Very true, at 733MHz, the Transmeta chips they are using are NO WHERE CLOSE to being as fast as a PIII or Athlon of the same clock speed. In reality, the performance of these chips have a tough time matching 400MHz Celeron processors, let alone anything that has been sold by Intel or AMD in the last 3 years.
For a thin client though, this might be enough computing power. A thin client really doesn't do a heck of a lot other than display simple graphics. Power consumption (and therefore heat produced) is also quite low, though a ULV mobile Celeron would offer comperable power consumption. The real reason why HP went with these chips is because they are cheap. I have to wonder why they didn't go for a VIA C3 instead though. Similar power consumption, low cost and much more widely available/better supported chipsets.
This sort of stat, of course, is absolutely impossible to determine. Just take a look at your firewall logs sometime (you DO have a firewall, RIGHT?!). My Linux firewall is still getting hit by regular SQL Slammer attacks, MSBlaster attacks, etc., as well as a number of odd ICMP packets, SYN packets and a variety of other garbage. I also get tons of Windows machines trying to share their files and printers with me (or connect to shares on my end).
I'd guess that the vast majority of publicly accessible IP addresses have SOME kind of probe/attack (almost all of which are automated) at least a dozen times a day, probably more.
Simple fact of the matter is that you WILL get hacked if you put a machine on the internet without security. Whether it takes 10 attempts or 10,000 doesn't matter all that much, the end result is the same.
Most Windows worms do not deface websites, and as such were not counted in these results from my understanding of their testing methodology (note: I couldn't read the original mi2g article as it costs a shitload to buy, only the Globe and Mail article and other posters here).
Bullshit or otherwise, Linux servers ARE being hacked, and frequently. People should take this as a clear leason that it doesn't matter what operating system you are running, you MUST apply patches and stay on top of security if you want a secure system.
Given the open-source nature of the operating system, I would guess that it is EASIER to hack an unpatched Linux system than an unpatched Windows system, since finding the vulnerability should be a lot easier once it's been announced.
National Post is a "bit downscale"?! Now that is an understatement if I've ever heard one! The National Post is only one step above the Sun, a paper known primarily for the scantly clad "Sunshine Girl" shown on Page 3.
The Globe and Mail is a very respectible paper. They do their best to report the news accurately and with a fair degree of journalistic integrity.
I don't live in the States either, but if I'm posting prices, I post in either US dollars or Euro. Why? Because those are two currencies that MOST people in the world are fairly familiar with.
The simple fact of the matter is that there are only three hard currencies in the world today, the US dollar, the Euro and the Yen. My Canadian dollars are just fine here in Canada, but everywhere else they're pretty useless. Same with the british pound, it's only good for Britain (hell, sometimes you can't even use Scotish-British pounds in England or Northern Ireland-British pounds anywhere aside from Northern Ireland!)
Fortunately it seems that the British parliment actually does appear to be considering taking their blinders off and switching to the Euro (something they SHOULD have done two years ago if they had been smart).
Re:Not an Athlon64, but an Opteron
on
AMD64 Preview
·
· Score: 1
The reason why Opteorn boards with AGP and PCI-X slots are few and far between is because of the chipset used. AMD's 8000 Opteron chipset has three chips, an AGP tunnel, a PCI-X tunnel and a legacy I/O hub (AMD uses the term "hypertransport tunnel" as opposed to the old "PCI bridge" terminology of north/southbridge days or Intel's "Accelerated Hub Architecture").
So, if you want to use AGP, you just use the AGP tunnel. If you want PCI-X, you just use the PCI-X tunnel. If you want BOTH, than you need to use both chips, which is a little tricky since they can't be chained together (the legacy I/O hub can be daisy-chained to the far side of either, but it uses a narrower HT connection). Instead you have to either tie them to a second Hypertransport link on your processor (Opteron's all come with 3 hypertransport links) or you put the AGP tunnel on one processor and the PCI-X tunnel on the other processor in a dual-processor machine. End result is a rather tricky design, not only from a BIOS point of view but also simply getting all the traces to go to the right place on the board.
Fortunately boards DO exist that make use of both chips. Tyan's Thunder K8W is the only one I've seen that is (apparently) shipping, but I've seen pictures of an MSI board as well.
Unfortunately this will NOT be possible with AMD's Athlon64 chips (either the Athlon64 or the Athlon64 FX), since these chips have only a single hypertransport link enabled. If you want AGP and PCI-X on an Athlon64 system, you'll have to wait until another chipset comes out to support said features.
Ohh, a quick note about the AthlonXP in an Athlon64 socket rumors, my guess is that if such a chip exists, it will be a fully-functional Athlon64 with 64-bit mode disabled in the BIOS.
