There's a lot more to it than just how fast of a chip they can produce. Namely, how can they make the most money?
Let's say that AMD could produce an Athlon64 4500+ right now. Would it be in their best interest to release it? Not really. The fastest chips tend to be the lowest yield - and it would greatly push down the cost of the lesser chips. Their best interest is to release a chip that's juuuuust fast enough to keep up with or beat Intel, and keep the pricing high enough to come in just a bit under intel's pricing for competitive models.
In fact, each time Intel has actually released a faster chip lately, AMD has released a faster one as if it were no trouble at all. The way they've been doing, I wouldn't be surprised if they could release faster chips if there were an economic incentive to do so.
One more thing that you've missed is that the variant that does appear very strongly to be transmitted through the consumption of cows is not the classical CJD, but a new variant called (uninterestingly) vCJD.
The new variant occurs at a MUCH higher rate than normal CJD, and strikes much younger people. The median age for CJD is around 68, the median age for vCJD is around 28.
Furthermore, of all diagnosed vCJD to date, virtually all of had multi-year expose in Britain during the peak of the BSE epidemic.
There has never been a single vCJD case in a country where cows did not have BSE.
It's not like there's one single prion disease. There are a lot of them. And what's more, even within one prion disease, there are many different variants. Within scrapie alone, there are at least 15 different variations.
vCJD is a little worrisome in that it does appear (very strongly!) to jump from cattle to humans. On the other hand, it appears to require direct consumption of the infected animal. Stop eating beef, problem completely solved.
A lot of people have heard of "Mad Cow". Some of them have even heard of BSE or CFD. And most people don't realize that this is nothing novel, nothing new, and not at all limitted to cows.
The result of these prions in the brain is spongiform encephaly - literally, holes being eaten in your brain by the prion's interaction. Not a very fun thing!
Now, prion-caused sponfigorm encephalies have been found in a good number of animals. At a minimum, humans, goats, sheep, cows, squirrels, deer, elk, etc..
In cows, the condition is called "BSE" ("Bovine Spongiform Encephaly"). In humans, it's usually called Creutzfelt-Jacob's Disease (I'm sure I murdered the spelling). Those are merely terms for the resultant condition from the prion infection.
Now, the prion responsible for BSE isn't all that bad, as far as infectious prions go. First, it's not really transmissible in cows without the direct ingestion of infected nervous tissue. That means that if we just didn't feed cows ground up cows or ground up sheep, a very large part of the problem is solved.
However, there are other prion agents that are a bit nastier. In the case of CWD and scrapie, the prions have been shown to be transmissible to other individuals through the environment if (a) a n infected carcass or (b) excreta from an infected animal is in the area. Even better, even after all of the animals have left the area, CWD and scrapie agents have been shown to remain and still be contagious to other individuals years later.
Here's the good part: Researchers have already found genes that cause resistance to prion infections, or at least to certain types of them. The genes are found most commonly (and most heavily) in populations that practiced (or still practice) cannibalism. On the down side, it's not something as nice as getting infected and developping an immunity - we're talking about the cannibalistic societies being mostly wiped out by prion-based diseases, leaving only those (luckily) able to resist as the sole survivors, to pass along the genes to their offspring.
(1) There has not been a single proven case of your skull being smashed by me hitting it with a mallet.
Just the same, we're reasonably sure that a smashed skull would result.
There are enough cases of spongiform encephalies being transmitted between species through ingestion of infected nervous tissue that your argument is a non-starter. In fact, if you look into the spread of prion-caused spongiform encephalies, you'll find that there are variants that don't require consumption of the infected animal at all.
One case in point being CWD and scrapie. Merely the presence of excrement from infected animals has been shown sufficient to infect other animals. To make it better, the prion agents remain in the environment (and infections) even after two years without any animals in the region at all.
Well, your line of thinking isn't all that unsound. But here's what we do know:
1. Prions do exist, and do infect the nervous tissue of cows (and a lot of other animals, actually.)
2. Prions are rarely (if ever) able to be found outside of the nervous tissues.
3. Consuming infected nervous tissue does appear (quite strongly, in fact) to infect the consumer.
So, our classical line of thinking about digestion and protein absorbtion doesn't quite fit what we're seeing. What does that mean, class? That's right, we'll have to keep researching both what's really going on with the digestion (or lack thereof) of prions, and the absorbtion/transmission.
Note that while BSE doesn't appeaar to be transmissible in cattle without the cannibalistic ingestion that results from human intervention, there are other related prion diseases such as CWD and scrapie that do appear to be transmitted through environmental factors. In fact, in scrapie epidemics, it has been shown that scrapie agents may remain in the local environment for years after the outbreak.
So, your hard drive uses 5v * 0.3A = 1.5 watts on the 5V line, and 12V * 0.5A = 6 watts on the +12V line. Total wattage: 7.5 watts. And those are *maximum* figures.
