Well, it all depends on just which applications you use and how patient you are.
A while back, I was using a K6-III+ 450MHz system with 192MB of RAM, and this seemed pretty speedy (the K6-III was a very speedy chip for all non-gaming type applications). But then I set up a new system for a friend of mine, and used it for a week before I dropped it off to her. It was only a 750MHz Duron with 256MB of memory, but the difference in performance was VERY noticeable to me. Browsing and reading e-mail was MUCH snappier. I also suddenly realized just how often my usenet reader paused for half a second or so, and how much quicker and easier using this program seemed on the faster system. Programs started up much quicker. Now, some may write this off to just differences in hard drive speed, but really they were using nearly identical drives (the K6-III+ had one of the infamous IBM 75GXP drive (knock on wood, still working great), while the Duron had an IBM 60GXP drive, slightly faster, but only by about 5% or so). Once I gave that system to my friend and went back to my "slow" 450MHz system, I found myself getting rather impatient with a lot of programs (particularly web browsers, mainly Mozilla and occasionally IE or Konqueror). Before long, I was looking into an upgrade.
I eventually got an Athlon 1700+ with 512MB of DDR SDRAM, and let me tell you, it's quick (for today at least!:> ). Unfortunately BOTH of my sticks of memory have gone bad, and guess what, I'm back on the old 450MHz system and boy does it ever seem SLOW! (trust me, I'm checking FedEx's website on an almost hourly basis to see how far my replacement sticks have made it!). Now, I'll be the first to admit, I'm VERY impatient when it comes to computers, however I think that a lot of other people would find that if they actually tried using a somewhat faster system, they would feel that the ugprade is worthwhile. Other people who are more patient or possibly just use less processor/ram intensive applications (the memory access speeds of PPros is quite poor, and the memory access speed of Pentiums is terrible as compared to modern systems, not to mention those also have butt-slow cache, so it's not just procesor-intensive apps that are faster on new systems).
Still, all that being said, if it came down to a choice between broadband and a decently fast computer, I'd take broadband any day. They'll have to pry my cable modem from my cold dead fingers dammit! Fortunately for me, this isn't an issue since broadband is cheaper then any dial-up I could ever find around here. Dial-up is definitely in it's dying days in Canada (more then half of Internet connections are now broadband).
You know, everyone complains about how x86 is just "hacks upon hacks", but guess what, it's the cheapest platform out there and it outperforms damn near everything else!
Right now, only Power4 and the Alpha EV68 are the only chips in the same ballpark as x86 when it comes to raw processing power. In Spec CINT2000, x86 (P4 2.53GHz) manages a score of 896, to beat out ALL other chips. The 1.3GHz Power4 scores 839, while the 1.0GHz Alpha EV68 scores 679. In Spec CFP2000, IBM and DEC/Compaq/HP redeem themselves slightly, with scores of 1266 and 960 respectively, as compared to 879 for the Intel chip.
Alpha has been on the chopping block for some time. Sun? They're SPARC chips and especially their systems as a whole have some advantages, but when it comes to raw processing power, $2000+ USIII chips are beaten by $100 x86 chips by a factor of almost 2 to 1 in most tests.
x86 may have it's problems, and it's certainly not the nicest architecture to write assembly for. But guess what, no one writes in assembly anymore, and compilers have been better optimized for x86 then for any other architecture. So why do people always complain about x86? Well let's see:
Complaint #1: Backwards compatibility holding it back.
Answer: As mentioned above, there are only two chips out there that are faster then x86 chips, and both of those cost a LOT more and neither are all that much faster. Backwards compatibility has also been the reason why x86 has sold so well over the years and is the best supported architecture with the widest selection of software.
Complaint #2: Not enough registers.
Answer: x86 is slightly limited by it's registers, but due to it's CISC nature it doesn't need as many registers as a RISC chip does and no where near as many as RISC chips. What's more, register renaming allows for x86 chips to have a lot more registers then it initially seems like it has (you just can't use them all at the same time).
Complaint #3: x87 stack is dumb
Answer: There is no answer to this, except maybe SSE2. The x87 stack is dumb (I'll grant the complainers this one!:> ). However even this has been mostly worked around by the compilers, to the point where x86 chips are now some of the fastest chips out there at floating point.
Complaint #4: Decoding CISC instructions to RISC instructions is a waste of time.
Answer: The decoding front-end of a moern x86 chip is now a rather small proportion of the total chip and can usually keep the backend well fed (assuming that the decoders can get data from the cache fast enough). What's more, perhaps the most interesting feature of the P4 is it's trace cache, which caches already decoded instructions, there by removing the decoders from ~95% of all instructions executed (AMD doesn't have this yet).
So where does AMD's x86-64 fit into all of this? Well, it maintains backwards compatibility with all previous chips (x86's real strength), it should offer rather impressive performance (particularly the Opteron and 64-bit Athlon should improve the I/O performance of x86 systems, which had traditionally been one of the weak points). It also helps on the register front, in that x86-64 doubles the number of available registers. It doesn't do anything to fix the x87 stack, but even that is mostly compensated for by the generally high quality of x86 compilers. Other then that, the real reason for it is to add seamless 64-bit support.
x86-64 is not just a new feature of x86, it's completely extending the architecture to 64-bits. The Hammer chips will boot a standard i386 install just fine, however there's an entire new architecture being added to the kernel to support these processors! This is a LOT more involved then just adding support for MTRRs, 3DNow, SSE and the like.
I'll be surprised if the Intel Itanium or even the AMD Hammer chips will compare favorably to comparable Alphas on floating point performance, which is very important in many high end applications.
At the moment, only the new 1.0GHz Alpha EV68 chips are faster then the current P4 processors, and the Itanium and Athlon are right up there as well. The rather substatial lead in FP performance that the Alpha used to have has virtually disapeared these days when compared to x86 chips. The very fastest EV67 chips are slower then the fastest x86 chips (note: I'm using Spec CFP2000 for comparison here, if you know of any FP benchmarks that run on both platforms I'd like to hear about them).
As for the future, Alpha's time on this planet is very limited. EV7 is still supposed to come out, and I've heard from reliable sources that it should post some very impressive scores for floating point due to it's HUGE memory bandwidth. However Intel's Itanium 2 is also supposed to post some rather impressive scores (they're talking about 1300-1350 in Spec CFP2000, which would put it ahead of the current champion Power4 processors from IBM). AMD's Hammer won't be any slouch either, as it's on-chip memory controller should boost it's score quite nicely.
Like it or not, Alpha is dead. It's been sold many times and basically salavaged for scrap (Intel now owns most of the old Alpha technology). You're mentioning Titanic is actually quite approriate, because that was the last gasp for Alpha. If you look at new movie production, they're moving to x86, just like all the rest of the world. MIPS might have a future in the embedded space, where it's currently second to ARM. Alpha technology might have a bit of a future in Intel's IA-64 (though I'm skeptical as to how well they'll be able to integrate the pure-RISC Alpha technology into the VLIW IA-64 technology), but as a product on it's own, stick a fork in it.
Hmm.. Just a thought, but maybe this has something to do with the fact that SuSE is also working very closely with AMD for their upcoming "Hammer" line of chips and x86-64 code. Perhaps a bit of a conspiracy theory, but I'm sure it didn't help there chances when it came to a deal with Intel.
