Not to mention it keeps the EXTREMELY high voltage used called the "screen" from killing you. >>>>>>>> Oh come on. We played with open monitors lots of times in high-school. They're like 20-30 thousand volts. It'll give you a good shock, but it won't kill you. The real danger is that it arcs several inches, so "insulated" wires look safer than they are.
Internet2 >>>>> He he. Hopefully I'll be out of college by then. I actually kinda enjoy the 300K+ per second I get whenever I connect to someone else on Internet2. Wonder what'll happen to the speed when other people get on:(
This is a bit oversimplified (and I'm afraid of networking), so please by nice with corrections:) When you connect to your ISP, you connect via something like DSL or cable or (dread) dial-up. Now your ISP is probably pretty small, so he has a fatter connection to some major telcom company (like AT&T, for example). The telcom companies maintain whats called the internet "backbone" (though its more like a web, I'd hate to meet the creature that has a backbone arranged like that:) These companies have fiber optic connections that link them to other major telcom companies. Its these high-speed links between major telcom companies (and probably government agencies and whatno) thats called the Internet backbone.
Computer programs are critically different from mathematical algorithms: they run in computers, which are essentially mechanical devices. >>>>>>>>>>> And mathematical algorithms can be done in hardware (not via computers, but mechanically). What's your point?
As for the valve example: if its a new timing, then hell no! That's patenting a numerical value just because you were the first to get it. If its a new type of valve, then sure. Key distinction between a discovery and an invention.
This fact presents a huge hurdle that an ivory-tower opponent has to overcome: that somehow writing a program is so different from designing a electronic or mechanical device that software patents have exactly the opposite effect that hardware patents have had. >>>>>>>> I'm not talking about software in general. I'm talking about software algorithms. Software algorithms are statements of truth, the design of a mechanical device are creative inventions. Take, for example, the Cleartype patent. Its is a patent on an algorithm that minimizes perceived color error. Now, by your logic, if that's patentable, so is the algorithms by which scientists derive a crystal structure from an electron density map. Do you know how a patent on that would have f**ked solid state physics? I'm not saying that software patents in general are bad. Say somebody invents a new way for the user to interact with the operating system. That's really not an algorithm in the general sense, and should be patentable. What I'm arguing against is patenting the low level algorithms in software, the parts that are really part of "Computer Science" rather than "The Software Industry."
And then presumably risen Lazarus-like a few years later, when the patent expired? >>>>>>>> Um, no, more like a few decades later. Patent terms are long and getting longer. The JPEG patents we issued, for example in the mid 1980s. And computer science would not have come to an end (that was me exaggerating for effect), but it would have set back the field significantly.
Patents have existed in some for or another for hundreds of years. Many scientists could have taken advantage of them. Generally, however, they have not, since the whole idea behind the way science is conducted is that people build upon the work of others. Patents throw a wrench into the whole process. Patents were made to protect inventions, not discoveries. These are clearly two different things, though they might overlap at many points. However it helps to think of it this way. Patenting allows someone control of a type of product, but not necessarily the principle behind it. Take for example, the lightbulb. The bulb itself could be patented, but the concept of running electricity through a wire to get heat really can't. In software, however, the product is itself the concept, expressed in code. While I suppose the code (the implementation) could be patented, the concept (the algorithm) should not. A good example is the Cleartype patent. The innovation involved in Cleartype starts with a mathematical model for human perception of color error. Then, it uses various mathematical techniques such as Fourier analysis to derive an algorithm that minimizes local color error. Sure its innovative, but its simply a mathematical transformation of scientific principles. Scientists do the exact same thing in other fields every single day, but only in the software world can something like this be patented.
As for the comment about people inventing new algorithms keeping them secret, I doubt it. After all, if that were the case, mathmatics simply wouldn't be a viable field without patents. And who ever heard of mathmatical discoveries being patented!
How is this a human right? Making money is not a human right. Now, let's talk about basic fairness. Say someone gets a patent on something. Five years later, somebody independently comes up with the same idea. Is it really fair for the second person not to be able to use HIS innovation, just because he approached the problem a few years later? Software isn't like most products. Usually, when you come up with a product, just selling it causes the information behind the patent to become known. Thus, the patent in this case really does protect the person's right to utilize his work. Software isn't like that. Thus, it is entirely conceiveble that certain software could use patented algorithms without anybody else knowing. Now the patent system here just prevents competition, instead of protecting the first person's right to his work.
You might want to look at the science angle. Computer algorithms aren't really different from mathematical algorithms. Can you imagine a mathematician patenting his method for finding large primes? Patenting software algorithms is exactly equivilent. Wouldn't it be terrible if Dijkstra had patented the semaphore? Computer science would have come to an end! Computer science is a science like any other. New discoveries should be credited to the people that discovered them, but that shouldn't prevent other people from using and building upon that work. It just stagnates the whole system.
I consider this a good thing, btw. All those type designers that assume I'm running at 75 dpi and bork the layout on my 133dpi computer don't NEED anymore control over how their webpages look on my computer.
You're right about the wattage, I was using older numbers from the 2.2 GHz P4. But, I'd also mention that the 1 GHz G4 draws a maximum of 30 watts, not 18. Either way, as long as its less than 100w, who cares? I mean most monitors use > 150w, so is the processor really an issue?
I will state that it isn't soundly trounced. >>>>>>>>>> 60% is soundly trounced, especially considering the cost of the machine and the fact that when moving to even more data-heavy benchmarks the slow bus speed of the G4 will probably hurt it even more.
2.5 times the clock speed, and only 60% better performance? >>>>> Yes. Performance doesn't scale linearly with clock-speed, even among chips within the same architecture. A 2 GHz Athlon XP is not 100% faster than a 1 Ghz Athlon. Its closer to 50-60% faster. But that's still very significant, even $800 HPs are being sold with 1.5 GHz Athlons these days.
As for the runtimes of the comparison, 1-5 seconds is 1 - 5 billion clock-cycles. Short, yes, but entirely sufficient for CPU benchmarking.
Now, there is a lot of talk about benchmarks not being important. Is 60% faster really sufficient to brand one processor a slowpoke and another one wickedly fast? I contend, yes. Its not necessarily to say that one system is better than another just because of the processor, but the processor itself should cry itself to sleep at nights in shame. 60% (or more given data heavy benchmarks) will let you finish your work quicker, and spend less time waiting for your computer to respond to things. Personally, I noticed a very visible improvement when I moved from my 1.5Ghz Athlon to my 2GHz P4. In reality, the P4 is probably 20% or less faster, but my machine feels less unusably slow./. isn't for average users who putter around with email and Word. Its for power users, and until every conceivable operation can be done in less than 0.3 seconds (human response time) then its still too slow.