As you can see, most of the power is drawn on the 12V line. I've seen cheap 400-watt power supplies that could supply 13 amps on the +12V line, but I've seen quality 300-watt power supplies that could supply 22(!) amps on the +12V line.
There are various limitations on power supplies: First, you don't want to draw more wattage overall than it's rated for, you're asking to start hearing "BANG". Second, you don't want to draw more current on one voltage line than it's rated for. Last, they often have combined wattage limits, such as "Max 200 watts between +12V and +5V lines".
Despite that, it's not all that tough to size a power supply: It takes a very serious machine to overload a reasonable power supply. Even with a relatively cheap 350- or 400-watt power supply, you're probably not going to overload it unless you've either got (a) dual processers, and a good assortment of other accessories, or (b) a VERY high-draw video card and a VERY high-draw processer.
Look at the drives. On the label, it will tell you how much current it draws on the +12V and +5V lines.
Now look at the power supply. On the label, it will tell you how much current it can provide on the +12V and +5V lines.
Of course, the hard drives aren't the only devices using the system: You'll have to look at other devices to get a better idea.
Here's an example: I just built a machine with 8x300gb drives, with dual 1.266GHz P3's. The el-cheapo Codegen 400-watt power supply that came with the case is more than enough to power everything, and it doesn't push nearly as many amps on the +12V line (where most of the HDD wattage goes) as even my 325-watt Enermax power supply.
I just built a 2+ terabyte machine, for a lot less money. It doesn't have to be that expensive. Here's how to do it for a lot less:
1. Throw out the idea of the 3ware card. I've used them. They're overpriced pieces of crap. Responsiveness under load is horrible. Linux's software RAID works much better. Savings: $770
2. Use 300-gig drives. Fewer of them means a much cheaper casew ($124 in my case). Savings: $200
3. Two 550-watt power supplies? Stop believing the crap in the overclocking forums. The relatively cheap 400-watt power supply that came with the case easily powers eight 300-gig drives and a dual processer motherboard. Savings: $204
(Out of a good number of single- and multi-CPU systems, I have exactly two that couldn't run on a 400-watt power supply: And they both have four processers in them.)
Total savings: >$1,000
If you're thinking of using Windows on this, well, maybe you need the hardware RAID controller. Under Linux, the software RAID implementation is second to none - and I can guarantee that your host CPU will crank out parity in amounts that one of 3ware's cards can't even dream of. Combined with Samba, nothing on your network will know that it's not talking to a Windows machine.
I bought one of these, eight 300-gig drives, and some of these to mount the drives in the 5.25" bays. Add in a couple of Highpoint IDE controllers, some cheap power splitters, and you're set.
Eight drives in RAID 5 gives me two terabytes of storage, mounting the drives in the 5.25" bays gives room for airflow, and the fans on the mounting brackets keep them somewhat cool for increased longevity. If memory serves, I could still shove in at least one extra drive.
Now the ultimate machine would have SGIs architecute (memory) and #CPUs per node using the PowerPC CPU
If we're going for ultimate, I'd rather use the Power line of chips than the PowerPC. The PowerPC is a good chip, but the Power line is an entirely new level. : )
Linux has, for quite some time, had a lot of effort put into support for machines other than x86 - which implies 64-bitness, more processers, NUMA, and all of those goodies.
Windows, on the other hand, has (with a few exceptions, now gone) been tied to the x86 architecture - meaning there's never even been a need to support more than a relatively small number of processers, and not even the possibility (until recently) of 64-bitness. (Yes, there was NT for the Alpha. It's gone.)
With Linux, there are people saying "I'd like this to not only run on a 32-CPU NUMA system, I'd like it to run well. With Windows, you have people saying "Ah, we'll never have to worry about more than four or maybe eight processers."
In particular to this machine, one of the large benefits of Linux is that it has pretty good support for NUMA (Non-Uniform Memory Architecture). IBM, who has a lot of experience in the area, has put a very large amount of development into it. That's paid off extremely well not only in supporting IBM's Power line, but also Opterons and other NUMAs.
With Windows, until the Opterons, NUMA was one of those things that Billy G. never even imagined supporting, because it just wasn't part of the x86 world.
As another example, the o(1) scheduler that was introduced along the 2.4 development vastly improved multi-processer scalability. Windows just doesn't have such cool things, and it certainly doesn't have the ability to switch to a new scheduler whenever they want!
But on a dollar-for-dollar basis, the integer work sucked. I had a Compaq rep trying to sell me on Alpha for database work. He loaned me a $25,000 Alpha for a week. Our $14,000 Xeon beat it very handily.