Fortunately Gnutella HAS become much better in the past few months. Two very important features have been added to at least three common Gnutella servants (Bearshare, Limewire and Gnucleus), Ultrapeers and swarming. Ultrapeers means that now a significantly lower percentage of the traffic on the Gnutella net is made up of pinging and ponging overhead, while swarming means that you can often download stuff FAST.
If you haven't tried a Gnutella client in the past four or five months (or tried an outdated one the last time), I'd recommend checking them out again. I personally use Gnucleus and find it to be the best of the three, though a lot of that is personal preference (though the lack of ad-ware/spyware helps, plus it's open source, which I like). Ohh, and if you want ultrapeers with Bearshare, you need to use the 3.0 betas, but I understand that those are starting to get reasonably stable now.
There's still a little ways to go, and it would really help if Morpheus used a halfway up-to-date client (they're still usingly mostly Gnucleus 1.6.0 code to the best of my knowledge, which isn't bad, but is missing many important features of the current Gnucleus 1.8.x code). I think that Bearshare also made the right choice by not allowing connections from any of the REALLY outdated clients, and if others did the same I suspect that the network would perform even better.
So, long story short, all those complaining about Gnutella should really give it another shot.
As for the legal aspects of Gnutella, really the RIAA can't stop it, but what's more important, they SHOULDN'T be allowed to stop it. Developers of Gnutella servants really and truly have no more liability to the software distributed on their network then Microsoft has for software transfered using IIS and Internet Explorer. We all know that the RIAA wouldn't THINK of going after Microsoft because MS has the money and legal might to stand up to them, and legally the Gnutella people shouldn't be any different. That being said, I HIGHLY expect the RIAA to sue many Gnutella developers in the near future expressly to put them out of business through legal costs (the fact that the RIAA doesn't have a valid case is of little consequence in the current American legal system.. long live the land of the free).
The dual G4 1Ghz macs also have a L3 cache that might help with RC5 key cracking.
L3 isn't going to do squat for RC5 key cracking. RC5 key cracking is a VERY small algorithm that fits entirely into the L1 cache of a processor. It uses only a handful of instructions VERY often. Basically it comes down to how well a processor is optimized at one or two very specific instructions.
RC5 numbers tell you exactly ONE thing, how good the processor is at RC5 key cracking. It tells you exactly NOTHING about how good the chip will do anything else because no other code is at all like those d.net clients. It doesn't even relate at all to any other encryption type code. Similarly, Seti@Home isn't much better, since it basically just tests how good a chip is at doing FFTs, ie it tests how good a general purpose chip is at acting like a DSP. This might have a very minor relevance to things like AC'97 audio and stuff like that, but that takes such a tiny fraction of a modern CPU's processing power that it's not worth worrying about anyway.
Long story short, Quake 3 isn't a very good benchmark to compare chips with, but it's a hell of a lot better then RC5 numbers are. Unless you buy a computer for the sole purpose of getting a high score on d.net's stats page, RC5 numbers are pretty much worthless.
Quick question: What do you think the fastest chip that Intel currently offers is?
Answer: The P4. Intel's new Itanium, which totally dropped the old compatibility layer and all, is quite a bit slower then the P4, x86 and all.
Simple fact of the matter is that x86 isn't NEARLY as bad as people make it out to be, and what's more, it's well known and VERy well optomized for. The end result is that most of the fastest chips in the world are x86 chips, right now only IBM's monstorous Power4 is faster when it comes to raw processing power (as per Spec benchmarks at least).
With the Hammer series, AMD is getting rid of two of the problems that people kept complaining about with x86. first, it adds 64-bit support, and second, it doubles the number of registers available. Ok, I'm sure that some people still want a lot more registers, though a good compiler should be able to make very effective use of what the Hammer offers.
It's not quite interleaved like we used to have back in the 32-bit SIMM days (or in the 8-bit SIMM days before that), but the end result is very similar. All current Intel RDRAM chipsets use a two-channel RDRAM controller, so it accesses two completely separate sets of RDRAM at the same time. Rambus designed their channel to allow for essentially an infinite number of channels to be stacked together to get as much bandwidth as is needed. This was really one of their big selling points of the technology. The Alpha people are supposed to be taking this one step further with their EV7 (if they ever get it to market), which has a whole bunch of RDRAM channels (8 of them?) on every processor.
As for the DDR (Double Data Rate), I don't know if it's THE generic term for sending singles on both edges of the clock, though it's pretty descriptive for such a task, so I think it works in most cases.
Your average incandescent lightbulb releases about 5% of it's energy as light, and 95% of it as heat. So, technically I suppose we're talking about a 60W light bulb producing ONLY about 57W of heat, while those "AMD space heaters" produce about 60W of heat. There's also the fact that a light bulb uses ONLY 60W of power, while a 60W processor get's it's power through a supply that's only about 85-90% efficient.
Still, all that being considered, the end result is that two light bulbs will heat up a room a LOT more then a single AMD Athlon processor will, and a single light bulb makes FAR more of a difference then one Athlon vs. one PIII, let alone when compared to a P4. In short, if you see anyone complaining about how much their Athlon heats up their room, they're either trolling, of just dumb (or more than likely, they're both).
Where exactly do you think that the heat is going to go from a light then? Sure, it will stay slightly more centralized around the lamp, but the difference is going to be REAL small. Free convection DOES work you know. Yes, forced convection does give you a lot more movement, but only for about 1 foot with most of these fans, maybe 2 feet at most. These are not some huge industrial fans that are going to blow hot air all over the room or anything like that.
High school physics is fine for starting to learn the concepts, university physics is better for figuring out when these concepts apply and when they can be ignored because they aren't going to change thing by any meaningful amount.
AMD replaced one TOTALLY MEANINGLESS measure of performance (clock speed) with another totally meaningless measure of performance (their model numbers).
Think of clock speed like the displacement of a car engine. If everything else is equal, a 3.0L engine will give you a lot more power then a 2.0L engine. However there are plenty of 3.0L engines out there that only produce 180HP or less, while there are some 2.0L engines that produce over 200HP, and that's without taking into account things like turbochargers. What's more, if you're interested in the actual speed of the car, there's a TON of other factors to consider outside of the engine itself.
So, does having a larger displacement make the car "faster"? It might, but it's only one factor in determining the power that the engine will give you, and that in turn is only one (relatively small) factor in determining how fast the car is.
Same thing with processors. Having a higher clock speed might mean a faster computer, but it's only one of MANY factors in determining how powerful the processor is, and that in turn is only one of MANY factors in determining how fast the overall system is.
Ohh.. And if your heatsink falls off, either your cursed with insanely bad luck, or your a moron. If you happen to run a website with the initials THG, then it's certainly the latter. Hint: Tom found the P4 running at 29C without a heatsink. The P4's thermal throttling kicks in at 60-70C. Do you see a problem? There's a reason why Tom has a degree in medicine, and not electrical engineering, computer science, or anything of the like.
Perhaps, but it's rather important to look at what chips they cut prices on.
The desktop P4's were cut by, at most, 29%. This means that it's likely that AMD desktop chips will still be cheaper. These P4 price drops were very typical of what Intel and AMD do every 3 months or so.