Re:Opteron Benchmarks, not Athlon 64
on
AMD64 Preview
·
· Score: 1
All Athlon64 and Opteron machines will support it. This is a GOOD THING IMO. Fortunately Intel's newest chipsets also support ECC, though it's up to the motherboard manfuacturer to not fuck things up and break said support. (so, for example, Asus would probably make this work, but companies like Abit probably would not, much like how Abit's old 430HX motherboards never worked with ECC memory). In the case of the Athlon64 and Opteron, the memory controller is on-die so the motherboard manufacturer's would REALLY have to go out of their way to break this feature.
For some reason though people seem to consistantly confuse ECC memory and registered memory. The Opteron (and presumably the Athlon64 FX) required REGISTERED memory, not ECC memory. This is why the memory is slower and costs more. ECC is only a few dollars more expensive than non-ECC memory and has virtually no effect on performance (it adds a cycle here and there, but nothing significant at all). Registered memory is what costs twice as much as non-registered (aka unbuffered) memory and is a tiny bit slower.
You can easily get a pair of G5's into a 1U server. Dell, HP and the like all ship 1U servers with dual 3.0GHz XeonDP chips in them, and those chips consume quite a bit more power than the G5. The G5 uses roughly the same amount of power as a similarly clocked AMD Opteron, and almost all of the dual-processor Opteron systems out there are 1U racks.
The G5 uses 9 fans so that they could use relatively slow spinning and variable speed fans. By all accounts their systems are VERY quiet while idle, though not abnormally quiet once they spin up to full speed. It's a rather nice touch IMO, though I also went out of my way and spent a little bit of extra cash to make my AthlonXP system very quiet (ie I bought a big heatsink that cost $50 more than the small heatsinks with tornado-style fans that most users buy), and even then I ended up with a power supply fan that is a bit louder than my liking.
Others who know more than I do about Myrinet have already pointed out that you don't seem to know what you're talking about, so I'll skip that.
Next point, FP performance. Let's see, Apple managed a score of 840 on Spec CFP2000 base on their 2.0GHz PowerPC using GCC. AMD managed a score of 1209 on their 2.0GHz Opteron using Intel's compiler, while IBM managed a slightly lower score of 1172 using GCC. If the G5 is so superior, it sure isn't doing a very good job of showing it.
The memory bus on the G5 is better? What are you smoking here?! Each Opteron chip has a low-latency integrated memory controller capable of managing 5.4GB/s of bandwidth. Two G5 shares a higher latency, off-chip memory controller capable of 6.4GB/s total. Their bus may allow memory and cache transfers at the same time, the Opteron has a completely separate buses for these tasks! The Opteron also has twice the L1 data cache and twice the L2 cache that the PowerPC 970 has.
I will, however, agree with you that the Opteron is no cheaper than the G5 if you can get Apple to give you a useful configuration of the boxes for a supercomputer (who the hell needs a high-end gaming video card on each super-computer node? Let alone iTunes!). As for building the machines themselves though, why would they do that? Why not just buy a machine from one of the dozens of companies that are selling dual-processor 1U Opteron servers now. Many can and have built clusters of these already. Speaking of '1U' and all, brings me to one fo the dumbest things about this supercomputer, it's ALL being built usind the damn PowerMac tower cases! Now THAT is just a DUMB idea! 1100 giant desktop cases using bog-standard power supplies and the works, now that is just ridiculous!
That "traditional advantage" doesn't apply here because the power consumption of IBM's PowerPC 970 and the AMD Opteron are nearly identical.
Exact numbers are few and far between on both sides (neither company has a publicly available document that describes their exact power consumption), but most estimates put them as being real close. Both chips consume about 45-50W typical power consumption, and somewhere in the order of 50-60W maximum power consumption. One or the other might be a few watts cooler, but the difference is quite small.
Both of these chips consume a lot more power than the G4 and both consume a lot less power than the AthlonXP or the Pentium 4 (or the Itanium for that matter, which is the current champion of high power consumption).
No, the G5 does NOT have Hypertransport, at least assumign that you are using Apple nomenclature of calling the IBM PowerPC 970 CPU the "G5" and the machine itself a "PowerMac".
The PowerMac G5 machines use Hypertransport to connect their motherboard chipsets together. This is nothing new, the XBox does this as well. Hypertransport is a very good, low-cost, high performance solution for connection chips together directly. However, connecting motherboard chipsets together doesn't do much of anything for getting data to/from the processor itself.
