Slashdot Mirror
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."
Although, Firefox doesn't run particularly fast on my G5 compared to my run-of-the-mill XP box at work.
It'll require a Mac brand intel, this has been announced. Whether this is through some kind of magic BIOS trick or something, or just by only producing drivers for exactly one set of hardware isn't known for sure.
G5 runs floating point and vector operations faster than Intel chips, and can address more memory (without weird segment changes) than Pentiums of the same vintage. Thus they run scientific code and large media manipulations faster than Pentiums of the same vintage. They do not run branchy program logic code faster than Pentiums of the same vintage. Shockingly, most users spend most of their time browsing the web and writing emails, not running simulations or media transcodes; those that do are going to Opteron based systems.
No. The version of OS X on the developer Macs may be compatible with other PCs, but the final product will be tied to an special Intel DRM chip that will prevent it from running on other machines.
The developer machines are loaners and will go back to Apple in two years, and will not continue to be supported.
omnia tua castra sunt nobis
I get 7 hours on my p4-m notebook.
so I'm thinking they may see an increase.
-nB
whois gawk date unzip strip find touch finger mount join nice man top fsck grep eject more yes exit umount sleep dump
"Since Intel has always been shaky in floating point "
I am not defending Intel, but your statement is wildly inaccurate. They did have an issue once and that was only with some purposefully obscure math. It was very difficult to reproduce and Intel recalled those processors. That was years ago. So I am not sure what you base that statement on.
I think by transparent they mean that it runs in OS X windows and looks and feels like an OS X application rather than the speed. Like, with Classic, all of the Apps ran after the Classic environment booted and they used the old Mac OS 9 widgets and windows and were just very seperate from OS X.
Methinks you've been had. This is an update of a quite famous rant made about one of the earlier PPC Macs, the 8600/300, as found here.
And tomorrow the stock exchange will be the human race
Memory test.
--
"Open source is good." - Steve Jobs
"Open source is evil." - Microsoft
It's likely that Rosetta is pulling a lot of tricks, I/O level and otherwise, and when you're reading data from a hard disk or the network, using a few spare CPU cycles to swap some bytes around isn't going to be noticed *at all*, because if you're doing that sort of I/O bound processing, you'll likely have plenty of CPU cycles laying around to swap bytes in memory.
If you're not talking about Rosetta, but about multi-platform applications more generally, the long answer is in the Universal Binary Programming Guidelines PDF. The short answer is that you abstract away from your own code byte-order issues where possible, and where not possible, you otherwise have separate code paths that do the right thing depending on the targeted platform.
Seasoned developers who coded applications under NeXTStep have been through this once before, remember. This is not a new problem. We've been here before... I avoid binary data formats where it's reasonable to do so.
> They did have an issue once and that was
> only with some purposefully obscure math.
Um, it was a division. Hardly obscure math...
If the divisor has some particular binary
pattern the result was incorrect. The problem
was that you could not *trust* the computer,
because you did not know if a particular Excel
spreadsheet hit that binary pattern or not.
Intel indeed recalled the chips, after first
denying the existence of the bug.
This problem was resolved about 12 years ago when NextStep shipped on Intel.
The IO libs just take care of it.
Also, it's worth noting that binary storage in a "native" format is always a Bad Idea and most serious projects would be expected to avoid it.
The reported threading problems were not, in fact, threading problems at all. Instead, they are the result of F_FULLFSYNC fcntl. You will probably find this thread interesting.
Surely the OS has to know what memory is in use, otherwise there would be chaos. How would it allocate blocks of memory if drivers were "secretly" allocating their own blocks?
As for detecting new hardware - that's such a rare event that it should be treated as an exception, rather than the norm. Most days I boot my computer, there's a marked absence of strange new hardware...
Both techniques (discarding unused memory and compressing used memory) are used by windows xp, which is why it is so fast to hibernate/resume (less than 10 secs for my p3/512mb laptop)
2 2GHZ processors does not equal 4GHZ. The percent performance gain for each additional processor is rarely even close to it's ghz rating, and it goes down sharply for each additional processor. You have things such as system bus and memory contention when you add multiple processors as well. AMD does a far better job with this in regards to the Opteron than Intel does with the Xeon. I think Apple/AMD would make a very interesting combination.
trippling disk i/o will half boot times.
http://overclockers.com/tips00806/
and trippling CPU speed will NOT half boot times, but maybe reduce by 1/3, showing that boot times are more dependent on disk i/o than CPU speed.
in a typical setup, CPU time is a MAJOR bottleneck, only to be matched by device initialization, which has little to do with cpu speed.
in fact, a k6-2 500Mhz machine will boot windows xp nearly as fast as a athlon xp 2500. the athlon xp machine takes about 2/3 the time as the k6-2 500Mhz with the same hard disk
i find the 1st 1/2 of your post completely incorrect, and the second 1/2 quite true.