Where they really cut the prices were in their mobile P4-M chips, where were just released. This isn't too surprising for two reasons. First, they're brand new and this is their first price cut (ie right after the "I gotta have it NOW at any cost" buyers have been taken care of). Secondly because they just aren't very good chips for notebooks. The P4-M consumes a LOT of power for a notebook chip, particularly the higher speed ones where Intel cut the prices the most. All of these are in the 30W+ range. I'm sure that OEMs like high GHz numbers for their marketing efforts, but designing the cooling for one of these chips must be a bitch! Meanwhile AMD just brought out their new notebook chips which consume 20-25W of power for comperable performance, they have more advanced power saving features and they're cheaper to boot.
In short, Intel probably just couldn't keep going charging the rather ridiculous prices for these laptop chips (they were QUITE expensive). When your competitors have a cooler running chip that performs the same or better and sells for less, marketing and clock speeds can only last you so long.
Hmm.. AMD's highest powered Athlons maxed out at roughly 70W, though typically they were consuming 60W or less.
In other words, I sure as hell hope you don't have the lights on in your room, because they're heating the place up as much or more then those Athlons are! We all KNOW how a single 60W light bulb turns ANY room into a sauna.. uhh.. do we?
The difference in power consumption between AMD's Athlon and Intel's P4, if run full-out, would amount to about 10W (if they're sitting idle they're essentially equal), assuming equal Athlon model number vs. P4 clock frequency.
Given a price of electricity of $0.10/KWH (probably about typical for the U.S.? I dunno, I pay $0.08 CDN, or about $0.05 U.S.). If we assume that the systems run 24/7 for an entire 30 day month, that amounts to 7.2KWH for the month, or $0.72. You would have to run these chips full out, 24/7 for over 5 years before you would save $50 on the Intel chip due to it's lower power consumption. Ohh, and since you use RDRAM, which consumes more power then DDR SDRAM, the difference for the system as a whole is likely to be even smaller.
Long story short, heat and power consumption are NOT an issue when comparing AMD's Athlon and Intel's P4. People may have valid complaints one way or the other, but saving money on your electric bill is not one, and claiming your room gets hot?! Well, I'll write that one off as a poor excuse of a troll post.
RDRAM is based on a new tech that gives you only 16 bits per clock cycle instead of 64 for SDRAM and 128 for DDR-SDRAM. The difference is that you can clock it way up. So there was PC600, PC700 and PC800 RDRAM, again based on MHz
One minor quibble to this otherwise great explination of the different technologies. RDRAM actually uses DDR technology for it's bus as well. This means that PC800 RDRAM actually runs on a 400MHz bus. End result, 400MHz, 16-bits wide and DDR give you the 12,800Mb/s that you mentioned.
FWIW, DDR is by no means specific to DDR SDRAM, or even memory in general. AGP used DDR, modern ATA controllers use DDR, AMD's Hypertransport uses DDR, etc. etc. It's actually a pretty simple way to double the bandwidth on a bus or I/O channel.
How exactly does RDRAM involve any different level of integrated between the processor and the memory?! In fact, how exactly does the processor fit into this at all?!
THE P4 DOES NOT HAVE AN INTEGRATED MEMORY CONTROLLER! (yes, I am yelling).
This means that the chip hasn't got a freaking clue as to what kind of memory is plugged into the chipset, it just makes requests and then waits for the chipset to send it data. This works EXACTLY the same for both DDR and RDRAM. The only difference is on the far side of the chipset, where one using Rambus' porotocol to get to data stored on RDRAM chips, while the other uses one of JEDEC's protocols to get to data stored on DDR SDRAM chips.
The difference is that RDRAM offers higher clock speeds (400/800 or 533/1066MHz DDR) but narrower buses (16 or 32-bits), while DDR has lower bus speeds (133/266 or 166/333MHz DDR) and wider buses (64-bits). Also, in current P4 chipsets, RDRAM uses a pair of memory buses while DDR uses only a single one.
End result? RDRAM offers more bandwidth. It also has slightly higher latency, so performance ends up being pretty close.
I wonder how expensive a graphics card with RDRAM would be, or if it would be any faster?
I've actually suggested that RDRAM might make a good low-cost alternative for video cards. Low cost you ask? Well, RDRAM is expensive as compared to the hugely mass-produced 100-166MHz DDR chips used in PCs, but not as compared to the 300MHz+ DDR chips used in video cards! I think that RDRAM would have a tough time matching the total bandwidth offered by some of the top-end video cards which use DDR (these cards currently have over 8.0GB/s of memory bandwidth), however it could be quite good for fairly low-cost cards which would require slightly less bandwidth.
With an integrated RDRAM memory controller and everything kept within the relative easy confines of a video card, you could probably got stock-PC1066 RDRAM chips to run at about 600MHz or more. That would offer 2.4GB/s of bandwidth per channel, so a dual-channel RDRAM video controller would have 4.8GB/s of bandwidth, putting it just a bit bellow the top-end cards. Now obviously there's more to it then just that, but it seems to me like this might be a decent idea.
Of course, RDRAM video cards HAVE been made before. If my memory serves, Cirrus Logic had an RDRAM based video card a while back, and as others have mentioned, the N64 also used RDRAM. However, these are not using Direct RDRAM like is used in PCs, and they've both long-gone now. The fact that no one else has bothered with RDRAM for video cards suggests to me that there are other technical reasons why it's an undeseriable memory technology for this use.
Hmm.. Interesting how an post that's flat out wrong got modded up... Ahh well..
As others have pointed out, the 5GHz chip was NOT a P4 at all, but just a stripped down portion of the P4. The whole processor is only expected to reduce power consumption by something like 23% in the integer unit, ie it'll do VERY little for the overall power consumption of the chip. A 5GHz P4 on.13um design rules is still going to require a LOT of power (though don't think for a second that Intel isn't doing plenty to reduce power consumption).
Also, that's a great quote, but if I can add another quote from the same article:
"This processor, note, is a 32-bit chip - it's a different presentation from the McKinley that we detailed above."
The 3MB cache is for the McKinley (aka "Itanium 2"), it has NOTHING to do with the 32-bit integer core mentioned above, and it certainly has nothing to do with the P4!
The Intel P4 processor does NOT contain a memory controller, of ANY type! It doesn't matter if you're using RDRAM or DDR memory, both go through the memory controller (on the chipset) and BOTH share the system bus with all other processor chipset I/O.
However, you are touching on an interesting point; integrated memory controllers. Sun does this now with their UltraSparc III (an SDRAM controller.. or is it a DDR controller?), and if (Digital/Compaq/HP/whatever they're called next week) ever gets the Alpha EV7 out, they'll have an integrated RDRAM controller on the processor. For x86 chips, AMD is leading the way here, with their next generation "Hammer" processors having an integrated DDR memory controller. These chips should be out around October or so, and should provide for some very impressive memory performance.
If anything it's AMD that's using the brute force method of processor design. AMD gets it's performance from the Athlon by having a whole whack of functional units (3 integer units and 3 floating point units) and 128K of L1 cache. Intel, on the other hand, is trying for a lot more finesse in their P4 design. They've only got 2 integer units, but they did some rather funky clock multiplying on-die to get them running faster, and they've only got one floating point unit (kinda sorta.. some could make an argument otherwise, particularly with SSE2). The L1 cache of the P4 is only 8K data and 12k uops of trace cache.