I think it might just be easier to make a giant 256x256 cursor in that case. Display resolutions go up slowly enough that "cursor scaling technology" really isn't something to bother about:)
You keep calling the PPC a weaklink, underperformer, etc yet continually fauly to provide a link to any benchmark that shows a P4 syatem beating a PPC system by anything close to the ratio of their clock speed difference. >>>>>>>>>> I pointed you to the ArsTechnica forum. They sum up the benchmarks much better than I ever could. There is a specific thread full of benchmark results. But if you want: http://www.digitalvideoediting.com/2002/05_ may/fea tures/cw_aeshowdown.htm http://www.digitalvideoed iting.com/2002/07_jul/fea tures/cw_macvspc2.htm http://www.heise.de/ct/engl ish/02/05/182/
Go to the ArsTechnica site for mroe info.
Your "treatise" is littered with opinion (apparently mostly quoted from Ars Technica). >>>>>> Like what? The G4 has 4 integer units? That's opinion? To feed the AltiVec units at full power you need 32GB/sec of bandwidth, while the Mac system bus only provides 1.3GB/sec of bandwidth, that's opinion? You have yet to give me a single factual statement that might confirm that the G4 is competitive with Pentium 4s. Point to some aspect of the architecture that indicates that might be the case. G4 proponents seem to believe that there is something magic in the PPC architecture that allows a G4 to process twice as much data in a given clock cycle than a Pentium 4, to make up for the clock speed difference. I'm showing you exactly how the G4 not only has no magic in it, but in many ways, its architecture is not as good as the P4's, clockspeed aside. Architecture aside, clock-speed aside, its not even theoretically possible for the G4 to be competitive with a P4. Most code doesn't have nearly enough parallism to feed an instructions per cycle number much higher than current x86 chips. In those special cases where where code does have that level of parallelism, the G4's memory bandwidth issues prevent the CPU from taking full advantage of its architectural good points.
Noplace has anyone shown (that I know of) that at 2.5 times the clock speed, that a P4 system is even twice as fast as a PowerMac. >>>>>>> Its not twice as fast as a PowerMac. Its more like 50-75% faster. In some cases its much closer to 100% (where memory bandwidth comes into play) and in same cases (much rarer) where the G4's L3 cache or vector permute unit becomes useful, the gap is much smaller. But remember that 50-75% is about the same as x86 chips with similar clock-speed differences. In this era of 2.5 GHz P4s, I consider a 1.5 GHz P4 to be a weakling/underperformer/etc, just as I consider a G4 to be an underperformer. And if you'd just go to the damn Ars forum, you'd see them.
I don't know why you say Itanium is slower than the P4. HP's testing shows [hp.com]that cycle for cycle the Itanium outperforms the P4 in the SPEC tests you like so much. Ex: a 1GHz Itanium is 2.1x as fast in FP than a 1.6GHz P4 Xenon. >>>>>>>> You're link pointed to an article about an Itanium2. I was talking about the Itanium-1. Read up on the architecture of the Itanium2 sometime. The thing has 6 integer units, 2 floating point units, a vector unit, megabytes of cache, 328 registers, can issue 11 instructions per cycle, and has 6.4 GB/second of bus bandwidth. OF COURSE ITS TWICE AS FAST AS A P4! But a G4 doesn't have all those things. What makes it faster? Nothing, because its not! Lets look at this mathematically. Well judge the performance of the G4 by making some relationships (see spec.org for benchmarks). The fastest P4 is more than 50% faster, in integer, than the fastest Alpha (1GHz). In order to be competitive with the Alpha, the G4 has to be faster than an Alpha at the same clock-speed. Both CPUs have 4 integer units. Both are RISC. However, the Alpha has twice the internal cache, 4x the external cache, many times the bus bandwidth, and is recognized as being one of the best CPU architectures ever built. Are you still going to tell me the G4 is faster clock-for-clock?
My opinion remains that Intel builds CPUs, support shipsets and mobos that are designed soley to achieve higher clock speeds with actual real-world performance being far less than should be expected for that transistor count and current draw. >>>>>>>. Nobody's arguing about which CPU is more efficient. But in the world of desktops, 40w vs 14w really doesn't make a single ounce of difference. That's the difference between one-sixth a lightbulb and one-half. What counts is ultimate, wall-clock performance.
AMD designs chips that are overall well balanced, and efficiet, >>>>>> Hah! Shows how much you know about CPUs! The AMD chips run hotter and draw more current than their Intel counterparts. Only for a short time (0.18 micron Willamette) did AMD draw less current. In terms of transistors, they're maybe 20% smaller, but that doesn't mean much.
Apple is now designing systems that are fairly well balanced and remove many previous bottlenecks (many with still exist in P4 systems). >>>>>>>> What kind of bottlenecks? Apple systems have some terribe bottlenecks. The measly 167Mhz bus being one of them. Everyone else and their mother has already moved to double and quad pumped busses, and Apple is stuck with technology slightly better than PC133. P4s currently have 4.2 GB/sec of memory bandwidth. AMD (with Clawhammer) is moving towards 6.4 GB/sec. Apple has 2.7 GB/sec (theoretical) and 1.3 GB/sec (actual). What bottleneck's does Apple fix? If you mean the stupid architecture "documentation" they've got on their site, its bull. Its a standard NorthBridge/SouthBridge architecture, and despite their claims of a "DirectPCI bus" its a standard PCI bus setup as well. In fact, the vanilla NorthBridge/SouthBridge connected via the PCI bus setup is quite dated. All modern x86 chipsets use either the hub architecture, or connect their North/South bridges together via dedicated high-speed busses. The only "bottleneck" breaking Apple does is add the L3 cache. Sure its significant, and allows the G4 to do well on certain benchmarks, but it only runs at 4GB/sec, slower than the main memory on modern P4 systems.