IMO the IBM PowerPC 970 (aka the G5) is actually not a bad chip to power a supercomputer. It seems to have quite a lot of processing power for a reasonable price and decent thermal characteristics (when you have a LOT of processors in small space, low power consumption is a very good thing, this is where most of Cray's real innovation was). However, the PowerMac's seem to be rather weak for this sort of application because of Apple's chipset, which is really a desktop/workstation chipset. The I/O bandwidth isn't all that impressive (1.6GB/s in each direction is the absolute best that can be managed) and latency should be even worse. Getting to a PCI-X card involves going through two separate chips and three buses. Compared to some of the proposed Opteron solutions which will hang networking hardware right off the processors Hypertransport buses (no chips in the middle and only a single bus with 3.2GB/s of bandwidth in each direction), the chip I/O looks pretty weak.
My guess is that this deal is mainly due to the fact that Apple gave them a real good deal. Given the rather high cost of the networking equipment that will be needed, VT is probably getting their 1100 servers for next to nothing.
Why not save yourself 2.7 million dollars and get the same computing power while only running the tests ONCE!
If they're spending that much money to buy that much equipment, you better believe that they expect to USE it! Running tests twice might be fine for you or me on our home computers which are going to be sitting idol 99% of the time anyway, but duplicating everything you do every time when your building a supercomputer is just wasting money, and a lot of it.
Do you run a simulation once? Of course not! But what's the point in running the simulation twice with the exact same input data and/or random seed? That doesn't help prove anything except that it works/doesn't work for one very specific case.
For the SPEC scores, the operating system plays virtually no role at all, however the compiler plays a major role. Here is where Apple really loses out, because the Mac compilers are no where near as good as compilers for x86, so even if Apple had a chip that was as fast it would not run applications as fast.
Of course, the data really isn't that close if you look at REAL results as opposed to what Apple paid some company to find. The scores that Apple showed for the x86 systems are WAY lower than what anyone else would get for an x86 system. The G5 is a fine chip, but it's only marginally faster than a P4 clock for clock. A 2.0GHz G5 would give you comperable performance to a 2.4GHz P4 with an 800MHz bus speed or possibly a slightly higher clocked P4 on a slower bus.
As for dual-processors, part of the problem is that Intel's implementation for SMP isn't so hot. The dual-Xeon systems use a shared multi-drop bus, which isn't a particularly good design these days. AMD's Opteron chips tend to scale MUCH better to multiple processors than Intel's Xeons.
FWIW a two-processor Opteron 246 (2.0GHz) system manages (base) SPEC CINT_rate scores of 28.8 and CFP_rate of 28.1. This makes the 2.0GHz PowerMacs 17.2 and 15.7 scores seem a touch low.
This is only marginally useful from a purely academic standpoint. In the real world, there are only two measures that are important: Highest maximum performance regardless of cost and performance for a given price point. How you get that performance, be it by higher clock speed or more work per clock is rather meaningless.
It's strange to see some people seeing a 2.0GHz G5 performing comperable to a 3.0GHz P4 (at least when Apple pays someone to do the benchmarks) and somehow assume that this means anything, or worse yet that somehow IBM/Apple are magically going to come out with a 3.0GHz PowerPC 970. This chip will NEVER make it to 3.0GHz on a 130nm fab process. It MIGHT manage to clock that high on a 90nm fab process, but I wouldn't even hold my breath for that if I were you. The P4 was designed to be able to clock high, which is why it's clocked at 3.2GHz on the same sort of fab process that has every other chip maxing out in the 1.5-2.5GHz range (I think AMD has the next fastest 130nm main processor at ~2.2GHz).
The numbers Apple got as STILL low for GCC, at least for the interger side of things (I haven't seen many numbers for CFP using GCC). Generally speaking, GCC is only about 5-10% slower than Intel's C compiler for x86, Apple's numbers are 20-25% lower.
Basically these benchmarks are a load of marketing crap. The IBM PowerPC 970/G5 is a perfectly fine chip, but it's definitely NOT the fastest PC processor in the world, not by a long shot. This is just another case of the "Steve Jobs Reality Distortion Field" in action.
It was for Quake3, and a 20-25% speedup was the goal, but the actual result was MUCH less, to the extent that it usually ran slower on dual-processor machines.
The simple fact of the matter is that in games, there is usually just ONE thread that is doing heavy processing at any given time, and it's extremely difficult to get around that. Sure, you can offload things like sound and the odd other thing to a secondary processor, but usually one chip ends up working full out 100% of the time with the other chip being idle 99% of the time. The nature of the work on most games just doesn't parallize all that well, besides which there is such a TINY market for dual-processor gaming machines that no one really bothers for it much.