Speaking of the trace cache, this is perhaps the most interesting thing that Intel brought out with the P4. This eliminates yet another of the downsides of the x86 architecture (you know, the architecture that everyone says is totally out of date and busted, yet still manages to deliver performance that matches or beats damn near everything else out there for a tiny fraction of the price? yeah, that x86). AMD would probably be wise to copy something like this down the road.
In short, AMD and Intel took different approaches to chip design with their P4 vs. the Athlon (and the Hammer for that matter, which has a processing core that is fairly similar to the Athlon). The end result though is roughly comperable performance and roughly comperable power consumption levels. The main difference is that the Athlons tend to sell for a lot less, which is the reason why I'm currently running an AMD-powered system.
Why is this such a popular misconception? Having 128KB L2 cache doesn't make it a Celeron.
The misconception comes from the fact that it's the same damn die! In fact, the Celeron and the PIII were the same die as well, the Celeron just had 128KB of it's cache disabled. The processor used in the XBox was also from the same die, and also had 128KB of it's cache disabled, however they disabled the cache in such a way so that it was still 8-way set associative.
Ohh, and 8-way vs. 4-way set associative typically makes about 1% difference in performance. The associativity of a cache is a matter of diminishing returns, and going much beyond 2 or 4 way doesn't give you much. Just how often do you expect that you'll need more then 4 chunks of 32-bytes at exact 256KB intervals but not the chunks of memory right beside them? It's pretty rare. I know that Anand made a little claim of a 10% performance increase, but he probably couldn't even explain what set associtivity is, let alone make an argument as to why it would make a 10% performance difference.
You're probably correct about TSMC's yeilds improving, but do you think that nVidia paid TSMC for the bad dies in the first place? Nope. TSMC eats the loss when they've got poor yeilds, and reaps the profits when they have good yeilds. At best they're only going to pass on a small percentage of the saving to nVidia, who in turn are probably only going to pass on a small percentage of their savings to Microsoft.
Next correction? Oh yeah, the memory. The X-Box uses Samsung K4D263238M-QC50 chips, although it could probably use some other identical chips from other manufacturers. You can find info about this chip from Samsung's website. Here's the link. As you can see, it's DDR memory rated for 200MHz, for 400MHz effective performance. I suppose it's not really "DDR400", and it's certainly not "PC4200" memory, however it IS 200/400MHz DDR.
There are major CPU manufacturing plants in the US, Ireland, Germany, Taiwan, etc. etc. There are CPU packaging plants in Malaysia, Puerto Rico, and many other locations.
However, when it comes to CPU design teams, there's VERY little going on outside the United States. The ARM design is all based out of the UK, which is probably the best example. Intel apperentely has some work being done in Isreal, and AMD has a new design center in Germany alongside their fab there, but both AMD and Intel do most of their design in the US. Other than that? Sun's in the US, MIPS is still all US-based to the best of my knowledge, HP and Compaq/Digital were both designing in the US, but they've given up on that now. Even VIA, which is a Taiwanese company, has their design team based out of Austin, Texas.
That being said, the point is probably moot since it makes no sense embedding this technology into the CPU's themselves. But then again, I suppose this law makes no sense at all in the first place.
I don't know that Microsoft's costs would have changed all that much.
Both Intel and nVidia sold MS some fairly low-production-cost chips. In the case of nVidia, they don't even make these chips, so they've got to pay whatever TSMC or UMC charge, and given that these were relatively low cost chips to begin with, the cost that TSMC/UMC charges isn't going to decrease too much. Even if the price does decrease, nVidia may decide to keep the extra profits for themselves, and keep charging MS the same amount. MS is pretty much locked in to using nVidia chips for the lifetime of the X-Box, so nVidia isn't really forced to lower their prices.
As for Intel, they were producing a dirt-cheap chip (a low speed Celeron processors built on a.18um fab line). They could (and possibly already have) decreased costs easily by switching to a.13um fab line, but that's only going to be a marginal decrease in costs given that it was a pretty cheap chip to produce in the first place. Further cost cutting measures are going to help less and less. To top it off, while the original chip was a run-of-the-mill Celeron die, which Intel was making in HUGE quantities, soon this chip will be a low-volume specialty part as Intel moves all it's Celerons first to a.13um fab process (they may or may not be able to use a standard.13um Celeron die for the X-Box, I dunno), and now they're moving to a completely different architecture (Celeron's will become semi-castrated P4s).
Same thing pretty much goes for the hard drive and DVD drive. These producets were all fairly low-cost models ot begin with, and cost cutting just isn't going to trim too much off the bottom line. What's more, in all of these cases MS is outsourcing production of each part to different OEMs, each of whom are going to look for a piece of the pie. I'd even hazard a guess that many of these OEMs took the contract with virtually no margins in the hope that this would turn into a very large volume deal, which it hasn't.
The one area that they can probably really cut costs down is memory. The memory that they're using is DDR400 memory, which used to be a pretty rare specialty part only for graphics cards, but now is becoming a LOT more commonplace and would probably have decreased in price significantly.
So, long story short, production costs probably have decreased somewhat since the initial release, but I doubt that they've dropped very significantly. My guess is that the drop in production cost is quite a bit less then this new drop in retail price.
I doubt that Debian would have the hardware yet (though they might), but SuSE probably does. AMD has been teamed up with SuSEE for quite some time in their x86-64 endeavors.
Ohh, and I really HOPE that the original post was supposed to be a joke, though no one seems to have gotten it. i386 binaries for x86-64 architecture? yeah, it'll work, but it won't help you much!
This is sort of what is happening with x86. x86 is a VERY well known entity. All it's ins and outs have been thoroughly explored a thousand times over. If you want to replace it, you should do so with something that you KNOW will be better in every way. IA-64 is almost certainly not it. Actually there are a LOT of people saying that IA-64 sucks worse then x86, and it's starting from scratch. Ditching x86 for IA-64 would be ditching the 20 years of work that hundreds of people have spent getting x86 to be the one of the fastest architectures on the planet (like it or not, right now the IBM Power4 is the ONLY chip out there with more raw processing power then an AthlonXP or P4).
Besides, the weaknesses of x86 have been known for ages and are slowly being worked around. Think SSE/SSEII/3Dnow! to replace that ridiculous stack-based floating point unit? And now with x86-64, not only is AMD extending the architecture to 64-bits, but they're also doubling the number of general purpose registers that can be used, thereby reducing another major weaknesses in x86. So what are we left with? An architecture that has TONS of funding and support, more then 10x as much software as the next closest competitor, compilers that have been optimized VERY well optimized, and to top it off, you can still run 20 year old applications on the thing if you so desire.
And you want to replace this beast? You better have a REAL good alternative lined up! IA-64 just isn't going to cut it!
The Sunblade gets you a system that's great for I/O stuff, but not raw processing power. In terms of actual processing power, current x86 chips have Sun beat rather solidly, and that gap is only going to widen with the above-mentioned Hammer chips. However raw processing power has never been Sun's strength. Their strength is in I/O and in their software (not to mention support).
Slashdot Mirror
Well, it all depends on just which applications you use and how patient you are.