In general businesses try to "segment" their market, and when you vertically own the market, it is in your best interest to "make" everything, including different socket types
The problem has been that in the high-dollar area (Xeons), they've been sticking the consumer with slower busses than the desktop line. Until recently, when there was no other option in the price range, people just sucked it up and dealt with it. Now that there is good competition, that's going to have to change. This is probably a way for them to get some fast busses into their high-dollar (Xeon and Itanium) lines without it looking like they could have done it all along.
The Itanium sucks for a lot of things, but when it comes to floating-point work, they're not half-bad - in fact, they are the tops of the currently published SpecFP scores.
Think of them like an Alpha: Very expensive, very fast at FP work, not so great at Integer work.
It's not admitting defeat, it's getting rid of something that hasn't made much sense.
The Xeon line has traditionally been both the more lucrative and the more "stable" line, on the idea that businesses wanted long-term stability. When you bought a Xeon motherboard, you knew that it would be compatible with some of the upcoming Xeons in FSB and slot, and with a VRM module, even for different voltages.
A very large side-effect is that the Xeon line was tied to slower FSB/memory rates, and a lot more expensive. In reality, most people don't upgrade their Xeon chips - they move to a new architecture (P2->P3->P4, etc.)
Looking at the P2/P3 Xeons, it was long after the shift to a 133MHz FSB when the Xeons played catch-up. And it's just now that the P4 Xeons are getting away from the 533MHz FSB.
So, they're taking their most expensive, most lucrative processers, and hindering the performance. While it worked while Intel was the 800-pound gorilla (and you had no other choice in 2-way, let alone 4-way machines), that's no longer the case. There's competition in the market - good competition. It was a real kick in the pants to shell out $5000 to $15,000 on Xeon systems that had a slower FSB than the $500 desktop counterparts!
So, now, they're going to have to do the sensible thing: Start giving actual performance in exchange for customer money. Rather than have the "high-end bus" and the "low-end" or "mid-range" bus, it makes a lot of sense to just have all of them use one bus design - theoretically, a fast bus.
The Athlon bus was designed so that you could, in theory, plug an Alpha into an Athlon board
I really don't think that was the idea. AMD was looking for a front-side bus design, and Intel's wasn't an option. The Alpha design was available for the right price, and so it was used.
Do you believe that x86-32 BIOS code is going to work to get an Itanium CPU
I believe that Intel has already stated that they want to make some radical changes to the way the BIOS works.
Sorry, Intel; I'd say that this is the beginning of the end.
Unless they're able to really shift direction, I'd agree.
I don't think that AMD will ever beat out Intel completely, but if they keep making the right moves, they will be able to force Intel to radically change how it does business.
Intel has traditionally enjoyed very, very healthy profit margins. AMD, on the other hand, has traditionally lost money - it's only been in the last quarter or two that they've even posted a profit. To bring a company from the times of the K6 and K6-2 to having an incredible offering like the Opteron, while having to be so cautious with money, shows that AMD can be resourceful, and compete even when it's an uphill battle.
Intel, on the other hand, has enough infrastructure, capital, partners, investment, and diversification that it will never be beat out entirely. However, as AMD starts to make more and more profits (by eating into Intel's most lucrative market!), they'll (theoretically) be able to step up the competition and give Intel an even stronger opponent. If they do that, Intel will have to start being much more frugal and resourceful.
Up until a few years ago, the AMD-Intel battle was pretty one-sided. Now, things are going to be a lot more fun to watch.
With Opterons coming in much less expensive than the Itaniums, moving to a common socket with the Xeons isn't going to help much.
On the other hand, with Opterons offering far better scalability and performance than Xeons, moving to a common socket with Itaniums isn't going to help out much.
Even on 2-way machines, Opterons show much better scalability than Xeons. As the number of CPUs increase, the Opteron architecture (when coupled with a supporting OS) allows it to shine more and more. With 8-way Opterons coming out fairly soon, Intel needs to come up with something fast - they're losing one of their most lucrative markets to AMD.
I was in a building just a couple of hours ago with two OC192s. That totals up to 20 gigabits per second.
The two OC192s easily handle all of the company's voice and data communication in the city, which is not at all a small amount.
Very few people realize just how much data 10 gigabits per second really is, and fewer still realize how much hardware it takes to be able to send/receive/route/process data at that rate without even looking at the network card.
storage area networks. at 10Gb/s, SAN's are not a slow data storage medium, but an ultra fast storage medium, that may ever be faster that your local hard drive.
You still think you're going to get 10 gigabits to/from that thing? Think again. That's over a gigabyte per second. It would take at least twenty disks in RAID 0 just to barely get up to that speed. However, putting twenty disks in a RAID 0 array means that your realistic MTBKYBG (Mean Time Between Kissing Your Butt Goodbye) would be something like a month.
