you get a "clean" system every time at the speed of a restore from hibernation. (And if something goes wrong, you still have the option to do a "full" boot.)

Some might say that you need to go through all the slow processing of a full boot in case anything's changed. Really? Restoring from hibernation seems to cope with that possibility.

More likely, most times, nothing will have changed. And for the times when it has, well, you do the extra configuration necessary after the restore - you're still no worse off.

So why are we still forced to sit through full boot cycles?

Re:Boot times disk/network bound

[imsabbel]

Your logic has a flaw:
A resume form hibernation has to read ALL of your memory from disc, as only saving "used" memory frames would just invide bugs from hell. (think about memory windows from drivers, or what is "clean state" ? which autostarts/services,ect)
And with 1GB, even on a very fast HD it would need 20 seconds... Not faster

Also consider that bootup is usually the time to detect new hardware.

In fact, im quite happy with the 20-30 seconds i get with windows xp.

Re:Dual Boot

[uncle_fausty]

For God's sake, will you please stop beating this issue to death? No, MacOS will not officially run on non-Apple hardware. Yes, l337 h4x0rs will probably find a way to make it happen. No, it will not be the rosy seamless computing experience MacOS provides on controlled hardware. Apple's success in OS development is in no small amount tied to their control of the hardware it runs on; don't expect that to go away anytime soon.

Re:Dual Boot

[keytoe]

the final product will be tied to an special Intel DRM chip that will prevent it from running on other machines.

I keep seeing this spouted around as fact, but I have yet to see anywhere where this was stated in any official capacity. They may do that. They may not - they haven't said one way or another. Outside of the iTMS, Apple hasn't done much with DRM, and I'm pretty sure that was only to make the labels happy - so I'd say it's still pretty much up in the air.

So stop saying it like it's a fact, please.

Re:64bit and vector code

[jiushao]

Actually you have it a bit backwards. The P4 is rather back to the roots for Apple, an excellent SIMD performer with a bit so-so performance on linear floating point. It was the initial issue with the P4, a weak x87 implementation with a great focus instead on the excellent performance of the SSE2 (and SSE3 with the prescott).

Compare this to the G4, another weak linear performer that Apple more or less specialized in getting to fly through good use of the excellent Altivec unit. The G5 on the other hand has a somewhat weak Altivec unit but a much beefed up set of single-element FPU units, yielding so-so vector performing but good linear performance. IBM did probably not focus much energy on the Altivec unit but instead threw it in since Apple required it (after all, the single-element FPU performance of the G5 almost puts the Altivec unit to shame).

Some might now be quick to point out that Altivec is a nicer instruction set than SSE2/SSE3, this is by most standards true, but if you are hand-coding assembly you can make do with either. On the flip side Intel has quite impressive auto-vectorization support in their compiler.

So, what does this add up to? The G5 is in a good place for beating the P4 on unoptimized unvectorized code, but the P4 really screams if things are tuned up a bit. Considering Apples history with Altivec I think we can safely assume that they won't be afraid of doing some hand-tuning to get good perfomance.

This all ends up looking quite favorable for the P4, I still don't think we will see a commercial Mac with a P4 derivative in it, but anyone who thinks the P4 is a weak performer has another thing coming. For a bit more on my opinion on the state of the x86 vs. PPC today see my earlier post in the "Apple Switch to Intel Not a Big Loss for IBM" story.

Re:Dual Boot

[drsmithy]

Steve Jobs also said that Apple had no plans to compete in the $500 computer market and then later Apple released the Mac mini.

But by the time they did that, the $500 market had become the $250 market...