If you want a gaming machine, you get a fast single processor machine with an even faster graphics card.
The floating point performance of the G5 is nothing to write home about. They only managed a score of 840 in SpecFP 2000, which is WAY bellow the performance of current x86 chips (as benchmarked by anyone OTHER than Apple/Apple paid testers).
As you can see, Apple's score of 840 in this benchmark isn't exactly impressive. Even one of the new 2.4C GHz P4 chips (800MHz bus speed) will score higher.
Now, I know that these scores are using a compiler that is more heavily optimized than that used for the Mac, but guess what, there ARE compilers that are more heavily optimized out there for x86 than those for the Mac. MUCH more heavily optimized. Does this make the comparison somehow unfair? Not at all, because the end result is the same, applications run a lot faster on the newest x86 chips than on Macs.
Of course, system performance has never really been the reason why people buy Macs vs. a PC.
Those sorts of upgrade cards exist for PCs too, but no one uses the for a very good reason, because they SUCK! Sure you might be upgrading the processor, but you end up severely limiting the available bandwidth of the chip to even less than what was available on the previous system usually. A G4 with less bandwidth than a G3 is hardly a worthwhile upgrade for anyone, just as the old PIII upgrades for Socket 7 boards were mostly useless.
These upgrades are fine for people who absolutely can not replace their motherboard because they have some super-expensive proprietary hardware in there, but otherwise they usually cost about the same as buying a new processor, motherboard and memory while offering FAR less performance.
Ohh this is just too perfect. To quote Manfred from the above linked messages:
"P.S.: I'd propose that the GPL rule that a change must be tagged with the name of the \ person who changes (or adds) a file is enforced - atealloc.c is only tagged with \ "SGI", thus I don't know who should be shot for writing that."
Looks like SCO is the one he should be shooting!:>
In my experience, most reboots aren't necessary, they're just the result of patch programmers being lazy. Microsoft only just recently stopped being lazy in this regard, and guess what? Their patches no longer require reboots, they just stop and restart the right services, run a few ocmmands and your set to go. Norton hasn't figured this out yet.
One thing I've found, for most Norton updates that "require" a reboot, you can just cancel their reboot and restart the Norton Anti-Virus service, and most of the time everything will work with all the latest and greatest updates. The odd time you need to register some component, and then it's usually just easier to reboot rather than trying to figure out all the undocumented commands.
Your comparing apples and orangutans here when it comes to power consumption. Intel's Thermal Design Power (which the maximum power consumption excluding thermal viruses) for the P4-M 2.0GHz is 32W. Transmeta's Thermal Design Power for the 1.0GHz Crusoe 5800 is 7.5W. The "1W" number that Transmeta likes to flaunt around is the amount of power that the chip consumes while it's sitting around twiddling it's thumbs. For comparison, a P4-M will consume about 5.0W doing the same thing.
That being said, the real competitor for the Transmeta chips is not the P4-M, but rather the ULV Mobile Celeron chips. Both chips will cost close to the same amount and consume close to the same amount of power (TDP for a ULV Celeron is 7.0W). Previously these Celerons, even at lower clock speeds, have had no trouble beating Transmeta chips. It remains to be seen if this new Transmeta chip has improved it's performance enough to compete with the Celerons. Otherwise it's just an odd-ball design with the same power consumption and similar price to a more widely available and better supported processor.
I don't know about that. They are promising only 50-80% improvements over their old Crusoe 5800 processor. That would put them as being about on-par with the chips that Intel had out two years ago when the Crusoe 5800 was first available, but it'll have a hell of a time competing with the chips that Intel is producing now, let alone 6+ months from now when this new Transmeta processor actually starts shipping.
I think that the real question will be how well this chip can compete with Intel's Ultra Low Voltage (ULV) Mobile Celeron line of processors. The two chips will have comperable power consumption (5-10W max, typical of under 5W) and probably won't be too far off one another in terms of price. Previous Transmeta chips have had a heck of a time keeping up with even the slowest mobile Celeron chips that Intel had available (read: they kind of kept up in MS Office, but got pretty well thrashed for everything else), but maybe this newer chip will bring performance up a bit.
Let's see, we'll replace the current ARM or MIPS chips, which consume about 500mW or less, with a Transmeta chip that consumes 5-10W, for what exactly? Transmeta may have decently low power consumption for an x86 chip, but it's off by an order of magnitude as compared to the chips used in most handhelds.
Slashdot Mirror
Very true, at 733MHz, the Transmeta chips they are using are NO WHERE CLOSE to being as fast as a PIII or Athlon of the same clock speed. In reality, the performance of these chips have a tough time matching 400MHz Celeron processors, let alone anything that has been sold by Intel or AMD in the last 3 years.