The Xeon chips are exactly like the P4 chips except with more cache. They have only slightly lower (usually a 100Mhz or so behind) clockspeed but a lot more cache. The pipelines and such are exactly the same length. As for the Itanium, the P4 outperforms the Itanium by a good margin, about 15% in SpecFp and 100% in SpecInt. I don't disagree with you that a lower clocked CPU can't outperform a much higher clocked CPU. Happens all the time. A 1Ghz Alpha 21264C is just as fast as a 2Ghz P4 in floating point, and 75% as fast in integer (which is a more important measure, btw for most desktop programs like compiling or whatnot). How is this possible? Well, the Alpha has 4 integer execution units and two fully independent floating point units. In comparison, the Pentium 4 has three integer units and a single floating point unit. All things being equal (they aren't, the P4's floating point unit is stack based, which makes it slower clock-for-clock than the Alpha's) the performance figures make sense! The P4 is faster in integer because the additional integer instruction that the Alpha can do per clock doesn't make up for the clock speed. The P4 is the same speed in floating point because the Alpha has twice as many (better) floating point units. Great example of a lower clocked CPU performing as well as a higher clocked one. BUT, the G4 is no Alpha. It simply does not have as many execution resources as the Alpha, and simply cannot hope to compete at its level, much less the level of a P4. The G4 has four integer units, and a measly one floating point unit. On top of that, the integer units aren't fully independent, three of them can only do simple operations like addition and subtration. The P4, in comparison, has two simple integer units clocked at TWICE core clock (5Ghz in a 2Ghz chip) and a single complex integer unit. Note, that the P4 has quite an advantage here architecturaly. Not only can it (theoretically) complete 25% more simple integer instructions per clock cycle, but because it has only two highly-clocked units rather than four lower-clocked units, it isn't as burdened with inter-instruction dependencies as the three units in the G4e. So even architecturally, the P4 has an advantage here in terms of integer performance. Now, the P4's floating point unit is slightly slower than the G4's clock for clock, but not 2.5x as slow. When all this is factored together, the end result is that the P4 is a great deal faster than the G4 running regular integer code, and a little bit faster running regular floating point code. Now, let's take SSE2 and AltiVec into account. Theoretically, AltiVec blows SSE2 out of the water. The G4e can theoretically issue 2 floating point instructions per cycle, and be crunching on 4 simultainously. The P4, meanwhile, can only do 1 per cycle. BUT, and there is a huge BUT here. Altivec's 4 excecution units are all different. 1 is a permute unit, 1 is simple integer, 1 is complex integer, 1 is floating point. In any given application, it is highly unlikely that more than two (or even that) will be usable at the same time. Thus, being generous, AltiVec can do 2 vector ops per cycle. But there's another catch: 2 vector ops per cycle = 32 bytes of source data per cycle. Running at 1 GHz, assuming full throughput from the vector unit, that equates to 32 gigabytes of data per second! But the processor bus (running at a mere 167 MHz and 64-bits) can only deliver 1.3 GB/sec! Now there lies the problem with AltiVec. The unit itself is great, but the current G4 implementation doesn't have nearly the bus-speed necessary to feed it. Now, with its 533Mhz 64-bit bus, (4.2 GB/sec) the P4 can feed its vector unit much more data, and thus crunch many more numbers, even though its actual vector unit is slower. But how, then, can the G4 ever come close to equaling the P4 (because it does, occasionally). The answer is whenever Apple can make up benchmarks that utilize the unique quirks of the G4. The G4 has a large L3 cache, running at 500MHz (4GB/sec with 64-bit bus). On benchmarks where the dataset fits in the 2MB of L3 cache, and where the data consists of integer values, the L3 cache will be able to feed the two integer pipes in the AltiVec unit quickly enough to get some serious performance. Guess what does this? Certain Photoshop benchmarks! Of course, its use in reality is rather limited. If the dataset is small (less than 512K) then the dataset fits in the L2 caches of both chips and the P4 wins because of the clock-speed advantage. If the dataset is bigger than 2MB, the data spills out of the cache of both chips, and the P4 wins by even more because of its greater memory bandwidth. If the data-set consists of floating point data, then the AltiVec unit can essentially do only one per cycle, and the P4 wins because of the clock-speed advantage.
There you have it. A veritable treatise on why the G4 is a CPU weakling. Again, let me note that I'm not trying to say that people shouldn't buy Macs, or that they aren't nice machines in other ways ("she has a nice personality!") merely pointing out that Apple's marketing department is fully of bull.
PS> If you want to understand CPUs at some level, a good place to start is ArsTechnica (www.arstechnica.com). They are one of the only websites that actually give technical detail in their articles, much more so than you'll find in any Slashdot thread. Much of the info in this post can be found in their article "An architectural comparison of the P4 and the G4e."
I never said that you couldn't do the same thing in Windows. I said that it speaks well of the UNIX APIs that such obscure functionality can be access through entirely generic APIs. The fact that you can just go ahead and use standard mmap functions to map random bits of memory is genuinely cool to those (like me) who value some elegence and beauty in software architecture. And thanks for the great example, by the way. DirectX is exactly what I'm talking about. Specific APIs to access generic functionality. Why should I have to use a different API to map graphics memory vs mapping regular memory vs mapping a file? Its just not clean! But it characterizes the Windows way of doing things.
PS> As for the comments about my intelligence and my spelling:
1) I've known DirectX since I was in the tenth grade. Learned Win32 just after I learned the C++ standard library.
2) I just got off summer break, it was late at night, and I was typing that on a laptop keyboard, so fuck off!
I'd like to point out how well this speaks of the Linux kernel. By making the architecture generic and orthagonal, it allows you to do cool (if useless) stuff like this. Contrast this to Windows, where every API is extremely specifc, and you'll realize why Linux/UNIX is infinately better.
The AGP interface, at 1 GB/sec, is less than 1/3'rd the speed of the memory interface (3.2 GB/sec). Now, that speed is only attainable through deep buffering (basically, block transfers). Writing and reading to video memory from the CPU is, and always will be, extremely, hideously, painfully slow. Slow to the point where the XRender guys (read their mailing list) are talking about pulling blocks of data of the vid card, compositing them with the CPU, and putting them back on the vidcard.
How can you state that Mhz != performance, then in the very next sentence argue that Mhz = performance? >>>>>> No, is said that MHz is a factor in performance. Performance = Instructions_Per_Cycle * Clock_Speed. If the Clock_Speed of a processor is 2.5x as much as that of another processor then the Instructions_Per_Cycle must be at least 2.5x as high to balance the difference. Even theoretically, the PowerPC chips cannot hit that kind of IPC.
The PPC can keep up when it requires fewer instructions to process any task, >>>>>>> Wrong! The PowerPC is a RISC chip. RISC chips, in general, require more instructions to perform a given task.
and when vector processing allows massive scaling of clock cycle to work performed. >>>>>>> True, but the Pentium4 has a vector unit as well. Not as good as the AltiVec unit, but in the current G4 the AltiVec unit is severely limited by the bandwidth of the PowerMac system bus. Thus, SSE2 is in the same performance league even at the same clockspeed, and on a P4, the floating point units (that processes the SSE2 instructions) have the advantage of running at 2.5x the frequency. Again, Performance = IPC * CLOCK, and with the clock being 2.5x as fast, IPC must be 2.5x as high, and AltiVec is nowhere near that much faster than SSE2.
Also note that there's nothing inherent in the PPC design that is limiting clock speed, it's Motorola's corporate issues that keep the speeds down. >>>>>>>>>> Actually, there are two things keeping the clock speed down, both technological. First, the G4's pipelines are short, which makes each pipeline stage more complex, which makes it harder to clock them faster. Second, Motorola's process technology isn't as good as Intel's.
The benchmarks in SPEC are artificial. They test particular oddball functions of a system. How often does an average user cross-compile pre-processesed C code files into Motorola 88100 processor machine language? >>>>>>>>>>>> For most of a compiler run, the source code is represented as a processor-independent parse tree. This what the optimizers work on, and this is what that SPEC test tries to measure. Only final instruction scheduling actually uses the specific architecture to be compiled, and the 88100 was used in this case because GCC's algorithms are optimized to various levels for generating code for the most popular architectures. Testing with something like x86 or PPC code would be unfair because the x86 instruction scheduler most likely has optimizations the others don't and vice versa. The 88100 is obscure enough that this isn't a problem.
When was the last time you needed to process the "lithography artwork needed for the production of microchips"? >>>>>> Read the whole thing. Its an optimization problem, and people in scientific computing do these sorts of things all the time. Hell, people running Quake run these sorts of optimization problems (in the AI routines) all the time.
