Slashdot Mirror
Pentium-Based Macs The Future of Apple?
seek3r writes "Found this interesting article on BusinessWeek.com regarding Apple's potential switch to Intel chips. I wonder what the implications this might have for Apple with regards to market share and software support. Have Motorola's chips really lagged behind Intel?"
...Or am I missing something? the G4 chip has been around for a long time...
There is a world of difference between using an x86 processor, and using an x86-based PC. I suspect Apple has closely investigated hooking up an x86-64 (or two ;) to a custom motherboard/infrastructure that would solve many of the interrupt/expansion complexities of PCs. For example, they could adopt HyperTransport, which would make multiprocessing affordable, easy to design around and most of all, leading edge, which is important to some people. Remember, Apple is expert at getting a lot out of a little - it would not surprise me if they tied a Hammer to a custom motherboard and created a whole new architecture. And that wouldn't be a bad thing.
The RISC vs CISC argument is all-but dead in the water by now. x86 chips are only CISC in the loosest sense of the word, for backwards-compatibility. They all run internal mini-RISC chips that convert CISC commands to RISC sequences via microcode.
But we're hitting the point where few people [*] can tell the difference between 1GHz and 2.8GHz and even hardware engineers are starting to realize this, so maybe it Just Doesn't Matter.
Definitely. PC manufacturers love to compete on Mhz, but a fast CPU is useless if it's starved of useful work by bottlenecks in I/O, memory bandwidth, etc. It's not unusual for a sub 1Ghz PC with good SCSI disks to handily outperform a 2Ghz+ machine with mere IDE.
Sun, SGI et al realized this years ago. Serious computing is limited not by clock speed of the CPU but by bus and memory bandwidth. That's why Sun sell systems with 300-400Mhz processors and gigaplane XB crossbar active backplanes. Nowadays with the increasing sophistication of consumer software (like the latest games), the same issues are recurring.
If you're buying a system in the near future, drop 500-1000 Mhz in CPU speed and buy faster disks or more memory with the money you saved.
From the article:
No.
I don't want a laptop that blows hot air like hair drier or desktops that have three fans. As people realizing (as another poster mentioned) the CPUs are fast enough, I don't see much point in abandoning the PowerPCs that are small, consume little energy, and hence run so much cooler. For me, computers that are quiet and cool are much preferable to the opposite.
Another thing the author of the column seems to forget is that PowerPC is not a chip solely from Motorola. The point that IBM is also a partner and develops PowerPC chips is completely missing.
The whole RISC vs CISC argument is completely outdated. There is no such thing as pure CISC anymore. Please get with the times. You may love mac and that's fine but at least use up to date reasons for why mac is superior to pc. That's about as bad as a PC user dumping on macs "cause they only have one mouse button".
- Toby
Such a move on Apple's part would complicate matters significantly. Consider that if hardware devices would STILL need mac specific drivers to meet whatever "hardware security" apple uses to make their machines proprietary--Meaning much hardware STILL won't function with OS X, whether it's on top of Intel or a PowerPC proc from Motorola or IBM.
My favorite uninformed reader was this guy:
This guy doesn't understand the term "switch." If he starts off running an Intel PC, and buys an "Intel mac" what has he really changed? Still using the same ancient hardware architecture kludged on top of a 32-bit chip sucking more juice that an a electric battleship.
Who did what now?
The only real advantage of the PPC at the moment is that it lacks a lot of backwards compatibility cruft and, because of its RISC design, consumes less power and spreads less heat. It is a fine notebook CPU (and Apple is a fine notebook manufacturer). But Apple seems to have had no other chance but giving up this advantage by selling its newest line of desktop G4 Macs with dual CPUs, keeping up with Intel at least halfway with such a "hack".
gopher://cramer.plaintext.cc http://cramer.plaintext.cc:70
It's not unusual for a sub 1Ghz PC with good SCSI disks to handily outperform a 2Ghz+ machine with mere IDE
If you said "a clustered array of RAID5 15,000 RPM drives versus a 5400RPM single drive", then that would have made sense, but to use SCSI versus IDE as the big differentiation is just silly: The intrinsic SCSI advantage has been disproven countless times.
Sun sell systems with 300-400Mhz processors and gigaplane XB crossbar active backplanes
That's pretty disingenuous: Sun sells systems with tens or hundreds of those "300-400Mhz" processors, disproving your "CPU power doesn't matter" BS. I guarantee you that if Sun weren't sliding behind in the CPU game (it's hard to compete with AMD and Intel with such a small niche market) they'd sell much more powerful CPUs. Instead they compensate by clustering dozens of them together.
If you're buying a system in the near future, drop 500-1000 Mhz in CPU speed and buy faster disks or more memory with the money you saved.
You'd save next to nothing. An Athlon 2200+ costs $220 Canadian here, and puts you in the upper realm of CPUs. Considering that most power PCs have 512MB of RAM (which is virtually never exhausted. Despite having several development tools open, and SQL Server running, and several different browsers, I currently have 370MB free. Adding more memory will merely increase the capacitive load of my PC). Secondly, adding a faster disk only matters if you do tasks which are heavily disk I/O intensive, which the overwhelming majority are not (especially because people have so much memory, and hence disk cache). It's like saying you'll get better video encoding performance by equipping your PC with a faster CD-ROM drive.