Apple did do a lot to improve the boot time: there's a boot cache, a kernel extension cache, hot file clustering, prebinding, on-the-fly defragmentation and more. Even old macs (e.g. 300 Mhz G3's) boot up OS X fairly quick.
Ethical concerns aside, for many companies Intel is such a desirable choice because they do the testing with components and certify a complete package for you. That's expensive work, but Intel does it because they know that people will pay for it.
Sure, I'd love to see AMD in macs. I have fond memories of my last AMD machine. But, only if AMD can give us a good, inexpensive motherboard that they certify and test. If AMD wants to compete on this front, they're going to need to offer a package deal.
Back to ethics, it's entirely possible that Intel has crossed a major line and is due for a DOJ-powered slap, but I'm going to reserve judgement on that until I see more evidence. Just because AMD says they are being anti-competitive doesn't prove it. We'll know soon enough.
Slashdot. It's Not For Common Sense
So if you think about waiting for an Intel PowerMac, think 1.5 years. But even then, Apple is not reknown for getting the first revision right, plus they have a major platform shift to deal with. And it will take years to have an Intel-native software catalog the size of the current PPC one.
The MWSF refresh could bring dual core dual processors. Or maybe not. But anyway it won't be too much of a difference from the current line, so I think now's a good time to buy.
Tuff that Smatters.
Rather than using DRM, if they use specific altered BIOS and/or a specific chipset then that is all that is necessary. The OS will not have the drivers/ability to work on other hardware.
Apple will most likely still be designing their own MB's, even if they do include Intel chips, so this is a very easy way for them to maintain sales on their own, still technically proprietary, machines. DRM may be used, but if it does then it will probably be a second line of attack, not the primary.
Shawn's Tech Articles
Altivec is not so much poor at double-precision calculations, as not supporting them at all.
It's been a few years since I programmed for the Altivec unit, but it didn't support any doubles then, only single-precision floats.
Apart from that (and it is a big thing), I've never heard a serious comparison of Altivec with SSE that resulted in SSE coming out equal or even close. Altivec is pretty damned amazing.
Shame we won't be able to play with it any longer.
(By the way, Apple's transition documentation shows how to convert from Altivec to SSE, and after reading through it briefly, SSE doesn't impress me. Hopefully raw clock frequency will overcome SSE 'suckiness' and we'll see good use of it in media apps.)
The PCI cards identify themselves with a vendor/device code that is part of the card. Some vendors are lazy and do not create a new revision for a new card. In these cases you have to perform extra gymnastics with the card to determine which variant it is and which initialization code to use. Keep in mind too that even if you have no addin PCI cards, many of the devices on the motherboard are still on the PCI bus. You need to initialize those bridgechips, those firewire/USB chips, serial chips, parallel port, built-in network i/f chips, etc. It adds up after a while and boot time is not something that most hardware vendors /device driver developers pay much attention to.
That's not a P4-m, that's a Pentium M. Similar names, big difference.
http://img82.imageshack.us/my.php?image=avvio18ep. jpg . jpg . jpg a 5zc.jpg t ot3ej.jpg d intel2pi.jpg y s5np.jpg o nt3ke.jpg e l6dl.jpg e c.jpg h d5iw.jpg j pg t e8gx.jpg o 4ta.jpg 1 vo.jpg t 4kw.jpg g c.jpg q n.jpg s y.jpg l 3id.jpg l xcode2wl.jpg o st3ri.jpg
http://img86.imageshack.us/my.php?image=avvio21sl
http://img86.imageshack.us/my.php?image=avvio31gz
http://img256.imageshack.us/my.php?image=barracud
http://img256.imageshack.us/my.php?image=centrale
http://img256.imageshack.us/my.php?image=dashboar
http://img256.imageshack.us/my.php?image=devtrans
http://img24.imageshack.us/my.php?image=internofr
http://img24.imageshack.us/my.php?image=macityint
http://img24.imageshack.us/my.php?image=pentium42
http://img24.imageshack.us/my.php?image=quicktime
http://img24.imageshack.us/my.php?image=retro7dz.
http://img109.imageshack.us/my.php?image=retropor
http://img17.imageshack.us/my.php?image=schedaalt
http://img17.imageshack.us/my.php?image=schedadvi
http://img17.imageshack.us/my.php?image=softdevki
http://img17.imageshack.us/my.php?image=sysprof18
http://img17.imageshack.us/my.php?image=sysprof24
http://img33.imageshack.us/my.php?image=sysprof35
http://img33.imageshack.us/my.php?image=tigerinte
http://img33.imageshack.us/my.php?image=tigerinte
http://img274.imageshack.us/my.php?image=ventolep
Bye!