Never would be a reasonable guess. Of course, the point is moot as I can't locate a single PPC chip test result, never mind a PowerMac system. >>>>>>>>> http://www.heise.de/ct/english /02/05/182/
A real-world benchmark in my opinion (for general use desktop systems) is using a program/code compiled and optimized for a given CPU to test some complete task that is commonly performed by the average user. >>>>>>> Apple does real world benchmarks, right? Like Photoshop (only 6.0, though, because in 7.0 Adobe added SSE2 support in addition to AltiVec and Apple loses now)? The way Apple pushes Photoshop benchmarks you'd think all those people with iMacs ran photoshop all day long! Okay, but I'll bite, here are compilation benchmarks: http://homepage.mac.com/nopea1/benchmark/ Guess what? They show a G4 running about 20% faster than a PIII at the same clock speed. That equates to about 30-35% faster than a P4 at the same clockspeed. Where is the 150% difference that's needed here to make up for the difference in clock?
Ex: Rip an audio CD to.mp3 format, make some alterations and burn out to CDR. >>>>>>>>>> Umm, you're mostly testing the speed of your CD drive here.
Load an image and perform some editing and convert it to a web suitable size/format. >>>>>>> Kinda like PSBench? Guess what, the G4 loses here too! (See ArsTechnica forum, search for PSBench)
Frame rates of games are pointless benchmarks. >>>>>>>> Uh, how? A 3D game is the perfect test for new media applications. The geometry transformation algorithms (a lot of 3D matrix multiplies) can take real advantage of stuff like AltiVec and SSE. Plus, the AI algorithms stress non-linear code performance (something the P4 is bad at) and the heavy use of textures really stresses memory bandwidth.
What's the point of having a graphics subsystem that can do 200FPS in quake when your monitor can only refresh at 120Hz max, and your eye can't see anything above 80hz anyway. >>>>>>>> Because when Doom3 comes out, a machine that can get 200fps in Quake III will be able to run it with full detail at playable speeds, while a machine that only gets 100fps in Quake III won't.
It's like comparing the top speed of a cars when they'll never go faster than 80mph anyway. The speed becomes insignificant compared to other aspects of the system or car. >>>>>>>> Other aspects, like how transparent it is? Seriously, though, this thread was never about transparency (or easy of use, or fludity of the GUI or workflow or whatnot) or other aspects of the system. It was about performance, an area in which Apple is sorely lacking.
Don't be ridiculous. Of course Mhz != performance. But the PPC loses in clockspeed by something like 2.5x. At that disparity, even if the PPC could keep its pipelines full 100% of the time due to its great architecture, it STILL wouldn't be as fast as the fastest P4. This is getting old. Go to ArsTechnica's forum, and look up G4 vs P4. Read the posts. A lot of people a whole lot smarter than you or me have put up some good info there. As for SPEC, go here. Yep, GCC and PERL sure sound like artificial benchmarks to me!
Ha ha, that's a good one. And I bet you believe that A G4 smokes an Alpha just because it gets higher RC5 scores? Don't make me laugh. The G4 get's killed in the SPEC benchmarks (which are real world, btw, gcc and mesa are part of them, among others) and most importantly, gets hammer in Quake III. I'm not going to dignify this post with a further response, except to quote my favorite ArsTechnica post: >>>>>>>>>>>>>>>&g t;
"MHz Myth"
Def: When faced with a computer that is significantly faster than a Mac.. simply scream out "MHz Myth" and wave your hands frantically. This will magically make the languishing G4 faster than the current fastest Intel or AMD Chip.
See "Up to Twice as Fast as any Pentium PC" for more details..
Whoops. I meant the 540s (2 channel). They're about $80, less on pricewatch. Being a Klipsch bigot, I'd never come near speakers made my a mouse manufacturer, but I hear they're quite good if you like the whole 2.1 thing...
Yes, the cheap Dell is better in some ways then the cheap Apple. But then again, you forgot to add the video editing bundle and the CD burner, which bumps the price up somewhat. Besides, I'd pay $200 extra to use OS X. >>>>>>>> Sorry, the link was bad. Dell order-form doesn't allow deep linking. I priced in a CDR at the $1120 price point. And you might like to use OS X, but Windows XP is pretty damn good for the average home user, and they would probably take the $200.
As for the high end - dual 1.25Ghz G4s aren't "vastly slower" than dual 2.4Ghz P4s. For your average office app or game, you won't notice a difference. >>>>>>> Oh please, that's bull. The Office app will be slower because OS X's GUI is slower. The game will be slower because G4s are nowhere near as fast as P4s. PC World did some benchmarks awhile ago that showed that Quake III was 50% faster on a P4 1.5 Ghz than on a G4 733 MHz, using EXACTLY THE SAME GRAPHICS CARD. That meant the the CPU on the P4 was a good deal more than 50% faster. Assuming for a moment that this scales with the different CPU types here (which it doesn't, because the Xeon has a much larger full speed cache as opposed to the G4's external, fractional speed cache) that still puts the dual P4s at more than 50% faster. And this doesn't take into account that the G4 doesn't effectively utilize DDR-SDRAM, and as games become more memory-bandwidth limited (Doom III) the effectivly PC-133 performance on the G4s will be blown away by the 3.2 GB/sec of bandwidth on the Xeons. And if you go into apps besides games and office (like gcc and whatnot) then you're covered by the SPEC benchmarks, and we all know how poorly the G4 does in those.
For multimedia purposes, the G4s are probably at least as fast, due to Altivec. Yes, I know, Altivec isn't an omnipotent silver bullet, but it still kicks ass. >>>>> Ha. There is a lot of talk on this (one giant 1200 post thread on ArsTechnica) and the general conclusion is that given the limitations of Apple's platform (namely memory bandwidth) AltiVec really isn't that much faster than SSE2 except in a few special cases where its permute instructions are useful. Otherwise, the P4s smoke the G4.
And if I have the choice between a Dell 20" flat panel and an Apple 23" flat panel, I think I know what I'd take. >>>>>>.. I wouldn't be too sure. I know Sharp makes some LCD screens that look noticibly better than Apple's, and are cheaper to boot. Since Dell rebrands them, I wouldn't be surprised to see one of these panels show up somewhere in its product line.
Well, what happens is that since the Inspiron 8000, Dell laptops have had the same graphics add-in card). So you could upgrade from a Geforce2 to a GeForce4 MX. We don't know yet if that will hold for the next Inspiron when the NV31-based mobile GPUs come out.
As for the LCD, its magnificent. 1600x1200 on a 15" screen gives 133 dpi, and with ClearType, its like reading a piece of paper.
As for the degrees of freedom, I meant that you can swing about thirty degrees to either side without losing the image on the LCD.
Slashdot Mirror
Not to mention it keeps the EXTREMELY high voltage used called the "screen" from killing you.
>>>>>>>>
Oh come on. We played with open monitors lots of times in high-school. They're like 20-30 thousand volts. It'll give you a good shock, but it won't kill you. The real danger is that it arcs several inches, so "insulated" wires look safer than they are.
