Socks and Sandals

If Intel and Apple merge, that’s my vote for the new name. ;-)

Cringely has some thoughts, but I think he’s mostly wrong, in part due to his misunderstanding of the state of technical affairs on the Apple-IBM side of things. The Altivec was only an advantage for certain applications on OS X and Apple has egg on their face by proxy due to IBM’s failure to get a 970 into a Powerbook. A commensurate feat, by the way, which Intel will have no problem with given their Centrino platform. Hannibal has some theories as to which CPUs we will see in the coming months and years in the Apple line.

Now Jobs does have that famous RDF (Reality Distortion Field) of his, but I don’t see this move as spite for IBM or an attempt to dethrone Microsoft. I see it as Apple finally moving to a platform that will allow them to garner a large amount of developers and one that opens up the possibility of multiple platforms to them (e.g. games, PDAs, servers that people might actually buy, etc). Apple is rapidly moving away from being a “computer” company and turning into a media services company. As such, they need a platform that is sufficiently flexible in order to provide the proper experience for customers. Sticking with IBM means that they are letting IBM call the shots on that score. I’m guessing Jobs no longer considers that acceptable.

I’m guessing they don’t really care about the Osbourne Effect because there are actually two types of people that won’t buy until the Intel’s come out:

• Those upgrading from a PPC-based Mac
• Those who have lost their last excuse to move off of Windows

My guess is that Apple will take the loss of some sales now to increase the overall sales and power of the Apple line by garnering a good customer base of people who are aggravated with Microsoft, but not so much so to switch to Linux. So, in that sense, Mr. Cringely, announcing this a year early does make sense without a merger. Apple is spotting disgruntled Windows users a year to come up with the loot to make the switch to a Mac with a guilt-free conscience.

I found a blog post today regarding Macs not being more expensive than the counterpart hardware running Windows. While the majority of the assertions in this essay appear to be within reason, there are some technical facts that are just plain wrong.

Windows 32-bit chips can address up to 3GB of RAM, and though that seems like a lot, it’s a puny amount compared with the theoretical 1,000GB capability of a 64-bit chip.

Negative.

aluminum:~> bc
bc 1.05
Copyright 1991, 1992, 1993, 1994, 1997, 1998 Free Softw...
This is free software with ABSOLUTELY NO WARRANTY.
For details type `warranty'.
2 ^ 64
18446744073709551616
. / 1024 / 1024 / 1024 / 1024
16777216
quit
aluminum:~>

That’s over 16 million terabytes. I’m guessing that the author has not been introduced to exponentiation.

Second, Microsoft just today released 64-bit versions of selected Windows operating systems. This allows Windows users full 64-bit computing, if they need or want it. As well, Linux and the BSDs have both run on 64-bit architectures long before Windows and Mac OS X, both. Simply supporting 64-bit architectures is no longer a deciding factor in an OS purchase decision since they all now have support for it.

As well, the entire discussion of 64-bit architectures is a bit of a red herring, since:

1. The G5 is the only fully 64-bit chip available in Apple’s whole line, and
2. The full power that 64-bit computing can yield is not yet available in hardware

This first point is illustrated as follows running some code on my 15-inch PowerBook G4:

aluminum:~> ./test
word size: 4 bytes
aluminum:~> cat test.c
#include <stdio.h>

int main( void)
{
     unsigned long *a;

     printf( "word size: %d bytes
", sizeof( a));
     return 0;
}
aluminum:~>

If the G4 were 64-bit, we would have seen ‘8’ in place of the ‘4’ that was printed.

The biggest technical issue with this post has to be this statement:

Basically, for every 7 steps a RISC chip (like the G4 or G5) takes in completing a process, a CISC (complex instruction set computing) chip like Intel’s or AMD’s takes about 15. This does not translate precisely to a simple conclusion such as “Divide a CISC chip clock speed in half to see a clock speed that’s comparable to a RISC chip,” but it’s close. What it does mean is that a consumer cannot look at a Dell laptop processor speed of 1.67 ghz and think you need an Apple processor speed of 1.67 to be comparable. In fact, a Dell laptop processor speed of 1.67 is roughly equivalent to an Apple processor speed of 1.0 ghz.

This is a gross oversimplification. If a CISC instruction can do more work than a RISC instruction, and the clock speed of the CISC chip allows it to execute more instructions per unit time than the RISC chip, how could it take the CISC chip more “steps” to complete a process?

Note: ISA = instruction set architecture

The answer comes from several points:

1. x86 is an extremely register-poor architecture.

Without enough registers available for some simple tasks (forget about complex tasks), the x86 compilers are forced to generate code that hits main memory more often than one would like. This reduces performance because it forces more cache misses and DRAM access is slower than cache access. Put another way, the register-rich PPC architecture actually makes better use of the caches by keeping them hotter because code can keep temporary values in registers rather than cluttering up the cache with tempvars that will become stale frequently.