SeqBox
If you turn on the 'boot logging' feature of Windows and examine the logs you can see where a lot of the time is spent on your system by the OS.
If your default boot disk is available then a normal Mac doesn't take anytime looking for other boot devices. These Intel dev Macs do support booting from USB drives so that is not the case anyway.
What makes you think they support booting from USB drives?
OSX kernel extensions are very rare. Almost no program uses them except Norton products(anti-virus and disk doctor). I recommend staying away from Norton stuff for this reason and using Alsoft's Disk Warrior. Third party kernel extensions are a bad idea on any OS.
Uhm, I hope you realize that apple includes many as well that aren't exactly usable on these Dev kits. Such as Bluetooth kexts, ATI kexts (or Nvidia kexts), Airport kexts, netboot, FWTDM kexts, Audio kexts (there are 8 audio kexts loaded on my G5), CHUD kexts, fan control kexts, slew, voltage, sensor kexts, and other kinds of kexts that either lack the hardware or software support on the Intel Dev Kits.
Then for third party kexts there are Logitech Drivers, Norton Utilities kexts, Virtual PC kexts, the Ambrosia kext, DiskWarrior kexts, and many other third party drivers and kexts that shouldn't be loading at startup and shouldn't even be kexts but are.
What features are lacking?? The Intel dev Macs have Firefox, iPhoto, iDVD, and Quicktime installed. The average user may install some extra dashboard widgets and a driver or two, but I doubt that would add more than a couple seconds to boot time.
What makes you think these dev kits have either iDVD or Firefox installed on them? Did you see iDVD in use during Steve Jobs' WWDC keynote?
See above for a large list. You don't need to load a kext for hardware that doesn't exist.
I think it will impact sales per se, but not any more then it is already affecting sales. Let me explain.
I bought a 15" Powerbook in September 2003 - this system is a G4 at 1.25 GHz. I love it but I would really like something faster (and with a better graphics chip) so it could be a real desktop replacement. But, this simply doesn't exist. Today's fastest powerbooks are still G4s at 1.67 GHz - an increase of just 34% in nearly 2 years. This isn't enough to make me buy a new machine. The way it was looking, I was waiting for Powerbook G5s - but it wasn't happening (and now, of course, it won't happen).
I am guessing that the significantly faster machines (both desktops and notebooks) with significantly help Apple sales, but will not hurt them more than the lack of speed was already. Increases in performance will correlate to sales, and if IBM was unable to deliver but Intel can, than I think it will help Apple immensely. If people need an Apple box they'll buy one, but right now they're just too slow or too expensive for people to consider (i.e., the fast machines are too expensive).
I look forward to finally replacing my Powerbook with a nice speedy Intel-powered machine in a year or two, and I bet many Apple users will be with me. The new speed will then make it a lot easier to get new switchers on board.
The space unintentionally left unblank.
G4 Optimized Firefox 1.1 pre-alpha nightlies (fast!)
http://homepage.mac.com/krmathis/
"Apple has really been pushing people to Sleep the machines instead of turning them off."
Then they need to fix the crash-on-wake problems. The absolute most reliable way to crash my dual processor G4 is to wake it up after it's been sleeping for awhile. If it doesn't kernal panic, then the USB ports don't come up and the keyboard and mouse do nothing. Then it's time to push the reset button.
Anyway, it boots faster than the XP box at work, so I'm not worried about booting every day.
I just so happen to be reading Gravity's Rainbow...
For instance, as ubiquitous as Creative Labs hardware is, Apple can't support it because it's really low quality compared to Apple's on board audio (designed by Bang Olufsen)
I won't get into the debate about the quality (or lack therof) of Creative's products, except to say that there is a thriving market for aftermarket audio products for the Mac platform (M-Audio being one such vendor).
My point of contention with your post is your assertation that Apple employs the aid of Bang and Olufsen for its built-in audio. While searching on Google reveals others using superlatives that compare Apple's design to the afformentioned company's products, there is no evidence of collaboration. There are aftermarket products made for lines such as the iPod, but that's as close as it gets.