Internet2 :(
>>>>>
He he. Hopefully I'll be out of college by then. I actually kinda enjoy the 300K+ per second I get whenever I connect to someone else on Internet2. Wonder what'll happen to the speed when other people get on
The impuse travels at 120-something miles per second vs 186 thousand miles per second for light.
This is a bit oversimplified (and I'm afraid of networking), so please by nice with corrections :) When you connect to your ISP, you connect via something like DSL or cable or (dread) dial-up. Now your ISP is probably pretty small, so he has a fatter connection to some major telcom company (like AT&T, for example). The telcom companies maintain whats called the internet "backbone" (though its more like a web, I'd hate to meet the creature that has a backbone arranged like that :) These companies have fiber optic connections that link them to other major telcom companies. Its these high-speed links between major telcom companies (and probably government agencies and whatno) thats called the Internet backbone.
Computer programs are critically different from mathematical algorithms: they run in computers, which are essentially mechanical devices.
>>>>>>>>>>>
And mathematical algorithms can be done in hardware (not via computers, but mechanically). What's your point?
As for the valve example: if its a new timing, then hell no! That's patenting a numerical value just because you were the first to get it. If its a new type of valve, then sure. Key distinction between a discovery and an invention.
This fact presents a huge hurdle that an ivory-tower opponent has to overcome: that somehow writing a program is so different from designing a electronic or mechanical device that software patents have exactly the opposite effect that hardware patents have had.
>>>>>>>>
I'm not talking about software in general. I'm talking about software algorithms. Software algorithms are statements of truth, the design of a mechanical device are creative inventions. Take, for example, the Cleartype patent. Its is a patent on an algorithm that minimizes perceived color error. Now, by your logic, if that's patentable, so is the algorithms by which scientists derive a crystal structure from an electron density map. Do you know how a patent on that would have f**ked solid state physics? I'm not saying that software patents in general are bad. Say somebody invents a new way for the user to interact with the operating system. That's really not an algorithm in the general sense, and should be patentable. What I'm arguing against is patenting the low level algorithms in software, the parts that are really part of "Computer Science" rather than "The Software Industry."
And then presumably risen Lazarus-like a few years later, when the patent expired?
>>>>>>>>
Um, no, more like a few decades later. Patent terms are long and getting longer. The JPEG patents we issued, for example in the mid 1980s. And computer science would not have come to an end (that was me exaggerating for effect), but it would have set back the field significantly.
Patents have existed in some for or another for hundreds of years. Many scientists could have taken advantage of them. Generally, however, they have not, since the whole idea behind the way science is conducted is that people build upon the work of others. Patents throw a wrench into the whole process. Patents were made to protect inventions, not discoveries. These are clearly two different things, though they might overlap at many points. However it helps to think of it this way. Patenting allows someone control of a type of product, but not necessarily the principle behind it. Take for example, the lightbulb. The bulb itself could be patented, but the concept of running electricity through a wire to get heat really can't. In software, however, the product is itself the concept, expressed in code. While I suppose the code (the implementation) could be patented, the concept (the algorithm) should not. A good example is the Cleartype patent. The innovation involved in Cleartype starts with a mathematical model for human perception of color error. Then, it uses various mathematical techniques such as Fourier analysis to derive an algorithm that minimizes local color error. Sure its innovative, but its simply a mathematical transformation of scientific principles. Scientists do the exact same thing in other fields every single day, but only in the software world can something like this be patented.
As for the comment about people inventing new algorithms keeping them secret, I doubt it. After all, if that were the case, mathmatics simply wouldn't be a viable field without patents. And who ever heard of mathmatical discoveries being patented!
How is this a human right? Making money is not a human right. Now, let's talk about basic fairness. Say someone gets a patent on something. Five years later, somebody independently comes up with the same idea. Is it really fair for the second person not to be able to use HIS innovation, just because he approached the problem a few years later? Software isn't like most products. Usually, when you come up with a product, just selling it causes the information behind the patent to become known. Thus, the patent in this case really does protect the person's right to utilize his work. Software isn't like that. Thus, it is entirely conceiveble that certain software could use patented algorithms without anybody else knowing. Now the patent system here just prevents competition, instead of protecting the first person's right to his work.
You might want to look at the science angle. Computer algorithms aren't really different from mathematical algorithms. Can you imagine a mathematician patenting his method for finding large primes? Patenting software algorithms is exactly equivilent. Wouldn't it be terrible if Dijkstra had patented the semaphore? Computer science would have come to an end! Computer science is a science like any other. New discoveries should be credited to the people that discovered them, but that shouldn't prevent other people from using and building upon that work. It just stagnates the whole system.
I consider this a good thing, btw. All those type designers that assume I'm running at 75 dpi and bork the layout on my 133dpi computer don't NEED anymore control over how their webpages look on my computer.
You're right about the wattage, I was using older numbers from the 2.2 GHz P4. But, I'd also mention that the 1 GHz G4 draws a maximum of 30 watts, not 18. Either way, as long as its less than 100w, who cares? I mean most monitors use > 150w, so is the processor really an issue?
/. isn't for average users who putter around with email and Word. Its for power users, and until every conceivable operation can be done in less than 0.3 seconds (human response time) then its still too slow.
I will state that it isn't soundly trounced.
>>>>>>>>>>
60% is soundly trounced, especially considering the cost of the machine and the fact that when moving to even more data-heavy benchmarks the slow bus speed of the G4 will probably hurt it even more.
2.5 times the clock speed, and only 60% better performance?
>>>>>
Yes. Performance doesn't scale linearly with clock-speed, even among chips within the same architecture. A 2 GHz Athlon XP is not 100% faster than a 1 Ghz Athlon. Its closer to 50-60% faster. But that's still very significant, even $800 HPs are being sold with 1.5 GHz Athlons these days.
As for the runtimes of the comparison, 1-5 seconds is 1 - 5 billion clock-cycles. Short, yes, but entirely sufficient for CPU benchmarking.
Now, there is a lot of talk about benchmarks not being important. Is 60% faster really sufficient to brand one processor a slowpoke and another one wickedly fast? I contend, yes. Its not necessarily to say that one system is better than another just because of the processor, but the processor itself should cry itself to sleep at nights in shame. 60% (or more given data heavy benchmarks) will let you finish your work quicker, and spend less time waiting for your computer to respond to things. Personally, I noticed a very visible improvement when I moved from my 1.5Ghz Athlon to my 2GHz P4. In reality, the P4 is probably 20% or less faster, but my machine feels less unusably slow.
I think it might just be easier to make a giant 256x256 cursor in that case. Display resolutions go up slowly enough that "cursor scaling technology" really isn't something to bother about :)
But cursors in current machines are hardware drawn (quite an good performance improvement actually) and current hardware doesn't do vector cursors.
