It's true the Cell has over 2.5 MB of cache/local storage, which is a lot, and doubtless accounts for many of the transistors. Memory transistors can also be packed more closely than logic, and take up less die space. The point is, it's still going to be a lot bigger, hotter and more expensive than a G5 - think P4 vs the Extreme Edition, only more so.
The 90nm Cell has a die size of 221 mm2. The 90nm G5 is only 66 mm2. 'Nuff said.
If you're thinking "Itanium" - do you know how much an Itanium costs?? Being cheaper than Itanium is not much of a recommendation.
The 90nm chip is only a prototype. The final Cell will be 65nm, and will be more practical at that size. If they can solve the transistor leakage problems that have been troubling Intel...
Details, indeed... The idea of transparent process migration to entirely different machines is not new; some mainframe systems have been doing it for maybe 30 years. It's not so much a function of the hardware (though it helps a lot if the systems are identical), it's primarily the OS that manages this - and it involves moving not just code but all the code's data, its entire local state, across to the new machine. The process may be unaware of this, but it can still take a hefty amount of time and/or bandwidth to do this. Not something you would be doing in the middle of a game, more when load-balancing a cluster of machines.
Having an OS that can do this does indeed make clustering easier, but it doesn't help as much for rendering farming (I wrote the render farm distribution code for a major AV processing package). Managing & synchronising multiple copies of a process across multiple CPUs and/or machines (for parallel processing, as with farms) requires application support, not just OS or hardware, at least for the great majority of applications.
None of this is particularly reliant on the CPU per se. Cell is great for number crunching, is well designed for multiple CPU systems, and IBM will no doubt churn out some killer (and cheap) supercomputers with it, but it'll take far more than just the hardware design presented to even approach the levels of hype ("Cell revolution" indeed) that Sony have been emitting....
The software might run without change on another Cell, but you still have to copy across all the state data - no getting around that, unless you want to manage shared access (across ethernet?). And migrating code to another Cell doesn't just happen automatically; it takes quite a lot of system support to migrate processes to different machines, not something you find often in a Mac or PC, let alone a console or a TV.
The primary advantage of consoles (to developers) is that they are a uniform environment, no hardware variations to complicate your software flow. I doubt any console developer would bother writing complex clustering code for the rare case that a player has an external device that can assist a little with the calculations, even if a high-bandwidth/low-latency two-way link existed between them.
It's true that Mac users treasure a boost in a Photoshop benchmark, even those users that rarely use Photoshop:-) AV processing does involve more than just number crunching, however - you need a lot of system and memory bandwidth to keep all those processors fed with enough data. Photoshop, After Effects et al are frequently limited more by bandwidth than by the CPU (which is why you see rather less than 2x increases in a dual system).
It seems to me that this is a natural for Apple - it will give them a 5x - 10x performance boost over anything that's on the drawing boards over at Intel.
That's a theoretical performance boost. Few apps will be able to take full advantage of 9 simultaneous processors, even after being coded with specific support for it. Still, it'd give a nice speedup to a couple of specific Photoshop tasks, and that's all you need to feed the Jobs RDF. If a Pentium III can speed up the Internet, then why not?
wouldn't it be great to run PS3 games on your Mac?
Like than all those Xbox games I run on my PC. Much better than playing it on my silly large-screen surround home theatre setup.
your PS3 could use your Mac to make its games run faster!
Yeah! The PS3 can run the physics code, and then copy all its internal state over the GigE link to a Mac every frame, which can render it! That's gotta result in a speed increase, and having to code for two different systems wouldn't be that much harder, would it?
If Sony can fit it in a console and sell a hundred million of them in a year, I'm sure Apple can...
Sony may be able to do that with the 65nm final design, when it arrives some time in 2006. Then we'll see.
Even then, there are other considerations that may make it a less-than-ideal fit for a general purpose computer - all those vector units are great for number crunching, but how much of that do you do each day? And when you're not, that's 3/4 of the cost of your chip sitting around idle. There are more cost-effective alternatives.
64-bit PPC on it has VMX. That's Altivec, baby. Sure, the SPE's don't have the full functionality of VMX but so what.
Read Part II of the article - it's not a full implementation of VMX (the SPEs don't have VMX at all - they have a different instruction set altogether). Hannibal believes the weak VMX implementation will be a major downside for Apple. Then there's the lack of out-of-order execution etc.
The biggest issue I see is that the Cell's design requires the programmer to have full control of the machine.
Not so. That's what operating systems are for. SPEs would be treated as a shared resource - you ask the OS to loan you one, and if you get it, you run your code on it. Or, you ask the OS to run your code, and it schedules it onto an available SPE when it can.
Scroll down a bit here, there's some more tasty tidbits.
e.g. 234 M transistors (!) That's why I don't think this will be replacing the G5 any time soon. The die size (at the current prototype's 90nm) is over 200 mm2.
It'll have to get a fair bit smaller/cheaper before the PS3 can use it without major subsidies, and I don't know why they think general consumer devices will want it. God knows how much power it dissipates with all 8 SPEs clocking over at 4 GHz...
So, estimates for the cost of the deorbit module range from $190M through $330M up to $1B - that's a lot of money just to toss something into the bin a little more accurately.
This might sound a little callous, but why not just let its orbit decay naturally (helped along with whatever its gyros can do), and put the some of the money into a global Hubble-damage insurance policy instead?
Chances are it'll just land in the water anyway, and even if it does hit the ground, it's still extremely unlikely to hit anything (or anyone) important. Skylab got away with it just fine.
Bad PR, you say? Announce a global Hubble Impact Lottery! Put $100M into the fund, and divide it up proportionately among those who are lucky enough to have to have any flaming wreckage plunging into their backyard. Highest allocations reserved for property damage or actual, provable injury. People will be praying for a "windfall" like that.