2. Some instructions on a CISC ISA take more than one clock cycle to complete.

Even though the newer x86 chips can issue more than one instruction per clock (superscalar execution), some instructions take more than one clock cycle to complete. There’s nothing that can speed that up; this is inherent in the complexity of a CISC ISA. More instructions doing more work means that each instruction is implemented less than optimally.

3. Optimal performance on the x86 requires that the pipeline be full all the time.

The pipeline on the x86 is extremely long, and when it stalls it causes a major performance degradation because the pipeline has to refill before the CPU can execute those instructions. Code that defeats x86 branch prediction will run poorly for sure. On the other hand, a RISC ISA is more forgiving of pipeline stalls.

The most important thing to note here, though, is not technical: Intel spent a fortune on advertising that reinforced the link between clock speed and quality of a CPU in the mind of the consumer. They reduced it to one metric that was easily bandied about at the watercooler and in doing so, pulled off the greatest technical marketing coup of this last decade. Anyone who has tried to convince a semi-technical man in his 30’s or 40’s to buy an Athlon instead of a Pentium will be familiar with how successful Intel was in this endeavor.

On a more editorial note, I would have liked the author to have compared more than just Dell. While Dell is the biggest target, there are others that employ the same tactics. As well, it would be interesting to do this comparison with PCs offered at brick-and-mortar outlets like Best Buy or Wal-Mart. There are generally less confusing options available in a store, if you’re willing to endure a higher-pressure sales experience.

I hope that people don’t start quoting the original author’s post verbatim, since it could easily be chopped apart by those who would do so.

I just tried Acquisition, a new P2P app from the same guy who brought you NewsFire, Mr. David Watanabe. It looks nice, but I already uninstalled it since it crashed twice in 30 seconds, before I could even attempt to download anything. I also don’t really understand why he’s charging for a beta product, but at least you know he’s going to charge for this one from the start. ;-)

Here’s an excellent interview from DrunkenBlog with John ‘The Wolf’ Rentzsch, Mac hacker extraordinaire. Really interesting stuff about mach_inject and mach_override, two of John’s many super hacks. Check out the links at the bottom for more info.

Also linked from that article is another DrunkenBlog gem; a dissertation about why SMP is made more difficult by Apple’s choice to bolt FreeBSD onto a Mach microkernel. DrunkenBlog continues to impress.

It sure looks like Microsoft is up to their old tricks again; in this case, the old tricks are stealing the latest Mac interface innovations to put in Windows. I wish they’d at least try to come up with something original… maybe they’ll buy Apple by the time Blackcomb comes out.

Two people replied to my last post about NewsFire’s move to shareware (one of them the author of the software, Mr. David Watanabe). Apparently, he had always stated that NewsFire was to be shareware; he just wasn’t actively charging for it until this latest version, 0.9. Now, I can’t confirm this was the case for myself via Google cache’s or the Wayback Machine, so someone providing a link to the source in question would be nice, but if that truly is the case then I will offer my apologies to Mr. Watanabe.

The Mr. Watanabe I am referring to is the author of NewsFire, a Mac OS X RSS reader. The bad move that he has made is switching his software to a shareware model, where it was previously freeware. I just switched back to NetNewsWire Lite this morning after finding out.

Now, I don’t have a problem with the guy making money from software; this is, in fact, how I make my own living. My issue is with the fact that there was never any notification that NewsFire might become shareware (i.e. for-pay). However, this increasingly common trick of publishing freeware that suddenly becomes shareware after there is a large-enough user base is not very endearing.

A more honest approach would be to make it shareware from day one and just increase the trial length in an effort to “hook” that potential revenue. Also, a software could have two versions: one full-featured version that is shareware or just plain costs money and another, partially-featured version that costs nothing. This enables users to become familiar with your product and decide on their own schedule whether or not its worth it for them to buy the full-featured version. This is what Ranchero does with their NetNewsWire product to good effect. I, as a software consumer, know that NNW Lite will be free to use, if somewhat limited in functionality. If I should need to edit blog posts from inside my reader or get other advanced functionality, I’ll pay for regular NNW. In fact, I’m using a shareware blog editor to write this very post! But, the important point is that I’m using it because it never claimed to be anything other than shareware.

Like I said, I don’t mind paying for software or having an author charge for their hard work in creating quality software. I do have a problem with what appears to be dishonesty. This trick is loathed when big corporations do it, but it seems that people are much more willing to let it go from small-time operations like Mr. Watanabe. At any rate, I’ll definitely be very wary of using another piece of “freeware” from Mr. Watanabe, now that I know what could happen after I get comfortable with it…