You keep calling the PPC a weaklink, underperformer, etc yet continually fauly to provide a link to any benchmark that shows a P4 syatem beating a PPC system by anything close to the ratio of their clock speed difference._ may/fea tures/cw_aeshowdown.htmd iting.com/2002/07_jul/fea tures/cw_macvspc2.html ish/02/05/182/
>>>>>>>>>>
I pointed you to the ArsTechnica forum. They sum up the benchmarks much better than I ever could. There is a specific thread full of benchmark results. But if you want:
http://www.digitalvideoediting.com/2002/05
http://www.digitalvideoe
http://www.heise.de/ct/eng
Go to the ArsTechnica site for mroe info.
Your "treatise" is littered with opinion
(apparently mostly quoted from Ars Technica).
>>>>>>
Like what? The G4 has 4 integer units? That's opinion? To feed the AltiVec units at full power you need 32GB/sec of bandwidth, while the Mac system bus only provides 1.3GB/sec of bandwidth, that's opinion? You have yet to give me a single factual statement that might confirm that the G4 is competitive with Pentium 4s. Point to some aspect of the architecture that indicates that might be the case. G4 proponents seem to believe that there is something magic in the PPC architecture that allows a G4 to process twice as much data in a given clock cycle than a Pentium 4, to make up for the clock speed difference. I'm showing you exactly how the G4 not only has no magic in it, but in many ways, its architecture is not as good as the P4's, clockspeed aside. Architecture aside, clock-speed aside, its not even theoretically possible for the G4 to be competitive with a P4. Most code doesn't have nearly enough parallism to feed an instructions per cycle number much higher than current x86 chips. In those special cases where where code does have that level of parallelism, the G4's memory bandwidth issues prevent the CPU from taking full advantage of its architectural good points.
Noplace has anyone shown (that I know of) that at 2.5 times the clock speed, that a P4 system is even twice as fast as a PowerMac.
>>>>>>>
Its not twice as fast as a PowerMac. Its more like 50-75% faster. In some cases its much closer to 100% (where memory bandwidth comes into play) and in same cases (much rarer) where the G4's L3 cache or vector permute unit becomes useful, the gap is much smaller. But remember that 50-75% is about the same as x86 chips with similar clock-speed differences. In this era of 2.5 GHz P4s, I consider a 1.5 GHz P4 to be a weakling/underperformer/etc, just as I consider a G4 to be an underperformer. And if you'd just go to the damn Ars forum, you'd see them.
I don't know why you say Itanium is slower than the P4. HP's testing shows [hp.com]that cycle for cycle the Itanium outperforms the P4 in the SPEC tests you like so much. Ex: a 1GHz Itanium is 2.1x as fast in FP than a 1.6GHz P4 Xenon.
>>>>>>>>
You're link pointed to an article about an Itanium2. I was talking about the Itanium-1. Read up on the architecture of the Itanium2 sometime. The thing has 6 integer units, 2 floating point units, a vector unit, megabytes of cache, 328 registers, can issue 11 instructions per cycle, and has 6.4 GB/second of bus bandwidth. OF COURSE ITS TWICE AS FAST AS A P4! But a G4 doesn't have all those things. What makes it faster? Nothing, because its not! Lets look at this mathematically. Well judge the performance of the G4 by making some relationships (see spec.org for benchmarks). The fastest P4 is more than 50% faster, in integer, than the fastest Alpha (1GHz). In order to be competitive with the Alpha, the G4 has to be faster than an Alpha at the same clock-speed. Both CPUs have 4 integer units. Both are RISC. However, the Alpha has twice the internal cache, 4x the external cache, many times the bus bandwidth, and is recognized as being one of the best CPU architectures ever built. Are you still going to tell me the G4 is faster clock-for-clock?
My opinion remains that Intel builds CPUs, support shipsets and mobos that are designed soley to achieve higher clock speeds with actual real-world performance being far less than should be expected for that transistor count and current draw.
>>>>>>>.
Nobody's arguing about which CPU is more efficient. But in the world of desktops, 40w vs 14w really doesn't make a single ounce of difference. That's the difference between one-sixth a lightbulb and one-half. What counts is ultimate, wall-clock performance.
AMD designs chips that are overall well balanced, and efficiet,
>>>>>>
Hah! Shows how much you know about CPUs! The AMD chips run hotter and draw more current than their Intel counterparts. Only for a short time (0.18 micron Willamette) did AMD draw less current. In terms of transistors, they're maybe 20% smaller, but that doesn't mean much.
Apple is now designing systems that are fairly well balanced and remove many previous bottlenecks (many with still exist in P4 systems).
>>>>>>>>
What kind of bottlenecks? Apple systems have some terribe bottlenecks. The measly 167Mhz bus being one of them. Everyone else and their mother has already moved to double and quad pumped busses, and Apple is stuck with technology slightly better than PC133. P4s currently have 4.2 GB/sec of memory bandwidth. AMD (with Clawhammer) is moving towards 6.4 GB/sec. Apple has 2.7 GB/sec (theoretical) and 1.3 GB/sec (actual). What bottleneck's does Apple fix? If you mean the stupid architecture "documentation" they've got on their site, its bull. Its a standard NorthBridge/SouthBridge architecture, and despite their claims of a "DirectPCI bus" its a standard PCI bus setup as well. In fact, the vanilla NorthBridge/SouthBridge connected via the PCI bus setup is quite dated. All modern x86 chipsets use either the hub architecture, or connect their North/South bridges together via dedicated high-speed busses. The only "bottleneck" breaking Apple does is add the L3 cache. Sure its significant, and allows the G4 to do well on certain benchmarks, but it only runs at 4GB/sec, slower than the main memory on modern P4 systems.
The Xeon chips are exactly like the P4 chips except with more cache. They have only slightly lower (usually a 100Mhz or so behind) clockspeed but a lot more cache. The pipelines and such are exactly the same length.
As for the Itanium, the P4 outperforms the Itanium by a good margin, about 15% in SpecFp and 100% in SpecInt.