Want 1+0? Well, then you better have at least FORTY disks in the SAN. Want RAID 5? Well... tough. It's going to be quite a while before you find any RAID 5 solution that's going to write at anything near a gigabyte per second. In fact, even READING at a gigabyte per second is going to be very expensive with RAID 5.
Sure, there are SANs that can dish out over one gigabit. But hitting ten gigabits is going to be tough, and not just because Ethernet has so many limitations.
OR, how about full diskless clients, with BIOS manufactures being able to map specific locations on the network to a local harddrive
Sorry. Virtually no desktop-level computers can even realistically take full advantage of even a one gigabit card at the moment. Slapping a bigger card in the machine isn't going to overcome architectural limitations.
CLUSTERING software. imagin a beow...., how about openmosix!
What? Serious clusters don't even use gigabit ethernet, it's too much of a hacked-together kludge. Bumping it up another order of magnitude will probably make things worse, rather than better.
EAEP scanners and video cameras = awesome. 1 640x480 camera compressing plain mpeg1 video can saturate a 100Mbps network pretty quickly
So, use MPEG2, or an even better CODEC. It'll be cheaper to implement that than to implement something that could handle a 10gigE network.
HDTV delivery, speeds like this would allow High Definition media delivery without bandwidth pains
Amazingly, there's already plenty of bandwidth to deliver HDTV.
Don't get me wrong, there are places in networks that really need 10 gigabits. There are just extremely few, and it's extremely expensive to be able to transmit/receive/process 10 gigabits even with a more advanced network than Ethernet.
Excuse me for a second while I burst into uncontrollable laughter. Designs do not stay static. Any time you add code to an SMP-capable operating system you have to continue to ensure that it is SMP-safe.
Yeah, well, that's the price of playing with the big boys. Put on your big-boy shorts and get with the program, or pack up and go home.
Really. If you're going to write "enterprise-level" applications, and you're not even willing to work in SMP, then you're just not going to cut it.
But, here's what my real point was: If you're going to write an SMP app, it doesn't matter whether it runs on 1 processer, 2 processers, or 128 processers: You did the coding, and it runs. Charging people more to use extra processers on the basis that it costs you more in development is an outright lie.
This is only true for programmers, and really only for those who can/want to work on the application or OS that's broken. For people who *use* the server, they would much rather have the company working their ass off trying to fix it.
No, people who use the server just want it fixed. They don't care who does it, they just want it working now.
The beauty of commercial software is that the company can provide focused support, subsidized by the money made by selling the software, in a quantity and quality greater than even a dedicated support company can provide for OSS.
That's a great theory, but until the company becomes very outstanding, and charges an obscene amount of money, it doesn't happen that way. I've gone through support routes with quite a number of commercial companies, and never found them to equal that which I can get from OSS.
We've installed, say, SAP. The company relies on it. It's software that using 5000 databases, a 14 million line code base and for some reason everyone has been given a random number of extra years on their hire date. Fix it.
That's not a fair comparison, you're using commercial software. I don't have the source, and I don't have 15-minute access to the developpers who wrote the thing. Too bad.
Oh, United's reservation system is wrong. It seems the "inconvience the passenger flag" is set for EVERYONE. Can you look into that? By the way, if it's not fixed by tomorrow, you and you're team are fired (and you just transferred to indianapolis and have no money to move back home).
By tomorrow? Child's play. In my work, it's more like a couple of hours.
Now I'm on an unusual side of the table for me. I'm furious that building PHP with LDAP and SNMP and MySQL and gettext
But finding a large support organization that will take my money and bust their butts when our servers crash running a huge mission critical (it dies, we go out of business, 25,000 people become unemployed, a town becomes a wasteland) database and web back end system is pretty easy to get with the big players. I've caused Sun patches.
Been there. The difference between Sun and OSS is that when I dealt with Sun on SEVERE problems running Sun Web Server on Solaris on Sparc, their developpers kept giving us patches for months, without ever solving the problem. We finally gave up on Sun. When I've had a problem with PostgreSQL, on the other hand, I've had a working patch from the developpers within a couple of hours.
But it's naive to preach that all problems can be solved by replacing your systems with OSS systems
Of course it is. But the areas where OSS is a viable replacement are growing all of the time. Think back 8 years - would you have even considered Linux on a 32-CPU machine? Of course not. You wouldn't have even thought about using it on a 4-way machine. Now, IBM does it regularly.
The number of CPUs that is supported is just one of a wide myriad of areas where Linux is reaching and/or surpassing commercial implementations - and Linux is just one of a number of OSS apps that are doing the same thing.
Slashdot Mirror
There's a lot more to it than just how fast of a chip they can produce. Namely, how can they make the most money?