For a thin client though, this might be enough computing power. A thin client really doesn't do a heck of a lot other than display simple graphics. Power consumption (and therefore heat produced) is also quite low, though a ULV mobile Celeron would offer comperable power consumption. The real reason why HP went with these chips is because they are cheap. I have to wonder why they didn't go for a VIA C3 instead though. Similar power consumption, low cost and much more widely available/better supported chipsets.
This sort of stat, of course, is absolutely impossible to determine. Just take a look at your firewall logs sometime (you DO have a firewall, RIGHT?!). My Linux firewall is still getting hit by regular SQL Slammer attacks, MSBlaster attacks, etc., as well as a number of odd ICMP packets, SYN packets and a variety of other garbage. I also get tons of Windows machines trying to share their files and printers with me (or connect to shares on my end).
I'd guess that the vast majority of publicly accessible IP addresses have SOME kind of probe/attack (almost all of which are automated) at least a dozen times a day, probably more.
Simple fact of the matter is that you WILL get hacked if you put a machine on the internet without security. Whether it takes 10 attempts or 10,000 doesn't matter all that much, the end result is the same.
Most Windows worms do not deface websites, and as such were not counted in these results from my understanding of their testing methodology (note: I couldn't read the original mi2g article as it costs a shitload to buy, only the Globe and Mail article and other posters here).
Bullshit or otherwise, Linux servers ARE being hacked, and frequently. People should take this as a clear leason that it doesn't matter what operating system you are running, you MUST apply patches and stay on top of security if you want a secure system.
Given the open-source nature of the operating system, I would guess that it is EASIER to hack an unpatched Linux system than an unpatched Windows system, since finding the vulnerability should be a lot easier once it's been announced.
National Post is a "bit downscale"?! Now that is an understatement if I've ever heard one! The National Post is only one step above the Sun, a paper known primarily for the scantly clad "Sunshine Girl" shown on Page 3.
The Globe and Mail is a very respectible paper. They do their best to report the news accurately and with a fair degree of journalistic integrity.
I don't live in the States either, but if I'm posting prices, I post in either US dollars or Euro. Why? Because those are two currencies that MOST people in the world are fairly familiar with.
The simple fact of the matter is that there are only three hard currencies in the world today, the US dollar, the Euro and the Yen. My Canadian dollars are just fine here in Canada, but everywhere else they're pretty useless. Same with the british pound, it's only good for Britain (hell, sometimes you can't even use Scotish-British pounds in England or Northern Ireland-British pounds anywhere aside from Northern Ireland!)
Fortunately it seems that the British parliment actually does appear to be considering taking their blinders off and switching to the Euro (something they SHOULD have done two years ago if they had been smart).
The reason why Opteorn boards with AGP and PCI-X slots are few and far between is because of the chipset used. AMD's 8000 Opteron chipset has three chips, an AGP tunnel, a PCI-X tunnel and a legacy I/O hub (AMD uses the term "hypertransport tunnel" as opposed to the old "PCI bridge" terminology of north/southbridge days or Intel's "Accelerated Hub Architecture").
So, if you want to use AGP, you just use the AGP tunnel. If you want PCI-X, you just use the PCI-X tunnel. If you want BOTH, than you need to use both chips, which is a little tricky since they can't be chained together (the legacy I/O hub can be daisy-chained to the far side of either, but it uses a narrower HT connection). Instead you have to either tie them to a second Hypertransport link on your processor (Opteron's all come with 3 hypertransport links) or you put the AGP tunnel on one processor and the PCI-X tunnel on the other processor in a dual-processor machine. End result is a rather tricky design, not only from a BIOS point of view but also simply getting all the traces to go to the right place on the board.
Fortunately boards DO exist that make use of both chips. Tyan's Thunder K8W is the only one I've seen that is (apparently) shipping, but I've seen pictures of an MSI board as well.
Unfortunately this will NOT be possible with AMD's Athlon64 chips (either the Athlon64 or the Athlon64 FX), since these chips have only a single hypertransport link enabled. If you want AGP and PCI-X on an Athlon64 system, you'll have to wait until another chipset comes out to support said features.
Ohh, a quick note about the AthlonXP in an Athlon64 socket rumors, my guess is that if such a chip exists, it will be a fully-functional Athlon64 with 64-bit mode disabled in the BIOS.