I don't disagree with you that a lower clocked CPU can't outperform a much higher clocked CPU. Happens all the time. A 1Ghz Alpha 21264C is just as fast as a 2Ghz P4 in floating point, and 75% as fast in integer (which is a more important measure, btw for most desktop programs like compiling or whatnot). How is this possible? Well, the Alpha has 4 integer execution units and two fully independent floating point units. In comparison, the Pentium 4 has three integer units and a single floating point unit. All things being equal (they aren't, the P4's floating point unit is stack based, which makes it slower clock-for-clock than the Alpha's) the performance figures make sense! The P4 is faster in integer because the additional integer instruction that the Alpha can do per clock doesn't make up for the clock speed. The P4 is the same speed in floating point because the Alpha has twice as many (better) floating point units. Great example of a lower clocked CPU performing as well as a higher clocked one. BUT, the G4 is no Alpha. It simply does not have as many execution resources as the Alpha, and simply cannot hope to compete at its level, much less the level of a P4. The G4 has four integer units, and a measly one floating point unit. On top of that, the integer units aren't fully independent, three of them can only do simple operations like addition and subtration. The P4, in comparison, has two simple integer units clocked at TWICE core clock (5Ghz in a 2Ghz chip) and a single complex integer unit. Note, that the P4 has quite an advantage here architecturaly. Not only can it (theoretically) complete 25% more simple integer instructions per clock cycle, but because it has only two highly-clocked units rather than four lower-clocked units, it isn't as burdened with inter-instruction dependencies as the three units in the G4e. So even architecturally, the P4 has an advantage here in terms of integer performance. Now, the P4's floating point unit is slightly slower than the G4's clock for clock, but not 2.5x as slow. When all this is factored together, the end result is that the P4 is a great deal faster than the G4 running regular integer code, and a little bit faster running regular floating point code. Now, let's take SSE2 and AltiVec into account. Theoretically, AltiVec blows SSE2 out of the water. The G4e can theoretically issue 2 floating point instructions per cycle, and be crunching on 4 simultainously. The P4, meanwhile, can only do 1 per cycle. BUT, and there is a huge BUT here. Altivec's 4 excecution units are all different. 1 is a permute unit, 1 is simple integer, 1 is complex integer, 1 is floating point. In any given application, it is highly unlikely that more than two (or even that) will be usable at the same time. Thus, being generous, AltiVec can do 2 vector ops per cycle. But there's another catch: 2 vector ops per cycle = 32 bytes of source data per cycle. Running at 1 GHz, assuming full throughput from the vector unit, that equates to 32 gigabytes of data per second! But the processor bus (running at a mere 167 MHz and 64-bits) can only deliver 1.3 GB/sec! Now there lies the problem with AltiVec. The unit itself is great, but the current G4 implementation doesn't have nearly the bus-speed necessary to feed it. Now, with its 533Mhz 64-bit bus, (4.2 GB/sec) the P4 can feed its vector unit much more data, and thus crunch many more numbers, even though its actual vector unit is slower. But how, then, can the G4 ever come close to equaling the P4 (because it does, occasionally). The answer is whenever Apple can make up benchmarks that utilize the unique quirks of the G4. The G4 has a large L3 cache, running at 500MHz (4GB/sec with 64-bit bus). On benchmarks where the dataset fits in the 2MB of L3 cache, and where the data consists of integer values, the L3 cache will be able to feed the two integer pipes in the AltiVec unit quickly enough to get some serious performance. Guess what does this? Certain Photoshop benchmarks! Of course, its use in reality is rather limited. If the dataset is small (less than 512K) then the dataset fits in the L2 caches of both chips and the P4 wins because of the clock-speed advantage. If the dataset is bigger than 2MB, the data spills out of the cache of both chips, and the P4 wins by even more because of its greater memory bandwidth. If the data-set consists of floating point data, then the AltiVec unit can essentially do only one per cycle, and the P4 wins because of the clock-speed advantage.
There you have it. A veritable treatise on why the G4 is a CPU weakling. Again, let me note that I'm not trying to say that people shouldn't buy Macs, or that they aren't nice machines in other ways ("she has a nice personality!") merely pointing out that Apple's marketing department is fully of bull.
PS> If you want to understand CPUs at some level, a good place to start is ArsTechnica (www.arstechnica.com). They are one of the only websites that actually give technical detail in their articles, much more so than you'll find in any Slashdot thread. Much of the info in this post can be found in their article "An architectural comparison of the P4 and the G4e."
I never said that you couldn't do the same thing in Windows. I said that it speaks well of the UNIX APIs that such obscure functionality can be access through entirely generic APIs. The fact that you can just go ahead and use standard mmap functions to map random bits of memory is genuinely cool to those (like me) who value some elegence and beauty in software architecture. And thanks for the great example, by the way. DirectX is exactly what I'm talking about. Specific APIs to access generic functionality. Why should I have to use a different API to map graphics memory vs mapping regular memory vs mapping a file? Its just not clean! But it characterizes the Windows way of doing things.
PS> As for the comments about my intelligence and my spelling:
1) I've known DirectX since I was in the tenth grade. Learned Win32 just after I learned the C++ standard library.
2) I just got off summer break, it was late at night, and I was typing that on a laptop keyboard, so fuck off!
I'd like to point out how well this speaks of the Linux kernel. By making the architecture generic and orthagonal, it allows you to do cool (if useless) stuff like this. Contrast this to Windows, where every API is extremely specifc, and you'll realize why Linux/UNIX is infinately better.
Actually, the extra memory space can be used by X to cache pixmaps, allowing them to be blitted to the frame buffer quicker.
The AGP interface, at 1 GB/sec, is less than 1/3'rd the speed of the memory interface (3.2 GB/sec). Now, that speed is only attainable through deep buffering (basically, block transfers). Writing and reading to video memory from the CPU is, and always will be, extremely, hideously, painfully slow. Slow to the point where the XRender guys (read their mailing list) are talking about pulling blocks of data of the vid card, compositing them with the CPU, and putting them back on the vidcard.
How can you state that Mhz != performance, then in the very next sentence argue that Mhz = performance?
h /02/05/182/
.mp3 format, make some alterations and burn out to CDR.
>>>>>>
No, is said that MHz is a factor in performance. Performance = Instructions_Per_Cycle * Clock_Speed. If the Clock_Speed of a processor is 2.5x as much as that of another processor then the Instructions_Per_Cycle must be at least 2.5x as high to balance the difference. Even theoretically, the PowerPC chips cannot hit that kind of IPC.
The PPC can keep up when it requires fewer instructions to process any task,
>>>>>>>
Wrong! The PowerPC is a RISC chip. RISC chips, in general, require more instructions to perform a given task.
and when vector processing allows massive scaling of clock cycle to work performed.
>>>>>>>
True, but the Pentium4 has a vector unit as well. Not as good as the AltiVec unit, but in the current G4 the AltiVec unit is severely limited by the bandwidth of the PowerMac system bus. Thus, SSE2 is in the same performance league even at the same clockspeed, and on a P4, the floating point units (that processes the SSE2 instructions) have the advantage of running at 2.5x the frequency. Again, Performance = IPC * CLOCK, and with the clock being 2.5x as fast, IPC must be 2.5x as high, and AltiVec is nowhere near that much faster than SSE2.
Also note that there's nothing inherent in the PPC design that is limiting clock speed, it's Motorola's corporate issues that keep the speeds down.
>>>>>>>>>>
Actually, there are two things keeping the clock speed down, both technological. First, the G4's pipelines are short, which makes each pipeline stage more complex, which makes it harder to clock them faster. Second, Motorola's process technology isn't as good as Intel's.
The benchmarks in SPEC are artificial. They test particular oddball functions of a system. How often does an average user cross-compile pre-processesed C code files into Motorola 88100 processor machine language?
>>>>>>>>>>>>
For most of a compiler run, the source code is represented as a processor-independent parse tree. This what the optimizers work on, and this is what that SPEC test tries to measure. Only final instruction scheduling actually uses the specific architecture to be compiled, and the 88100 was used in this case because GCC's algorithms are optimized to various levels for generating code for the most popular architectures. Testing with something like x86 or PPC code would be unfair because the x86 instruction scheduler most likely has optimizations the others don't and vice versa. The 88100 is obscure enough that this isn't a problem.