Let's say that AMD could produce an Athlon64 4500+ right now. Would it be in their best interest to release it? Not really. The fastest chips tend to be the lowest yield - and it would greatly push down the cost of the lesser chips. Their best interest is to release a chip that's juuuuust fast enough to keep up with or beat Intel, and keep the pricing high enough to come in just a bit under intel's pricing for competitive models.
In fact, each time Intel has actually released a faster chip lately, AMD has released a faster one as if it were no trouble at all. The way they've been doing, I wouldn't be surprised if they could release faster chips if there were an economic incentive to do so.
steve
One more thing that you've missed is that the variant that does appear very strongly to be transmitted through the consumption of cows is not the classical CJD, but a new variant called (uninterestingly) vCJD.
The new variant occurs at a MUCH higher rate than normal CJD, and strikes much younger people. The median age for CJD is around 68, the median age for vCJD is around 28.
Furthermore, of all diagnosed vCJD to date, virtually all of had multi-year expose in Britain during the peak of the BSE epidemic.
There has never been a single vCJD case in a country where cows did not have BSE.
It's not like there's one single prion disease. There are a lot of them. And what's more, even within one prion disease, there are many different variants. Within scrapie alone, there are at least 15 different variations.
vCJD is a little worrisome in that it does appear (very strongly!) to jump from cattle to humans. On the other hand, it appears to require direct consumption of the infected animal. Stop eating beef, problem completely solved.
steve
A lot of people have heard of "Mad Cow". Some of them have even heard of BSE or CFD. And most people don't realize that this is nothing novel, nothing new, and not at all limitted to cows.
The result of these prions in the brain is spongiform encephaly - literally, holes being eaten in your brain by the prion's interaction. Not a very fun thing!
Now, prion-caused sponfigorm encephalies have been found in a good number of animals. At a minimum, humans, goats, sheep, cows, squirrels, deer, elk, etc..
In cows, the condition is called "BSE" ("Bovine Spongiform Encephaly"). In humans, it's usually called Creutzfelt-Jacob's Disease (I'm sure I murdered the spelling). Those are merely terms for the resultant condition from the prion infection.
Now, the prion responsible for BSE isn't all that bad, as far as infectious prions go. First, it's not really transmissible in cows without the direct ingestion of infected nervous tissue. That means that if we just didn't feed cows ground up cows or ground up sheep, a very large part of the problem is solved.
However, there are other prion agents that are a bit nastier. In the case of CWD and scrapie, the prions have been shown to be transmissible to other individuals through the environment if (a) a n infected carcass or (b) excreta from an infected animal is in the area. Even better, even after all of the animals have left the area, CWD and scrapie agents have been shown to remain and still be contagious to other individuals years later.
Here's the good part: Researchers have already found genes that cause resistance to prion infections, or at least to certain types of them. The genes are found most commonly (and most heavily) in populations that practiced (or still practice) cannibalism. On the down side, it's not something as nice as getting infected and developping an immunity - we're talking about the cannibalistic societies being mostly wiped out by prion-based diseases, leaving only those (luckily) able to resist as the sole survivors, to pass along the genes to their offspring.
steve
(1) There has not been a single proven case of your skull being smashed by me hitting it with a mallet.
Just the same, we're reasonably sure that a smashed skull would result.
There are enough cases of spongiform encephalies being transmitted between species through ingestion of infected nervous tissue that your argument is a non-starter. In fact, if you look into the spread of prion-caused spongiform encephalies, you'll find that there are variants that don't require consumption of the infected animal at all.
One case in point being CWD and scrapie. Merely the presence of excrement from infected animals has been shown sufficient to infect other animals. To make it better, the prion agents remain in the environment (and infections) even after two years without any animals in the region at all.
steve
Well, your line of thinking isn't all that unsound. But here's what we do know:
1. Prions do exist, and do infect the nervous tissue of cows (and a lot of other animals, actually.)
2. Prions are rarely (if ever) able to be found outside of the nervous tissues.
3. Consuming infected nervous tissue does appear (quite strongly, in fact) to infect the consumer.
So, our classical line of thinking about digestion and protein absorbtion doesn't quite fit what we're seeing. What does that mean, class? That's right, we'll have to keep researching both what's really going on with the digestion (or lack thereof) of prions, and the absorbtion/transmission.
Note that while BSE doesn't appeaar to be transmissible in cattle without the cannibalistic ingestion that results from human intervention, there are other related prion diseases such as CWD and scrapie that do appear to be transmitted through environmental factors. In fact, in scrapie epidemics, it has been shown that scrapie agents may remain in the local environment for years after the outbreak.
steve
Watts = Amperage X Voltage
So, your hard drive uses 5v * 0.3A = 1.5 watts on the 5V line, and 12V * 0.5A = 6 watts on the +12V line. Total wattage: 7.5 watts. And those are *maximum* figures.