All Athlon64 and Opteron machines will support it. This is a GOOD THING IMO. Fortunately Intel's newest chipsets also support ECC, though it's up to the motherboard manfuacturer to not fuck things up and break said support. (so, for example, Asus would probably make this work, but companies like Abit probably would not, much like how Abit's old 430HX motherboards never worked with ECC memory). In the case of the Athlon64 and Opteron, the memory controller is on-die so the motherboard manufacturer's would REALLY have to go out of their way to break this feature.
For some reason though people seem to consistantly confuse ECC memory and registered memory. The Opteron (and presumably the Athlon64 FX) required REGISTERED memory, not ECC memory. This is why the memory is slower and costs more. ECC is only a few dollars more expensive than non-ECC memory and has virtually no effect on performance (it adds a cycle here and there, but nothing significant at all). Registered memory is what costs twice as much as non-registered (aka unbuffered) memory and is a tiny bit slower.
You can easily get a pair of G5's into a 1U server. Dell, HP and the like all ship 1U servers with dual 3.0GHz XeonDP chips in them, and those chips consume quite a bit more power than the G5. The G5 uses roughly the same amount of power as a similarly clocked AMD Opteron, and almost all of the dual-processor Opteron systems out there are 1U racks.
The G5 uses 9 fans so that they could use relatively slow spinning and variable speed fans. By all accounts their systems are VERY quiet while idle, though not abnormally quiet once they spin up to full speed. It's a rather nice touch IMO, though I also went out of my way and spent a little bit of extra cash to make my AthlonXP system very quiet (ie I bought a big heatsink that cost $50 more than the small heatsinks with tornado-style fans that most users buy), and even then I ended up with a power supply fan that is a bit louder than my liking.
Others who know more than I do about Myrinet have already pointed out that you don't seem to know what you're talking about, so I'll skip that.
Next point, FP performance. Let's see, Apple managed a score of 840 on Spec CFP2000 base on their 2.0GHz PowerPC using GCC. AMD managed a score of 1209 on their 2.0GHz Opteron using Intel's compiler, while IBM managed a slightly lower score of 1172 using GCC. If the G5 is so superior, it sure isn't doing a very good job of showing it.
The memory bus on the G5 is better? What are you smoking here?! Each Opteron chip has a low-latency integrated memory controller capable of managing 5.4GB/s of bandwidth. Two G5 shares a higher latency, off-chip memory controller capable of 6.4GB/s total. Their bus may allow memory and cache transfers at the same time, the Opteron has a completely separate buses for these tasks! The Opteron also has twice the L1 data cache and twice the L2 cache that the PowerPC 970 has.
I will, however, agree with you that the Opteron is no cheaper than the G5 if you can get Apple to give you a useful configuration of the boxes for a supercomputer (who the hell needs a high-end gaming video card on each super-computer node? Let alone iTunes!). As for building the machines themselves though, why would they do that? Why not just buy a machine from one of the dozens of companies that are selling dual-processor 1U Opteron servers now. Many can and have built clusters of these already. Speaking of '1U' and all, brings me to one fo the dumbest things about this supercomputer, it's ALL being built usind the damn PowerMac tower cases! Now THAT is just a DUMB idea! 1100 giant desktop cases using bog-standard power supplies and the works, now that is just ridiculous!
That "traditional advantage" doesn't apply here because the power consumption of IBM's PowerPC 970 and the AMD Opteron are nearly identical.
Exact numbers are few and far between on both sides (neither company has a publicly available document that describes their exact power consumption), but most estimates put them as being real close. Both chips consume about 45-50W typical power consumption, and somewhere in the order of 50-60W maximum power consumption. One or the other might be a few watts cooler, but the difference is quite small.
Both of these chips consume a lot more power than the G4 and both consume a lot less power than the AthlonXP or the Pentium 4 (or the Itanium for that matter, which is the current champion of high power consumption).
<heard from the bowels of Intel's headquarters>
Damn, Apple beat our Itanium shipments by 100 machines!
No, the G5 does NOT have Hypertransport, at least assumign that you are using Apple nomenclature of calling the IBM PowerPC 970 CPU the "G5" and the machine itself a "PowerMac".
The PowerMac G5 machines use Hypertransport to connect their motherboard chipsets together. This is nothing new, the XBox does this as well. Hypertransport is a very good, low-cost, high performance solution for connection chips together directly. However, connecting motherboard chipsets together doesn't do much of anything for getting data to/from the processor itself.