:> ). Unfortunately BOTH of my sticks of memory have gone bad, and guess what, I'm back on the old 450MHz system and boy does it ever seem SLOW! (trust me, I'm checking FedEx's website on an almost hourly basis to see how far my replacement sticks have made it!). Now, I'll be the first to admit, I'm VERY impatient when it comes to computers, however I think that a lot of other people would find that if they actually tried using a somewhat faster system, they would feel that the ugprade is worthwhile. Other people who are more patient or possibly just use less processor/ram intensive applications (the memory access speeds of PPros is quite poor, and the memory access speed of Pentiums is terrible as compared to modern systems, not to mention those also have butt-slow cache, so it's not just procesor-intensive apps that are faster on new systems).
A while back, I was using a K6-III+ 450MHz system with 192MB of RAM, and this seemed pretty speedy (the K6-III was a very speedy chip for all non-gaming type applications). But then I set up a new system for a friend of mine, and used it for a week before I dropped it off to her. It was only a 750MHz Duron with 256MB of memory, but the difference in performance was VERY noticeable to me. Browsing and reading e-mail was MUCH snappier. I also suddenly realized just how often my usenet reader paused for half a second or so, and how much quicker and easier using this program seemed on the faster system. Programs started up much quicker. Now, some may write this off to just differences in hard drive speed, but really they were using nearly identical drives (the K6-III+ had one of the infamous IBM 75GXP drive (knock on wood, still working great), while the Duron had an IBM 60GXP drive, slightly faster, but only by about 5% or so). Once I gave that system to my friend and went back to my "slow" 450MHz system, I found myself getting rather impatient with a lot of programs (particularly web browsers, mainly Mozilla and occasionally IE or Konqueror). Before long, I was looking into an upgrade.
I eventually got an Athlon 1700+ with 512MB of DDR SDRAM, and let me tell you, it's quick (for today at least!
Still, all that being said, if it came down to a choice between broadband and a decently fast computer, I'd take broadband any day. They'll have to pry my cable modem from my cold dead fingers dammit! Fortunately for me, this isn't an issue since broadband is cheaper then any dial-up I could ever find around here. Dial-up is definitely in it's dying days in Canada (more then half of Internet connections are now broadband).
You know, everyone complains about how x86 is just "hacks upon hacks", but guess what, it's the cheapest platform out there and it outperforms damn near everything else!
:> ). However even this has been mostly worked around by the compilers, to the point where x86 chips are now some of the fastest chips out there at floating point.
Right now, only Power4 and the Alpha EV68 are the only chips in the same ballpark as x86 when it comes to raw processing power. In Spec CINT2000, x86 (P4 2.53GHz) manages a score of 896, to beat out ALL other chips. The 1.3GHz Power4 scores 839, while the 1.0GHz Alpha EV68 scores 679. In Spec CFP2000, IBM and DEC/Compaq/HP redeem themselves slightly, with scores of 1266 and 960 respectively, as compared to 879 for the Intel chip.
Alpha has been on the chopping block for some time. Sun? They're SPARC chips and especially their systems as a whole have some advantages, but when it comes to raw processing power, $2000+ USIII chips are beaten by $100 x86 chips by a factor of almost 2 to 1 in most tests.
x86 may have it's problems, and it's certainly not the nicest architecture to write assembly for. But guess what, no one writes in assembly anymore, and compilers have been better optimized for x86 then for any other architecture. So why do people always complain about x86? Well let's see:
Complaint #1: Backwards compatibility holding it back.
Answer: As mentioned above, there are only two chips out there that are faster then x86 chips, and both of those cost a LOT more and neither are all that much faster. Backwards compatibility has also been the reason why x86 has sold so well over the years and is the best supported architecture with the widest selection of software.
Complaint #2: Not enough registers.
Answer: x86 is slightly limited by it's registers, but due to it's CISC nature it doesn't need as many registers as a RISC chip does and no where near as many as RISC chips. What's more, register renaming allows for x86 chips to have a lot more registers then it initially seems like it has (you just can't use them all at the same time).
Complaint #3: x87 stack is dumb
Answer: There is no answer to this, except maybe SSE2. The x87 stack is dumb (I'll grant the complainers this one!
Complaint #4: Decoding CISC instructions to RISC instructions is a waste of time.
Answer: The decoding front-end of a moern x86 chip is now a rather small proportion of the total chip and can usually keep the backend well fed (assuming that the decoders can get data from the cache fast enough). What's more, perhaps the most interesting feature of the P4 is it's trace cache, which caches already decoded instructions, there by removing the decoders from ~95% of all instructions executed (AMD doesn't have this yet).
So where does AMD's x86-64 fit into all of this? Well, it maintains backwards compatibility with all previous chips (x86's real strength), it should offer rather impressive performance (particularly the Opteron and 64-bit Athlon should improve the I/O performance of x86 systems, which had traditionally been one of the weak points). It also helps on the register front, in that x86-64 doubles the number of available registers. It doesn't do anything to fix the x87 stack, but even that is mostly compensated for by the generally high quality of x86 compilers. Other then that, the real reason for it is to add seamless 64-bit support.
I think you're missing a few things here.
x86-64 is not just a new feature of x86, it's completely extending the architecture to 64-bits. The Hammer chips will boot a standard i386 install just fine, however there's an entire new architecture being added to the kernel to support these processors! This is a LOT more involved then just adding support for MTRRs, 3DNow, SSE and the like.
I'll be surprised if the Intel Itanium or even the AMD Hammer chips will compare favorably to comparable Alphas on floating point performance, which is very important in many high end applications.
At the moment, only the new 1.0GHz Alpha EV68 chips are faster then the current P4 processors, and the Itanium and Athlon are right up there as well. The rather substatial lead in FP performance that the Alpha used to have has virtually disapeared these days when compared to x86 chips. The very fastest EV67 chips are slower then the fastest x86 chips (note: I'm using Spec CFP2000 for comparison here, if you know of any FP benchmarks that run on both platforms I'd like to hear about them).
As for the future, Alpha's time on this planet is very limited. EV7 is still supposed to come out, and I've heard from reliable sources that it should post some very impressive scores for floating point due to it's HUGE memory bandwidth. However Intel's Itanium 2 is also supposed to post some rather impressive scores (they're talking about 1300-1350 in Spec CFP2000, which would put it ahead of the current champion Power4 processors from IBM). AMD's Hammer won't be any slouch either, as it's on-chip memory controller should boost it's score quite nicely.
Like it or not, Alpha is dead. It's been sold many times and basically salavaged for scrap (Intel now owns most of the old Alpha technology). You're mentioning Titanic is actually quite approriate, because that was the last gasp for Alpha. If you look at new movie production, they're moving to x86, just like all the rest of the world. MIPS might have a future in the embedded space, where it's currently second to ARM. Alpha technology might have a bit of a future in Intel's IA-64 (though I'm skeptical as to how well they'll be able to integrate the pure-RISC Alpha technology into the VLIW IA-64 technology), but as a product on it's own, stick a fork in it.
Hmm.. Just a thought, but maybe this has something to do with the fact that SuSE is also working very closely with AMD for their upcoming "Hammer" line of chips and x86-64 code. Perhaps a bit of a conspiracy theory, but I'm sure it didn't help there chances when it came to a deal with Intel.