As you can see, most of the power is drawn on the 12V line. I've seen cheap 400-watt power supplies that could supply 13 amps on the +12V line, but I've seen quality 300-watt power supplies that could supply 22(!) amps on the +12V line.
There are various limitations on power supplies: First, you don't want to draw more wattage overall than it's rated for, you're asking to start hearing "BANG". Second, you don't want to draw more current on one voltage line than it's rated for. Last, they often have combined wattage limits, such as "Max 200 watts between +12V and +5V lines".
Despite that, it's not all that tough to size a power supply: It takes a very serious machine to overload a reasonable power supply. Even with a relatively cheap 350- or 400-watt power supply, you're probably not going to overload it unless you've either got (a) dual processers, and a good assortment of other accessories, or (b) a VERY high-draw video card and a VERY high-draw processer.
steve
The funny part is that the hardware RAID cards are no faster (and often slower!) than Linux's software-based RAID.
steve
Look at the drives. On the label, it will tell you how much current it draws on the +12V and +5V lines.
Now look at the power supply. On the label, it will tell you how much current it can provide on the +12V and +5V lines.
Of course, the hard drives aren't the only devices using the system: You'll have to look at other devices to get a better idea.
Here's an example: I just built a machine with 8x300gb drives, with dual 1.266GHz P3's. The el-cheapo Codegen 400-watt power supply that came with the case is more than enough to power everything, and it doesn't push nearly as many amps on the +12V line (where most of the HDD wattage goes) as even my 325-watt Enermax power supply.
steve
I just built a 2+ terabyte machine, for a lot less money. It doesn't have to be that expensive. Here's how to do it for a lot less:
1. Throw out the idea of the 3ware card. I've used them. They're overpriced pieces of crap. Responsiveness under load is horrible. Linux's software RAID works much better. Savings: $770
2. Use 300-gig drives. Fewer of them means a much cheaper casew ($124 in my case). Savings: $200
3. Two 550-watt power supplies? Stop believing the crap in the overclocking forums. The relatively cheap 400-watt power supply that came with the case easily powers eight 300-gig drives and a dual processer motherboard. Savings: $204
(Out of a good number of single- and multi-CPU systems, I have exactly two that couldn't run on a 400-watt power supply: And they both have four processers in them.)
Total savings: >$1,000
If you're thinking of using Windows on this, well, maybe you need the hardware RAID controller. Under Linux, the software RAID implementation is second to none - and I can guarantee that your host CPU will crank out parity in amounts that one of 3ware's cards can't even dream of. Combined with Samba, nothing on your network will know that it's not talking to a Windows machine.
steve
I bought one of these, eight 300-gig drives, and some of these to mount the drives in the 5.25" bays. Add in a couple of Highpoint IDE controllers, some cheap power splitters, and you're set.
Eight drives in RAID 5 gives me two terabytes of storage, mounting the drives in the 5.25" bays gives room for airflow, and the fans on the mounting brackets keep them somewhat cool for increased longevity. If memory serves, I could still shove in at least one extra drive.
steve
Yes, but the idea was an "ultimate" setup, not a real-world, cost-constrained setup.
steve
Now the ultimate machine would have SGIs architecute (memory) and #CPUs per node using the PowerPC CPU
If we're going for ultimate, I'd rather use the Power line of chips than the PowerPC. The PowerPC is a good chip, but the Power line is an entirely new level. : )
steve
Linux has, for quite some time, had a lot of effort put into support for machines other than x86 - which implies 64-bitness, more processers, NUMA, and all of those goodies.
Windows, on the other hand, has (with a few exceptions, now gone) been tied to the x86 architecture - meaning there's never even been a need to support more than a relatively small number of processers, and not even the possibility (until recently) of 64-bitness. (Yes, there was NT for the Alpha. It's gone.)
With Linux, there are people saying "I'd like this to not only run on a 32-CPU NUMA system, I'd like it to run well. With Windows, you have people saying "Ah, we'll never have to worry about more than four or maybe eight processers."
In particular to this machine, one of the large benefits of Linux is that it has pretty good support for NUMA (Non-Uniform Memory Architecture). IBM, who has a lot of experience in the area, has put a very large amount of development into it. That's paid off extremely well not only in supporting IBM's Power line, but also Opterons and other NUMAs.
With Windows, until the Opterons, NUMA was one of those things that Billy G. never even imagined supporting, because it just wasn't part of the x86 world.