A few more little known facts: Apple uses all Bose speakers even down to the smallest speakers in their systems. And Bose designs the case section where the speaker will be housed in order to provide high quality, room filling sound no matter what system you have.
Furthermore, Bose has only had a limited relationship with Apple, starting and ending with the PowerMac 6400 family. And for the record, the logevity of this particular piece of design has been lacking. Apple has collaborated with other companies on their speaker design, such as utilizing Harman Kardon enclosures/speakers starting with the iMac DV.
Your references to startup tones are tangential, more a matter of taste and style than metrics. Apple succeeds in the the fields of arts primarily because of their decent first-party applications, and additionally because of their ISVs and aftermarket hardware. Many creative types still call the Mac home, and I don't think it's because of trendy start-up tones or hardware companies.
As to Apple's success and appeal, I wholly agree it is due in large part to the bottom-to-top control they have of the platform. I may have made mistakes in calling you out on some of these facts you are presenting. I have spent nearly 10 years working in and around Apple's various offerings, and have admired their industrial design (with a number of exceptions). That said, I would like to think that whatever zeal that I may have for them is grounded in reality. If you can find reference to any of your above claims, I would appreciate that they be presented for sake of perusal.
This article is thin on everything. In fact, it's little more than a mutated form of the inevitable discussions of increased "snappiness" that occur every time Apple updates either hardware or system software. The information in the article is all vague: "as little as 10 seconds," "It's fast," etc. Most ludicrous of all is the claim that the PPC build of Firefox runs just as quickly on the x86 Mac as it does on native hardware. Bollocks, sez I. Rosetta's nice, but it's no replacement for native and never will be. Like Classic, it's value will diminish with time. It's intended to ease a transition, in this case to universal binaries. When Apple deems that transition complete, Rosetta will, I think, be deprecated, if not abandoned altogether, barring any decisions in Cupertino to switch to, say, sparc.
I've got access to a Mactel dev box, and the performance is good, but it's not so much better as to be revelatory. Compiling the source for several projects I work on is faster on the dual G5 2GHz machine than the Mactel (gcc4 on both machines). While not a great measure of performance, at least it's tangible. Of course, if you prefer to accept the nebulous claims from one of several notorious Mac rumor sites, be my guest.
AltiVec emulation can still be made faster than the linear code path.
The reason for this is that the AltiVec emulation can avoid many of the problems with the linear code. To list some...
* AltiVec has only alligned loads and stores, so you do not have to worry about emulating cross-page accesses
* AltiVec can load more information into your register map faster than your generic code (less address translation per word loaded/stored)
* AltiVec allows the emulating processor to parallel execute many of the instructions of the mathematical operations. So, you can make better use of the superscalar design of the emulating processor
There's a ton more, but I hope you get the idea. With PearPC there's a noticable increase in performance from ~1 MOPS for scalar integer arithmetic, and worse even MFLOPS for scalar floating point arithmetic to ~300 MFLOPS with even the scalar AltiVec emulation code.
Everyone seems to think emulating PowerPC and AltiVec on x86 and SSE/SSE2 (SSE3 provides no useful operations for AltiVec emulation) will make it slower than dirt. People think that emulating AltiVec with scalar operations will only slow the emulation down.
Why don't people actually look at some empirical data first, befor making such claims.
I am unamerican, and proud of it!
* Enhanced Write Filter
Have fun: Join D.N.A. (National Dyslexics Association)
I am not very knowledgeable in this area, but it was explained to me by one of the techs where I work that the Windows registry is the culprit for this. Being a binary file that is subject to fragmentation, the very act of installing software adds to the registry (of course), which then in turn causes an across-the-board speed hit every time some Windows function calls said registry... which is often. Educate me if I am wrong about this.
If Jesus wants me it knows where to find me.
Nope.
...and iTunes is still running through Rosetta.
The shipping devkits only have iPhoto and iTunes installed.
iMovie an iDVD work nicely as PPC binaries loaded from a standard OS X/iLife install.
"Now those of is in the here see that as nonsense, since filling up the hard drive with not-currently-executing code does not have any impact on memory usage or CPU usage."
Sorry, but you're wrong. Installing MS Office on a Windows machine loads a TON of resident and shared DLLs that just sit on the system and use resources. The more apps you throw on, the more shared files show up. The more files that show up, the more need to be processed by the OS when loading DLLs.
Take a clean XP machine. Measure its boot time. Install all of the programs I mentioned in the first post. Remove any of them from startup. Reboot and time how long it takes to get into Windows. I guarantee it takes five to ten seconds longer to boot than on the clean system. Such is the way that Windows processes its system files and installed applications.