This BS "CPUs are faster than we'll ever need" nonsense is as tired of an argument as it was a decade ago when contrarians were assuring us that a 386 was more power than any reasonable man would ever need. History has shown their claims to be absurd, yet as they say: History repeats itself. Take a man who claims that his Pentium 667 is "faster than I'll ever need" and give him a P4 2.2 to use for a week. Put him back on his 667. 9 times out of 10 he'll be on the phone to Dell to upgrade his PC. Most people who claim that they don't need better say so because they've never SEEN better.
Additionally, try doing some video editing on your PC. While the hard drive is a factor (because massive amounts of data are read and written), the processor is massively more an influence: An Athlon 2200+ will perform the task that much quicker than a Athlon 1500+, again thoroughly reputing your claims that processors are overpowered. That's especially telling as video processing is one of the most disk and memory bound activities.
This highly depends on the application. A single SCSI drive against a single IDE drive performing a single task may show the same performance. However, when you add multiple tasks and a lot of disk access , SCSI beats IDE hands down. As you add drives (don't even bring up RAID yet), tag command queing and parallel data paths blows away IDE no question. Now, add RAID into the equation, especially looking a the huge caching controllers available for SCSI with no IDE counterpart and you see that SCSI is certainly the way to go. Computer manufactures aren't idiots; IDE is cheaper and if it were on equal footing with SCSI no one would offer SCSI solutions. That having been said, no high-performance workstations or servers use IDE.
Sun, HP, etc., have for years sold small MHz machines that outperform the GHz machines available mainly because they use RISC technology and aligned instructions. Clustering has not been a large part of Sun's business -- ever! And, as far as multiple CPU's in a single box, yes, all these systems offer and endorse this, but then so does Intel if you read their journals. Intel ran themselves into needing GHz clocking because of poor chip design (backward compatible to x86). Sun and others don't design chips in those ranges because of the cooling requirements and heat failure rates. It is far easier for them to make lower MHz machines with multiple processors because they run OS's and software that can work UMP or SMP, where Intel has issues in the common market environment (example: Windows 95/98 unable to work SMP).
Click here or here.
I've chimed in on this myself with the following story on Forbes.com:
Will Apple Put Intel Inside?
August 9, 2002
Rumors are buzzing that Apple computers may one day be stamped "Intel Inside." It won't happen.
http://www.forbes.com/2002/08/09/0809apple.html
It looks like a lot of people have already challenged your assertion that IDE is just as good as SCSI. However, no one brought up the particular issue that plagues my experience with IDE - hardware write caching.
Most modern IDE drives have write caching enabled by default. However, under every OS I've tested this configuration can lose data, even with a journaling filesystem. The problem is that the filesystem thinks that the data is successfully written to disk, but it's actually in the drive's cache buffer. If you lose power at the wrong moment, you lose that data. I've reproduced this problem with Western Digital, Seagate, and Maxtor 7200rpm 4MB buffer 80GB IDE drives under both Linux 2.4.X kernels and Microsoft 98/2000/XP platforms.
I've written in to each of those drive manufacturers and they have confirmed that the cache buffer isn't backed by some battery or other type of power reserve, and that data can be lost when power is removed.
Apparently this isn't an issue in SCSI land because SCSI drives respect a flush command, while some IDE drives do not.
The bottom line is that if you want a reliable system with IDE drives you need to disable write caching, which drastically increases disk access latency and results in reduced throughput for many tasks.
I'd love it if a kernel hacker can provide some more details as to why journaling filesystems can't forceably flush the IDE disk's buffer... I've found many older threads on the issue on the linux kernel list but haven't found any definitive resolution or action items recently.
As the situation stands now, my iozone benchmarks show a 15k RPM 80GB SCSI drive performing 2x to 3x better across all tests than a 7.2k RPM 80GB IDE drive with write caching disabled, DMA turned on, and all other hdparm options optimized for maximal performance. That is a pretty large difference. Yes, I did verify that the hdparm tuning options were working correctly.
And yes, the 3ware IDE RAID controllers have the exact same problem. They have an on-board raid cache, but it's not battery backed, so it is not a good idea to enable write caching in most cases. The 3ware cards are great and cheap, but they don't perform as well as their scsi equivalents.
Before someone tries to flame me, yes I have heard of a UPS, but for the machines I'm trying to protect I can't trust that the UPS will be properly maintained, not overloaded, strong enough to survive a long outage, or that the customer won't hit the power button themselves out of ignorance when they think that the system has "hung".
I looked into this once tryong to convince myself to buy an LCD panel instead of a CRT for my next computer.
For an 18" LCD vs a 19" CRT, I'd have to keep the LCD for something like 17 years before it paid for the price difference in power costs, compared to a CRT running 9 hours per day. (Home computer, where I'm at work or asleep most of the day and so the CRT is turned off or in powersave mode when I'm not in front of it.) $200 for the CRT, $600 for the LCD, and 7cents per kilowatt-hour. Don't remember where I got the power usage figures, I think it was from NEC CRT and LCD monitor spec sheets.
I did not include power costs for running the AC extra in the summer. Bear in mind, though, that you run the heater less in winter too, so it is possible that you will balance this out. I didn't look into this, but it may be a wash.
This let me know that power consumption/cost alone was not a reason to get an LCD monitor for a desktop computer.
You don't realize how little power really costs... An extra 50watts, used 24 hours per day, increases a power bill about $30 per year. Takes a long time to make up for $1000 difference in system price at $30 per year.
This is my sig. There are many like it but this one is... Oops. Frank, I've got your sig again! Where's mine?