IMO the IBM PowerPC 970 (aka the G5) is actually not a bad chip to power a supercomputer. It seems to have quite a lot of processing power for a reasonable price and decent thermal characteristics (when you have a LOT of processors in small space, low power consumption is a very good thing, this is where most of Cray's real innovation was). However, the PowerMac's seem to be rather weak for this sort of application because of Apple's chipset, which is really a desktop/workstation chipset. The I/O bandwidth isn't all that impressive (1.6GB/s in each direction is the absolute best that can be managed) and latency should be even worse. Getting to a PCI-X card involves going through two separate chips and three buses. Compared to some of the proposed Opteron solutions which will hang networking hardware right off the processors Hypertransport buses (no chips in the middle and only a single bus with 3.2GB/s of bandwidth in each direction), the chip I/O looks pretty weak.
My guess is that this deal is mainly due to the fact that Apple gave them a real good deal. Given the rather high cost of the networking equipment that will be needed, VT is probably getting their 1100 servers for next to nothing.
Why not save yourself 2.7 million dollars and get the same computing power while only running the tests ONCE!
If they're spending that much money to buy that much equipment, you better believe that they expect to USE it! Running tests twice might be fine for you or me on our home computers which are going to be sitting idol 99% of the time anyway, but duplicating everything you do every time when your building a supercomputer is just wasting money, and a lot of it.
Do you run a simulation once? Of course not! But what's the point in running the simulation twice with the exact same input data and/or random seed? That doesn't help prove anything except that it works/doesn't work for one very specific case.
For the SPEC scores, the operating system plays virtually no role at all, however the compiler plays a major role. Here is where Apple really loses out, because the Mac compilers are no where near as good as compilers for x86, so even if Apple had a chip that was as fast it would not run applications as fast.
Of course, the data really isn't that close if you look at REAL results as opposed to what Apple paid some company to find. The scores that Apple showed for the x86 systems are WAY lower than what anyone else would get for an x86 system. The G5 is a fine chip, but it's only marginally faster than a P4 clock for clock. A 2.0GHz G5 would give you comperable performance to a 2.4GHz P4 with an 800MHz bus speed or possibly a slightly higher clocked P4 on a slower bus.
As for dual-processors, part of the problem is that Intel's implementation for SMP isn't so hot. The dual-Xeon systems use a shared multi-drop bus, which isn't a particularly good design these days. AMD's Opteron chips tend to scale MUCH better to multiple processors than Intel's Xeons.
FWIW a two-processor Opteron 246 (2.0GHz) system manages (base) SPEC CINT_rate scores of 28.8 and CFP_rate of 28.1. This makes the 2.0GHz PowerMacs 17.2 and 15.7 scores seem a touch low.
Ohh, and those Opteron scores were in Windows.
This is only marginally useful from a purely academic standpoint. In the real world, there are only two measures that are important: Highest maximum performance regardless of cost and performance for a given price point. How you get that performance, be it by higher clock speed or more work per clock is rather meaningless.
It's strange to see some people seeing a 2.0GHz G5 performing comperable to a 3.0GHz P4 (at least when Apple pays someone to do the benchmarks) and somehow assume that this means anything, or worse yet that somehow IBM/Apple are magically going to come out with a 3.0GHz PowerPC 970. This chip will NEVER make it to 3.0GHz on a 130nm fab process. It MIGHT manage to clock that high on a 90nm fab process, but I wouldn't even hold my breath for that if I were you. The P4 was designed to be able to clock high, which is why it's clocked at 3.2GHz on the same sort of fab process that has every other chip maxing out in the 1.5-2.5GHz range (I think AMD has the next fastest 130nm main processor at ~2.2GHz).
The numbers Apple got as STILL low for GCC, at least for the interger side of things (I haven't seen many numbers for CFP using GCC). Generally speaking, GCC is only about 5-10% slower than Intel's C compiler for x86, Apple's numbers are 20-25% lower.
Basically these benchmarks are a load of marketing crap. The IBM PowerPC 970/G5 is a perfectly fine chip, but it's definitely NOT the fastest PC processor in the world, not by a long shot. This is just another case of the "Steve Jobs Reality Distortion Field" in action.
It was for Quake3, and a 20-25% speedup was the goal, but the actual result was MUCH less, to the extent that it usually ran slower on dual-processor machines.
The simple fact of the matter is that in games, there is usually just ONE thread that is doing heavy processing at any given time, and it's extremely difficult to get around that. Sure, you can offload things like sound and the odd other thing to a secondary processor, but usually one chip ends up working full out 100% of the time with the other chip being idle 99% of the time. The nature of the work on most games just doesn't parallize all that well, besides which there is such a TINY market for dual-processor gaming machines that no one really bothers for it much.
If you want a gaming machine, you get a fast single processor machine with an even faster graphics card.
The floating point performance of the G5 is nothing to write home about. They only managed a score of 840 in SpecFP 2000, which is WAY bellow the performance of current x86 chips (as benchmarked by anyone OTHER than Apple/Apple paid testers).