Fortunately Gnutella HAS become much better in the past few months. Two very important features have been added to at least three common Gnutella servants (Bearshare, Limewire and Gnucleus), Ultrapeers and swarming. Ultrapeers means that now a significantly lower percentage of the traffic on the Gnutella net is made up of pinging and ponging overhead, while swarming means that you can often download stuff FAST.
If you haven't tried a Gnutella client in the past four or five months (or tried an outdated one the last time), I'd recommend checking them out again. I personally use Gnucleus and find it to be the best of the three, though a lot of that is personal preference (though the lack of ad-ware/spyware helps, plus it's open source, which I like). Ohh, and if you want ultrapeers with Bearshare, you need to use the 3.0 betas, but I understand that those are starting to get reasonably stable now.
There's still a little ways to go, and it would really help if Morpheus used a halfway up-to-date client (they're still usingly mostly Gnucleus 1.6.0 code to the best of my knowledge, which isn't bad, but is missing many important features of the current Gnucleus 1.8.x code). I think that Bearshare also made the right choice by not allowing connections from any of the REALLY outdated clients, and if others did the same I suspect that the network would perform even better.
So, long story short, all those complaining about Gnutella should really give it another shot.
As for the legal aspects of Gnutella, really the RIAA can't stop it, but what's more important, they SHOULDN'T be allowed to stop it. Developers of Gnutella servants really and truly have no more liability to the software distributed on their network then Microsoft has for software transfered using IIS and Internet Explorer. We all know that the RIAA wouldn't THINK of going after Microsoft because MS has the money and legal might to stand up to them, and legally the Gnutella people shouldn't be any different. That being said, I HIGHLY expect the RIAA to sue many Gnutella developers in the near future expressly to put them out of business through legal costs (the fact that the RIAA doesn't have a valid case is of little consequence in the current American legal system.. long live the land of the free).
The dual G4 1Ghz macs also have a L3 cache that might help with RC5 key cracking.
L3 isn't going to do squat for RC5 key cracking. RC5 key cracking is a VERY small algorithm that fits entirely into the L1 cache of a processor. It uses only a handful of instructions VERY often. Basically it comes down to how well a processor is optimized at one or two very specific instructions.
RC5 numbers tell you exactly ONE thing, how good the processor is at RC5 key cracking. It tells you exactly NOTHING about how good the chip will do anything else because no other code is at all like those d.net clients. It doesn't even relate at all to any other encryption type code. Similarly, Seti@Home isn't much better, since it basically just tests how good a chip is at doing FFTs, ie it tests how good a general purpose chip is at acting like a DSP. This might have a very minor relevance to things like AC'97 audio and stuff like that, but that takes such a tiny fraction of a modern CPU's processing power that it's not worth worrying about anyway.
Long story short, Quake 3 isn't a very good benchmark to compare chips with, but it's a hell of a lot better then RC5 numbers are. Unless you buy a computer for the sole purpose of getting a high score on d.net's stats page, RC5 numbers are pretty much worthless.
Quick question: What do you think the fastest chip that Intel currently offers is?
Answer: The P4. Intel's new Itanium, which totally dropped the old compatibility layer and all, is quite a bit slower then the P4, x86 and all.
Simple fact of the matter is that x86 isn't NEARLY as bad as people make it out to be, and what's more, it's well known and VERy well optomized for. The end result is that most of the fastest chips in the world are x86 chips, right now only IBM's monstorous Power4 is faster when it comes to raw processing power (as per Spec benchmarks at least).
With the Hammer series, AMD is getting rid of two of the problems that people kept complaining about with x86. first, it adds 64-bit support, and second, it doubles the number of registers available. Ok, I'm sure that some people still want a lot more registers, though a good compiler should be able to make very effective use of what the Hammer offers.
It's not quite interleaved like we used to have back in the 32-bit SIMM days (or in the 8-bit SIMM days before that), but the end result is very similar. All current Intel RDRAM chipsets use a two-channel RDRAM controller, so it accesses two completely separate sets of RDRAM at the same time. Rambus designed their channel to allow for essentially an infinite number of channels to be stacked together to get as much bandwidth as is needed. This was really one of their big selling points of the technology. The Alpha people are supposed to be taking this one step further with their EV7 (if they ever get it to market), which has a whole bunch of RDRAM channels (8 of them?) on every processor.
As for the DDR (Double Data Rate), I don't know if it's THE generic term for sending singles on both edges of the clock, though it's pretty descriptive for such a task, so I think it works in most cases.
Your average incandescent lightbulb releases about 5% of it's energy as light, and 95% of it as heat. So, technically I suppose we're talking about a 60W light bulb producing ONLY about 57W of heat, while those "AMD space heaters" produce about 60W of heat. There's also the fact that a light bulb uses ONLY 60W of power, while a 60W processor get's it's power through a supply that's only about 85-90% efficient.
Still, all that being considered, the end result is that two light bulbs will heat up a room a LOT more then a single AMD Athlon processor will, and a single light bulb makes FAR more of a difference then one Athlon vs. one PIII, let alone when compared to a P4. In short, if you see anyone complaining about how much their Athlon heats up their room, they're either trolling, of just dumb (or more than likely, they're both).
Where exactly do you think that the heat is going to go from a light then? Sure, it will stay slightly more centralized around the lamp, but the difference is going to be REAL small. Free convection DOES work you know. Yes, forced convection does give you a lot more movement, but only for about 1 foot with most of these fans, maybe 2 feet at most. These are not some huge industrial fans that are going to blow hot air all over the room or anything like that.
High school physics is fine for starting to learn the concepts, university physics is better for figuring out when these concepts apply and when they can be ignored because they aren't going to change thing by any meaningful amount.
AMD replaced one TOTALLY MEANINGLESS measure of performance (clock speed) with another totally meaningless measure of performance (their model numbers).
Think of clock speed like the displacement of a car engine. If everything else is equal, a 3.0L engine will give you a lot more power then a 2.0L engine. However there are plenty of 3.0L engines out there that only produce 180HP or less, while there are some 2.0L engines that produce over 200HP, and that's without taking into account things like turbochargers. What's more, if you're interested in the actual speed of the car, there's a TON of other factors to consider outside of the engine itself.
So, does having a larger displacement make the car "faster"? It might, but it's only one factor in determining the power that the engine will give you, and that in turn is only one (relatively small) factor in determining how fast the car is.
Same thing with processors. Having a higher clock speed might mean a faster computer, but it's only one of MANY factors in determining how powerful the processor is, and that in turn is only one of MANY factors in determining how fast the overall system is.
Ohh.. And if your heatsink falls off, either your cursed with insanely bad luck, or your a moron. If you happen to run a website with the initials THG, then it's certainly the latter. Hint: Tom found the P4 running at 29C without a heatsink. The P4's thermal throttling kicks in at 60-70C. Do you see a problem? There's a reason why Tom has a degree in medicine, and not electrical engineering, computer science, or anything of the like.
Perhaps, but it's rather important to look at what chips they cut prices on.
The desktop P4's were cut by, at most, 29%. This means that it's likely that AMD desktop chips will still be cheaper. These P4 price drops were very typical of what Intel and AMD do every 3 months or so.