As another example, the o(1) scheduler that was introduced along the 2.4 development vastly improved multi-processer scalability. Windows just doesn't have such cool things, and it certainly doesn't have the ability to switch to a new scheduler whenever they want!
steve
But on a dollar-for-dollar basis, the integer work sucked. I had a Compaq rep trying to sell me on Alpha for database work. He loaned me a $25,000 Alpha for a week. Our $14,000 Xeon beat it very handily.
steve
In general businesses try to "segment" their market, and when you vertically own the market, it is in your best interest to "make" everything, including different socket types
The problem has been that in the high-dollar area (Xeons), they've been sticking the consumer with slower busses than the desktop line. Until recently, when there was no other option in the price range, people just sucked it up and dealt with it. Now that there is good competition, that's going to have to change. This is probably a way for them to get some fast busses into their high-dollar (Xeon and Itanium) lines without it looking like they could have done it all along.
steve
The Itanium sucks for a lot of things, but when it comes to floating-point work, they're not half-bad - in fact, they are the tops of the currently published SpecFP scores.
Think of them like an Alpha: Very expensive, very fast at FP work, not so great at Integer work.
steve
It's not admitting defeat, it's getting rid of something that hasn't made much sense.
The Xeon line has traditionally been both the more lucrative and the more "stable" line, on the idea that businesses wanted long-term stability. When you bought a Xeon motherboard, you knew that it would be compatible with some of the upcoming Xeons in FSB and slot, and with a VRM module, even for different voltages.
A very large side-effect is that the Xeon line was tied to slower FSB/memory rates, and a lot more expensive. In reality, most people don't upgrade their Xeon chips - they move to a new architecture (P2->P3->P4, etc.)
Looking at the P2/P3 Xeons, it was long after the shift to a 133MHz FSB when the Xeons played catch-up. And it's just now that the P4 Xeons are getting away from the 533MHz FSB.
So, they're taking their most expensive, most lucrative processers, and hindering the performance. While it worked while Intel was the 800-pound gorilla (and you had no other choice in 2-way, let alone 4-way machines), that's no longer the case. There's competition in the market - good competition. It was a real kick in the pants to shell out $5000 to $15,000 on Xeon systems that had a slower FSB than the $500 desktop counterparts!
So, now, they're going to have to do the sensible thing: Start giving actual performance in exchange for customer money. Rather than have the "high-end bus" and the "low-end" or "mid-range" bus, it makes a lot of sense to just have all of them use one bus design - theoretically, a fast bus.
The Athlon bus was designed so that you could, in theory, plug an Alpha into an Athlon board
I really don't think that was the idea. AMD was looking for a front-side bus design, and Intel's wasn't an option. The Alpha design was available for the right price, and so it was used.
Do you believe that x86-32 BIOS code is going to work to get an Itanium CPU
I believe that Intel has already stated that they want to make some radical changes to the way the BIOS works.
Sorry, Intel; I'd say that this is the beginning of the end.
Unless they're able to really shift direction, I'd agree.
steve
I don't think that AMD will ever beat out Intel completely, but if they keep making the right moves, they will be able to force Intel to radically change how it does business.
Intel has traditionally enjoyed very, very healthy profit margins. AMD, on the other hand, has traditionally lost money - it's only been in the last quarter or two that they've even posted a profit. To bring a company from the times of the K6 and K6-2 to having an incredible offering like the Opteron, while having to be so cautious with money, shows that AMD can be resourceful, and compete even when it's an uphill battle.
Intel, on the other hand, has enough infrastructure, capital, partners, investment, and diversification that it will never be beat out entirely. However, as AMD starts to make more and more profits (by eating into Intel's most lucrative market!), they'll (theoretically) be able to step up the competition and give Intel an even stronger opponent. If they do that, Intel will have to start being much more frugal and resourceful.
Up until a few years ago, the AMD-Intel battle was pretty one-sided. Now, things are going to be a lot more fun to watch.
steve
With Opterons coming in much less expensive than the Itaniums, moving to a common socket with the Xeons isn't going to help much.
On the other hand, with Opterons offering far better scalability and performance than Xeons, moving to a common socket with Itaniums isn't going to help out much.
Even on 2-way machines, Opterons show much better scalability than Xeons. As the number of CPUs increase, the Opteron architecture (when coupled with a supporting OS) allows it to shine more and more. With 8-way Opterons coming out fairly soon, Intel needs to come up with something fast - they're losing one of their most lucrative markets to AMD.
steve
I was in a building just a couple of hours ago with two OC192s. That totals up to 20 gigabits per second.
The two OC192s easily handle all of the company's voice and data communication in the city, which is not at all a small amount.
Very few people realize just how much data 10 gigabits per second really is, and fewer still realize how much hardware it takes to be able to send/receive/route/process data at that rate without even looking at the network card.
steve
storage area networks. at 10Gb/s, SAN's are not a slow data storage medium, but an ultra fast storage medium, that may ever be faster that your local hard drive.
You still think you're going to get 10 gigabits to/from that thing? Think again. That's over a gigabyte per second. It would take at least twenty disks in RAID 0 just to barely get up to that speed. However, putting twenty disks in a RAID 0 array means that your realistic MTBKYBG (Mean Time Between Kissing Your Butt Goodbye) would be something like a month.