Here are some official SpecFP (baes) scores:
The P4 3.2GHz : 1252
AMD Opteron 246 (2.0GHz) : 1209
IBM Power4+ 1.7GHz : 1598
HP/Intel Itanium2 1.5GHz : 2119
As you can see, Apple's score of 840 in this benchmark isn't exactly impressive. Even one of the new 2.4C GHz P4 chips (800MHz bus speed) will score higher.
Now, I know that these scores are using a compiler that is more heavily optimized than that used for the Mac, but guess what, there ARE compilers that are more heavily optimized out there for x86 than those for the Mac. MUCH more heavily optimized. Does this make the comparison somehow unfair? Not at all, because the end result is the same, applications run a lot faster on the newest x86 chips than on Macs.
Of course, system performance has never really been the reason why people buy Macs vs. a PC.
Those sorts of upgrade cards exist for PCs too, but no one uses the for a very good reason, because they SUCK! Sure you might be upgrading the processor, but you end up severely limiting the available bandwidth of the chip to even less than what was available on the previous system usually. A G4 with less bandwidth than a G3 is hardly a worthwhile upgrade for anyone, just as the old PIII upgrades for Socket 7 boards were mostly useless.
These upgrades are fine for people who absolutely can not replace their motherboard because they have some super-expensive proprietary hardware in there, but otherwise they usually cost about the same as buying a new processor, motherboard and memory while offering FAR less performance.
Ohh this is just too perfect. To quote Manfred from the above linked messages:
:>
"P.S.: I'd propose that the GPL rule that a change must be tagged with the name of the \
person who changes (or adds) a file is enforced - atealloc.c is only tagged with \
"SGI", thus I don't know who should be shot for writing that."
Looks like SCO is the one he should be shooting!
In my experience, most reboots aren't necessary, they're just the result of patch programmers being lazy. Microsoft only just recently stopped being lazy in this regard, and guess what? Their patches no longer require reboots, they just stop and restart the right services, run a few ocmmands and your set to go. Norton hasn't figured this out yet.
One thing I've found, for most Norton updates that "require" a reboot, you can just cancel their reboot and restart the Norton Anti-Virus service, and most of the time everything will work with all the latest and greatest updates. The odd time you need to register some component, and then it's usually just easier to reboot rather than trying to figure out all the undocumented commands.
Your comparing apples and orangutans here when it comes to power consumption. Intel's Thermal Design Power (which the maximum power consumption excluding thermal viruses) for the P4-M 2.0GHz is 32W. Transmeta's Thermal Design Power for the 1.0GHz Crusoe 5800 is 7.5W. The "1W" number that Transmeta likes to flaunt around is the amount of power that the chip consumes while it's sitting around twiddling it's thumbs. For comparison, a P4-M will consume about 5.0W doing the same thing.
That being said, the real competitor for the Transmeta chips is not the P4-M, but rather the ULV Mobile Celeron chips. Both chips will cost close to the same amount and consume close to the same amount of power (TDP for a ULV Celeron is 7.0W). Previously these Celerons, even at lower clock speeds, have had no trouble beating Transmeta chips. It remains to be seen if this new Transmeta chip has improved it's performance enough to compete with the Celerons. Otherwise it's just an odd-ball design with the same power consumption and similar price to a more widely available and better supported processor.
I don't know about that. They are promising only 50-80% improvements over their old Crusoe 5800 processor. That would put them as being about on-par with the chips that Intel had out two years ago when the Crusoe 5800 was first available, but it'll have a hell of a time competing with the chips that Intel is producing now, let alone 6+ months from now when this new Transmeta processor actually starts shipping.
I think that the real question will be how well this chip can compete with Intel's Ultra Low Voltage (ULV) Mobile Celeron line of processors. The two chips will have comperable power consumption (5-10W max, typical of under 5W) and probably won't be too far off one another in terms of price. Previous Transmeta chips have had a heck of a time keeping up with even the slowest mobile Celeron chips that Intel had available (read: they kind of kept up in MS Office, but got pretty well thrashed for everything else), but maybe this newer chip will bring performance up a bit.
Ugg, Transmeta in a handheld?! BLEACH! Why?
Let's see, we'll replace the current ARM or MIPS chips, which consume about 500mW or less, with a Transmeta chip that consumes 5-10W, for what exactly? Transmeta may have decently low power consumption for an x86 chip, but it's off by an order of magnitude as compared to the chips used in most handhelds.
I dunno, Microsoft has been giving us users the shaft for long enough, maybe it's about time to return the favor? :>