Where they really cut the prices were in their mobile P4-M chips, where were just released. This isn't too surprising for two reasons. First, they're brand new and this is their first price cut (ie right after the "I gotta have it NOW at any cost" buyers have been taken care of). Secondly because they just aren't very good chips for notebooks. The P4-M consumes a LOT of power for a notebook chip, particularly the higher speed ones where Intel cut the prices the most. All of these are in the 30W+ range. I'm sure that OEMs like high GHz numbers for their marketing efforts, but designing the cooling for one of these chips must be a bitch! Meanwhile AMD just brought out their new notebook chips which consume 20-25W of power for comperable performance, they have more advanced power saving features and they're cheaper to boot.
In short, Intel probably just couldn't keep going charging the rather ridiculous prices for these laptop chips (they were QUITE expensive). When your competitors have a cooler running chip that performs the same or better and sells for less, marketing and clock speeds can only last you so long.
Hmm.. AMD's highest powered Athlons maxed out at roughly 70W, though typically they were consuming 60W or less.
In other words, I sure as hell hope you don't have the lights on in your room, because they're heating the place up as much or more then those Athlons are! We all KNOW how a single 60W light bulb turns ANY room into a sauna.. uhh.. do we?
The difference in power consumption between AMD's Athlon and Intel's P4, if run full-out, would amount to about 10W (if they're sitting idle they're essentially equal), assuming equal Athlon model number vs. P4 clock frequency.
Given a price of electricity of $0.10/KWH (probably about typical for the U.S.? I dunno, I pay $0.08 CDN, or about $0.05 U.S.). If we assume that the systems run 24/7 for an entire 30 day month, that amounts to 7.2KWH for the month, or $0.72. You would have to run these chips full out, 24/7 for over 5 years before you would save $50 on the Intel chip due to it's lower power consumption. Ohh, and since you use RDRAM, which consumes more power then DDR SDRAM, the difference for the system as a whole is likely to be even smaller.
Long story short, heat and power consumption are NOT an issue when comparing AMD's Athlon and Intel's P4. People may have valid complaints one way or the other, but saving money on your electric bill is not one, and claiming your room gets hot?! Well, I'll write that one off as a poor excuse of a troll post.
RDRAM is based on a new tech that gives you only 16 bits per clock cycle instead of 64 for SDRAM and 128 for DDR-SDRAM. The difference is that you can clock it way up. So there was PC600, PC700 and PC800 RDRAM, again based on MHz
One minor quibble to this otherwise great explination of the different technologies. RDRAM actually uses DDR technology for it's bus as well. This means that PC800 RDRAM actually runs on a 400MHz bus. End result, 400MHz, 16-bits wide and DDR give you the 12,800Mb/s that you mentioned.
FWIW, DDR is by no means specific to DDR SDRAM, or even memory in general. AGP used DDR, modern ATA controllers use DDR, AMD's Hypertransport uses DDR, etc. etc. It's actually a pretty simple way to double the bandwidth on a bus or I/O channel.
How exactly does RDRAM involve any different level of integrated between the processor and the memory?! In fact, how exactly does the processor fit into this at all?!
THE P4 DOES NOT HAVE AN INTEGRATED MEMORY CONTROLLER! (yes, I am yelling).
This means that the chip hasn't got a freaking clue as to what kind of memory is plugged into the chipset, it just makes requests and then waits for the chipset to send it data. This works EXACTLY the same for both DDR and RDRAM. The only difference is on the far side of the chipset, where one using Rambus' porotocol to get to data stored on RDRAM chips, while the other uses one of JEDEC's protocols to get to data stored on DDR SDRAM chips.
The difference is that RDRAM offers higher clock speeds (400/800 or 533/1066MHz DDR) but narrower buses (16 or 32-bits), while DDR has lower bus speeds (133/266 or 166/333MHz DDR) and wider buses (64-bits). Also, in current P4 chipsets, RDRAM uses a pair of memory buses while DDR uses only a single one.
End result? RDRAM offers more bandwidth. It also has slightly higher latency, so performance ends up being pretty close.
I wonder how expensive a graphics card with RDRAM would be, or if it would be any faster?
I've actually suggested that RDRAM might make a good low-cost alternative for video cards. Low cost you ask? Well, RDRAM is expensive as compared to the hugely mass-produced 100-166MHz DDR chips used in PCs, but not as compared to the 300MHz+ DDR chips used in video cards! I think that RDRAM would have a tough time matching the total bandwidth offered by some of the top-end video cards which use DDR (these cards currently have over 8.0GB/s of memory bandwidth), however it could be quite good for fairly low-cost cards which would require slightly less bandwidth.
With an integrated RDRAM memory controller and everything kept within the relative easy confines of a video card, you could probably got stock-PC1066 RDRAM chips to run at about 600MHz or more. That would offer 2.4GB/s of bandwidth per channel, so a dual-channel RDRAM video controller would have 4.8GB/s of bandwidth, putting it just a bit bellow the top-end cards. Now obviously there's more to it then just that, but it seems to me like this might be a decent idea.
Of course, RDRAM video cards HAVE been made before. If my memory serves, Cirrus Logic had an RDRAM based video card a while back, and as others have mentioned, the N64 also used RDRAM. However, these are not using Direct RDRAM like is used in PCs, and they've both long-gone now. The fact that no one else has bothered with RDRAM for video cards suggests to me that there are other technical reasons why it's an undeseriable memory technology for this use.
Hmm.. Interesting how an post that's flat out wrong got modded up... Ahh well..
.13um design rules is still going to require a LOT of power (though don't think for a second that Intel isn't doing plenty to reduce power consumption).
As others have pointed out, the 5GHz chip was NOT a P4 at all, but just a stripped down portion of the P4. The whole processor is only expected to reduce power consumption by something like 23% in the integer unit, ie it'll do VERY little for the overall power consumption of the chip. A 5GHz P4 on
Also, that's a great quote, but if I can add another quote from the same article:
"This processor, note, is a 32-bit chip - it's a different presentation from the McKinley that we detailed above."
The 3MB cache is for the McKinley (aka "Itanium 2"), it has NOTHING to do with the 32-bit integer core mentioned above, and it certainly has nothing to do with the P4!
Ummm.. What the hell are you talking about?!?!
The Intel P4 processor does NOT contain a memory controller, of ANY type! It doesn't matter if you're using RDRAM or DDR memory, both go through the memory controller (on the chipset) and BOTH share the system bus with all other processor chipset I/O.
However, you are touching on an interesting point; integrated memory controllers. Sun does this now with their UltraSparc III (an SDRAM controller.. or is it a DDR controller?), and if (Digital/Compaq/HP/whatever they're called next week) ever gets the Alpha EV7 out, they'll have an integrated RDRAM controller on the processor. For x86 chips, AMD is leading the way here, with their next generation "Hammer" processors having an integrated DDR memory controller. These chips should be out around October or so, and should provide for some very impressive memory performance.
Err, "brute force"?
If anything it's AMD that's using the brute force method of processor design. AMD gets it's performance from the Athlon by having a whole whack of functional units (3 integer units and 3 floating point units) and 128K of L1 cache. Intel, on the other hand, is trying for a lot more finesse in their P4 design. They've only got 2 integer units, but they did some rather funky clock multiplying on-die to get them running faster, and they've only got one floating point unit (kinda sorta.. some could make an argument otherwise, particularly with SSE2). The L1 cache of the P4 is only 8K data and 12k uops of trace cache.