When was the last time you needed to process the "lithography artwork needed for the production of microchips"?
>>>>>>
Read the whole thing. Its an optimization problem, and people in scientific computing do these sorts of things all the time. Hell, people running Quake run these sorts of optimization problems (in the AI routines) all the time.
Never would be a reasonable guess. Of course, the point is moot as I can't locate a single PPC chip test result, never mind a PowerMac system.
>>>>>>>>>
http://www.heise.de/ct/englis
A real-world benchmark in my opinion (for general use desktop systems) is using a program/code compiled and optimized for a given CPU to test some complete task that is commonly performed by the average user.
>>>>>>>
Apple does real world benchmarks, right? Like Photoshop (only 6.0, though, because in 7.0 Adobe added SSE2 support in addition to AltiVec and Apple loses now)? The way Apple pushes Photoshop benchmarks you'd think all those people with iMacs ran photoshop all day long! Okay, but I'll bite, here are compilation benchmarks: http://homepage.mac.com/nopea1/benchmark/ Guess what? They show a G4 running about 20% faster than a PIII at the same clock speed. That equates to about 30-35% faster than a P4 at the same clockspeed. Where is the 150% difference that's needed here to make up for the difference in clock?
Ex: Rip an audio CD to
>>>>>>>>>>
Umm, you're mostly testing the speed of your CD drive here.
Load an image and perform some editing and convert it to a web suitable size/format.
>>>>>>>
Kinda like PSBench? Guess what, the G4 loses here too! (See ArsTechnica forum, search for PSBench)
Frame rates of games are pointless benchmarks.
>>>>>>>>
Uh, how? A 3D game is the perfect test for new media applications. The geometry transformation algorithms (a lot of 3D matrix multiplies) can take real advantage of stuff like AltiVec and SSE. Plus, the AI algorithms stress non-linear code performance (something the P4 is bad at) and the heavy use of textures really stresses memory bandwidth.
What's the point of having a graphics subsystem that can do 200FPS in quake when your monitor can only refresh at 120Hz max, and your eye can't see anything above 80hz anyway.
>>>>>>>>
Because when Doom3 comes out, a machine that can get 200fps in Quake III will be able to run it with full detail at playable speeds, while a machine that only gets 100fps in Quake III won't.
It's like comparing the top speed of a cars when they'll never go faster than 80mph anyway. The speed becomes insignificant compared to other aspects of the system or car.
>>>>>>>>
Other aspects, like how transparent it is? Seriously, though, this thread was never about transparency (or easy of use, or fludity of the GUI or workflow or whatnot) or other aspects of the system. It was about performance, an area in which Apple is sorely lacking.
And don't forget, the first 20 collers get a free Elbrus E2K CPU!
Don't be ridiculous. Of course Mhz != performance. But the PPC loses in clockspeed by something like 2.5x. At that disparity, even if the PPC could keep its pipelines full 100% of the time due to its great architecture, it STILL wouldn't be as fast as the fastest P4. This is getting old. Go to ArsTechnica's forum, and look up G4 vs P4. Read the posts. A lot of people a whole lot smarter than you or me have put up some good info there. As for SPEC, go here. Yep, GCC and PERL sure sound like artificial benchmarks to me!
Ha ha, that's a good one. And I bet you believe that A G4 smokes an Alpha just because it gets higher RC5 scores? Don't make me laugh. The G4 get's killed in the SPEC benchmarks (which are real world, btw, gcc and mesa are part of them, among others) and most importantly, gets hammer in Quake III. I'm not going to dignify this post with a further response, except to quote my favorite ArsTechnica post:
>>>>>>>>>>>>>>>&g t;
"MHz Myth"
Def: When faced with a computer that is significantly faster than a Mac.. simply scream out "MHz Myth" and wave your hands frantically. This will magically make the languishing G4 faster than the current fastest Intel or AMD Chip.
See "Up to Twice as Fast as any Pentium PC" for more details..
Whoops. I meant the 540s (2 channel). They're about $80, less on pricewatch. Being a Klipsch bigot, I'd never come near speakers made my a mouse manufacturer, but I hear they're quite good if you like the whole 2.1 thing...
Yes, the cheap Dell is better in some ways then the cheap Apple. But then again, you forgot to add the video editing bundle and the CD burner, which bumps the price up somewhat. Besides, I'd pay $200 extra to use OS X.
>>>>>>>>
Sorry, the link was bad. Dell order-form doesn't allow deep linking. I priced in a CDR at the $1120 price point. And you might like to use OS X, but Windows XP is pretty damn good for the average home user, and they would probably take the $200.
As for the high end - dual 1.25Ghz G4s aren't "vastly slower" than dual 2.4Ghz P4s. For your average office app or game, you won't notice a difference.
>>>>>>>
Oh please, that's bull. The Office app will be slower because OS X's GUI is slower. The game will be slower because G4s are nowhere near as fast as P4s. PC World did some benchmarks awhile ago that showed that Quake III was 50% faster on a P4 1.5 Ghz than on a G4 733 MHz, using EXACTLY THE SAME GRAPHICS CARD. That meant the the CPU on the P4 was a good deal more than 50% faster. Assuming for a moment that this scales with the different CPU types here (which it doesn't, because the Xeon has a much larger full speed cache as opposed to the G4's external, fractional speed cache) that still puts the dual P4s at more than 50% faster. And this doesn't take into account that the G4 doesn't effectively utilize DDR-SDRAM, and as games become more memory-bandwidth limited (Doom III) the effectivly PC-133 performance on the G4s will be blown away by the 3.2 GB/sec of bandwidth on the Xeons. And if you go into apps besides games and office (like gcc and whatnot) then you're covered by the SPEC benchmarks, and we all know how poorly the G4 does in those.
For multimedia purposes, the G4s are probably at least as fast, due to Altivec. Yes, I know, Altivec isn't an omnipotent silver bullet, but it still kicks ass.
>>>>>
Ha. There is a lot of talk on this (one giant 1200 post thread on ArsTechnica) and the general conclusion is that given the limitations of Apple's platform (namely memory bandwidth) AltiVec really isn't that much faster than SSE2 except in a few special cases where its permute instructions are useful. Otherwise, the P4s smoke the G4.
And if I have the choice between a Dell 20" flat panel and an Apple 23" flat panel, I think I know what I'd take.
>>>>>>..
I wouldn't be too sure. I know Sharp makes some LCD screens that look noticibly better than Apple's, and are cheaper to boot. Since Dell rebrands them, I wouldn't be surprised to see one of these panels show up somewhere in its product line.
Well, what happens is that since the Inspiron 8000, Dell laptops have had the same graphics add-in card). So you could upgrade from a Geforce2 to a GeForce4 MX. We don't know yet if that will hold for the next Inspiron when the NV31-based mobile GPUs come out.
As for the LCD, its magnificent. 1600x1200 on a 15" screen gives 133 dpi, and with ClearType, its like reading a piece of paper.
As for the degrees of freedom, I meant that you can swing about thirty degrees to either side without losing the image on the LCD.