Want 1+0? Well, then you better have at least FORTY disks in the SAN. Want RAID 5? Well... tough. It's going to be quite a while before you find any RAID 5 solution that's going to write at anything near a gigabyte per second. In fact, even READING at a gigabyte per second is going to be very expensive with RAID 5.
Sure, there are SANs that can dish out over one gigabit. But hitting ten gigabits is going to be tough, and not just because Ethernet has so many limitations.
OR, how about full diskless clients, with BIOS manufactures being able to map specific locations on the network to a local harddrive
Sorry. Virtually no desktop-level computers can even realistically take full advantage of even a one gigabit card at the moment. Slapping a bigger card in the machine isn't going to overcome architectural limitations.
CLUSTERING software. imagin a beow...., how about openmosix!
What? Serious clusters don't even use gigabit ethernet, it's too much of a hacked-together kludge. Bumping it up another order of magnitude will probably make things worse, rather than better.
EAEP scanners and video cameras = awesome. 1 640x480 camera compressing plain mpeg1 video can saturate a 100Mbps network pretty quickly
So, use MPEG2, or an even better CODEC. It'll be cheaper to implement that than to implement something that could handle a 10gigE network.
HDTV delivery, speeds like this would allow High Definition media delivery without bandwidth pains
Amazingly, there's already plenty of bandwidth to deliver HDTV.
Don't get me wrong, there are places in networks that really need 10 gigabits. There are just extremely few, and it's extremely expensive to be able to transmit/receive/process 10 gigabits even with a more advanced network than Ethernet.
steve
you insensitive clo...
oh, wait. I guess I'm the insensitive clod.
steve
Excuse me for a second while I burst into uncontrollable laughter. Designs do not stay static. Any time you add code to an SMP-capable operating system you have to continue to ensure that it is SMP-safe.
Yeah, well, that's the price of playing with the big boys. Put on your big-boy shorts and get with the program, or pack up and go home.
Really. If you're going to write "enterprise-level" applications, and you're not even willing to work in SMP, then you're just not going to cut it.
But, here's what my real point was: If you're going to write an SMP app, it doesn't matter whether it runs on 1 processer, 2 processers, or 128 processers: You did the coding, and it runs. Charging people more to use extra processers on the basis that it costs you more in development is an outright lie.
steve
This is only true for programmers, and really only for those who can/want to work on the application or OS that's broken. For people who *use* the server, they would much rather have the company working their ass off trying to fix it.
No, people who use the server just want it fixed. They don't care who does it, they just want it working now.
The beauty of commercial software is that the company can provide focused support, subsidized by the money made by selling the software, in a quantity and quality greater than even a dedicated support company can provide for OSS.
That's a great theory, but until the company becomes very outstanding, and charges an obscene amount of money, it doesn't happen that way. I've gone through support routes with quite a number of commercial companies, and never found them to equal that which I can get from OSS.
steve
We've installed, say, SAP. The company relies on it. It's software that using 5000 databases, a 14 million line code base and for some reason everyone has been given a random number of extra years on their hire date.
Fix it.
That's not a fair comparison, you're using commercial software. I don't have the source, and I don't have 15-minute access to the developpers who wrote the thing. Too bad.
Oh, United's reservation system is wrong. It seems the "inconvience the passenger flag" is set for EVERYONE. Can you look into that? By the way, if it's not fixed by tomorrow, you and you're team are fired (and you just transferred to indianapolis and have no money to move back home).
By tomorrow? Child's play. In my work, it's more like a couple of hours.
Now I'm on an unusual side of the table for me. I'm furious that building PHP with LDAP and SNMP and MySQL and gettext
But finding a large support organization that will take my money and bust their butts when our servers crash running a huge mission critical (it dies, we go out of business, 25,000 people become unemployed, a town becomes a wasteland) database and web back end system is pretty easy to get with the big players. I've caused Sun patches.
Been there. The difference between Sun and OSS is that when I dealt with Sun on SEVERE problems running Sun Web Server on Solaris on Sparc, their developpers kept giving us patches for months, without ever solving the problem. We finally gave up on Sun. When I've had a problem with PostgreSQL, on the other hand, I've had a working patch from the developpers within a couple of hours.
But it's naive to preach that all problems can be solved by replacing your systems with OSS systems
Of course it is. But the areas where OSS is a viable replacement are growing all of the time. Think back 8 years - would you have even considered Linux on a 32-CPU machine? Of course not. You wouldn't have even thought about using it on a 4-way machine. Now, IBM does it regularly.
The number of CPUs that is supported is just one of a wide myriad of areas where Linux is reaching and/or surpassing commercial implementations - and Linux is just one of a number of OSS apps that are doing the same thing.
steve