Speaking of the trace cache, this is perhaps the most interesting thing that Intel brought out with the P4. This eliminates yet another of the downsides of the x86 architecture (you know, the architecture that everyone says is totally out of date and busted, yet still manages to deliver performance that matches or beats damn near everything else out there for a tiny fraction of the price? yeah, that x86). AMD would probably be wise to copy something like this down the road.
In short, AMD and Intel took different approaches to chip design with their P4 vs. the Athlon (and the Hammer for that matter, which has a processing core that is fairly similar to the Athlon). The end result though is roughly comperable performance and roughly comperable power consumption levels. The main difference is that the Athlons tend to sell for a lot less, which is the reason why I'm currently running an AMD-powered system.
Why is this such a popular misconception? Having 128KB L2 cache doesn't make it a Celeron.
The misconception comes from the fact that it's the same damn die! In fact, the Celeron and the PIII were the same die as well, the Celeron just had 128KB of it's cache disabled. The processor used in the XBox was also from the same die, and also had 128KB of it's cache disabled, however they disabled the cache in such a way so that it was still 8-way set associative.
Ohh, and 8-way vs. 4-way set associative typically makes about 1% difference in performance. The associativity of a cache is a matter of diminishing returns, and going much beyond 2 or 4 way doesn't give you much. Just how often do you expect that you'll need more then 4 chunks of 32-bytes at exact 256KB intervals but not the chunks of memory right beside them? It's pretty rare. I know that Anand made a little claim of a 10% performance increase, but he probably couldn't even explain what set associtivity is, let alone make an argument as to why it would make a 10% performance difference.
You're probably correct about TSMC's yeilds improving, but do you think that nVidia paid TSMC for the bad dies in the first place? Nope. TSMC eats the loss when they've got poor yeilds, and reaps the profits when they have good yeilds. At best they're only going to pass on a small percentage of the saving to nVidia, who in turn are probably only going to pass on a small percentage of their savings to Microsoft.
Next correction? Oh yeah, the memory. The X-Box uses Samsung K4D263238M-QC50 chips, although it could probably use some other identical chips from other manufacturers. You can find info about this chip from Samsung's website. Here's the link. As you can see, it's DDR memory rated for 200MHz, for 400MHz effective performance. I suppose it's not really "DDR400", and it's certainly not "PC4200" memory, however it IS 200/400MHz DDR.
There are major CPU manufacturing plants in the US, Ireland, Germany, Taiwan, etc. etc. There are CPU packaging plants in Malaysia, Puerto Rico, and many other locations.
However, when it comes to CPU design teams, there's VERY little going on outside the United States. The ARM design is all based out of the UK, which is probably the best example. Intel apperentely has some work being done in Isreal, and AMD has a new design center in Germany alongside their fab there, but both AMD and Intel do most of their design in the US. Other than that? Sun's in the US, MIPS is still all US-based to the best of my knowledge, HP and Compaq/Digital were both designing in the US, but they've given up on that now. Even VIA, which is a Taiwanese company, has their design team based out of Austin, Texas.
That being said, the point is probably moot since it makes no sense embedding this technology into the CPU's themselves. But then again, I suppose this law makes no sense at all in the first place.
I don't know that Microsoft's costs would have changed all that much.
.18um fab line). They could (and possibly already have) decreased costs easily by switching to a .13um fab line, but that's only going to be a marginal decrease in costs given that it was a pretty cheap chip to produce in the first place. Further cost cutting measures are going to help less and less. To top it off, while the original chip was a run-of-the-mill Celeron die, which Intel was making in HUGE quantities, soon this chip will be a low-volume specialty part as Intel moves all it's Celerons first to a .13um fab process (they may or may not be able to use a standard .13um Celeron die for the X-Box, I dunno), and now they're moving to a completely different architecture (Celeron's will become semi-castrated P4s).
Both Intel and nVidia sold MS some fairly low-production-cost chips. In the case of nVidia, they don't even make these chips, so they've got to pay whatever TSMC or UMC charge, and given that these were relatively low cost chips to begin with, the cost that TSMC/UMC charges isn't going to decrease too much. Even if the price does decrease, nVidia may decide to keep the extra profits for themselves, and keep charging MS the same amount. MS is pretty much locked in to using nVidia chips for the lifetime of the X-Box, so nVidia isn't really forced to lower their prices.
As for Intel, they were producing a dirt-cheap chip (a low speed Celeron processors built on a
Same thing pretty much goes for the hard drive and DVD drive. These producets were all fairly low-cost models ot begin with, and cost cutting just isn't going to trim too much off the bottom line. What's more, in all of these cases MS is outsourcing production of each part to different OEMs, each of whom are going to look for a piece of the pie. I'd even hazard a guess that many of these OEMs took the contract with virtually no margins in the hope that this would turn into a very large volume deal, which it hasn't.
The one area that they can probably really cut costs down is memory. The memory that they're using is DDR400 memory, which used to be a pretty rare specialty part only for graphics cards, but now is becoming a LOT more commonplace and would probably have decreased in price significantly.
So, long story short, production costs probably have decreased somewhat since the initial release, but I doubt that they've dropped very significantly. My guess is that the drop in production cost is quite a bit less then this new drop in retail price.
I doubt that Debian would have the hardware yet (though they might), but SuSE probably does. AMD has been teamed up with SuSEE for quite some time in their x86-64 endeavors.
Ohh, and I really HOPE that the original post was supposed to be a joke, though no one seems to have gotten it. i386 binaries for x86-64 architecture? yeah, it'll work, but it won't help you much!
vs. the devil you don't?
This is sort of what is happening with x86. x86 is a VERY well known entity. All it's ins and outs have been thoroughly explored a thousand times over. If you want to replace it, you should do so with something that you KNOW will be better in every way. IA-64 is almost certainly not it. Actually there are a LOT of people saying that IA-64 sucks worse then x86, and it's starting from scratch. Ditching x86 for IA-64 would be ditching the 20 years of work that hundreds of people have spent getting x86 to be the one of the fastest architectures on the planet (like it or not, right now the IBM Power4 is the ONLY chip out there with more raw processing power then an AthlonXP or P4).
Besides, the weaknesses of x86 have been known for ages and are slowly being worked around. Think SSE/SSEII/3Dnow! to replace that ridiculous stack-based floating point unit? And now with x86-64, not only is AMD extending the architecture to 64-bits, but they're also doubling the number of general purpose registers that can be used, thereby reducing another major weaknesses in x86. So what are we left with? An architecture that has TONS of funding and support, more then 10x as much software as the next closest competitor, compilers that have been optimized VERY well optimized, and to top it off, you can still run 20 year old applications on the thing if you so desire.
And you want to replace this beast? You better have a REAL good alternative lined up! IA-64 just isn't going to cut it!
The Sunblade gets you a system that's great for I/O stuff, but not raw processing power. In terms of actual processing power, current x86 chips have Sun beat rather solidly, and that gap is only going to widen with the above-mentioned Hammer chips. However raw processing power has never been Sun's strength. Their strength is in I/O and in their software (not to mention support).