Now, where did I put that giant electromagnet anyway...
Many Australian ISPs allow unmetered transfers (unlimited volume) between customers, usually within the same state - it offsets the bandwidth caps and makes the plans more attractive. A lot of those are connecting to PIPE, a dark fibre backbone that allows fixed-cost peering, which is often unmetered too.
However, transfers are still made at your upstream bandwidth limit - e.g. with 1536/256 ADSL like mine, I get around 25 KB/s to upload to others, whether on the same net or not.
Best link I had was in the Bad Old Days of early Telstra cable... up to 30 Mbit/s down and 768 Kbit/s up. This was rather drastically offset by the 100 MB (yes, megabyte) monthly volume cap, which I could exceed at the so-reasonable price of 35c/MB...
Sounds like they're just covering their arses to me.
I see no particular reason why they couldn't add dualcore support to their BIOS - the hardware is pin-compatible, their customers are certainly going to want it, and AMD would probably help them ahead of anyone to get it right.
But, as always, buy hardware for only what it does today, and you'll never be disappointed tomorrow.
A 2200 chip doesn't have the necessary 32 PCIe lanes, nor does it have dual GbE. Only way to get that is with a 2200 and a 2050 as well, and you won't see that under $200.
Besides, if SLI gaming is your big thing, two full x16 slots is overkill, and won't affect your framerates more than 1% at best. Two x8 slots will be plenty for anything around or on the horizon.
Sounds to me like you want one of the existing nForce4 SLI boards from Gigabyte, with a drop-in dual-core Opteron to go with it. All you need, nothing you don't.
Pentium..........(original)
Pentium II.......(twice as good)
Pentium III......(3x as good)
Pentium IV......(4x as good)
Pentium M.......(1000x as good)
Pentium MMX.(2010x as good)
Apparently the Pentium III-M and IV-M were fairly good chips too.
...and the 15K drives can do 142 MB/s, which is more than marginal - while both 10K and 15K drives provide much lower seek times than any 7200 rpm drive.
RAM cache is fine, but it only partially hides latencies (and not for all applications), unless you're caching the entire disk. Best is a faster spinning drive and a RAM cache.
There is generally a tradeoff between speed and capacity, with the faster spindle drives (both ATA and SCSI) usually being smaller, but still delivering faster transfers and seeks despite the slightly greater density of the larger drives.
5) interest in 10K and 15K RPM is misplaced for most applications. Speed affects rotational delay and nothing else.
Hardly. All else being equal, if you double the rotational speed, you double the bulk transfer rate. Also, rotational delay is the major factor in the drive's average seek times.
Bulk transfer rate is more important in most applications
I would contend that seek time is more important for most applications (transaction servers, database access, web servers, booting PCs etc) than transfer rate (used mostly by media-intensive industries).
If it spins twice as fast but has half the density, it has the same bulk transfer rate.
6) interest in SCSI is outdated. SATA with one (competent) controller per disk has better characeristics.
Not according to this article, and others. It's improving, SATA NCQ/TCQ is a big help, but its only advantage is price, still. 10K and 15K SCSI drives have real advantages in both transfer rate and seek times.
Now that the US's copyright extensions, DMCA provisions and other restrictions to the public's rights have brought our wild and carefree legal system under control, do we at least get the Fair Use paragraphs to go with it, to decriminalise all the iPod users?
Is there an up side, or did we just end up with the worst of both worlds?
In the 19th century there were theories called "vitalism" that said that there was something about human life that was so fantastic that it was not understandable to science. The modern environmentalist movement has this same idea, that somehow humans are different from the rest of nature.
Actually, most religions believe that too. Souls are the ineffable quality that sets humans apart from the rest of creation.
You can certainly take the view that anything humans do is as natural as anything else. We're not going to screw up the planet any worse than a ELE-asteroid slamming into the crust, and that's happened more than once already.
But it's going to cause some inconvenience for us (drastic climate change, flooding etc), and for the species that go extinct as a result. It makes sense to minimise the effect we're having on the environment simply to give us (and the rest of nature) more time to adapt to the changes. The cost of dealing with global climate shifts will be an awful lot higher than losing a few jobs.
It depends entirely on the price, just like the CD/SACD discs.
If the DVD/HD-DVD hybrid disc is the same or very close to the DVD-only disc price, people will buy the HD version on the off-chance it will be useful to them in the future. But if it costs significantly more, it'll get completely ignored.
This could be a chance for Toshiba & partners to lock in a good chunk of the market before the market even exists - but they'll never give up their precious new-product premiums. No way will they let their fancy new HD content go for the same price as existing SD content (especially as the hybrid disc will no doubt cost a little more to manufacture).
That's right, and they still haven't, in their view.
Their 64 bit architecture is Itanium, and that's not in desktops yet. Prescott & other EM64T-enabled chips are "32 bit chips with 64 bit extensions" - hence their terminology of IA-32e. It's just a normal 32 bit architecture with some 64 bit address & data registers bolted on and a few new instructions to access them, not a real 64 bit architecture.
So by that logic, I figure that Prescott is really a 4 bit CPU, with extensions...
Indeed. I live in Sydney, and dutifully went to the shop to buy a copy on Tuesday morning, only to be told it wasn't available (here) until Wednesday - AUD$90 for the plain (HL2 only) version, and AUD$130 for the Collector's Edition, which also included CS:Source, HL:Source, other Valve games and a t-shirt.
So I went with Steam, where for US$60 (only AUD$73) I could get everything except the t-shirt. And I didn't have to wait until Wednesday either.
Oh, and rewarding Valve instead of Vivendi (manufacturer of fine buggy whips) was a nice bonus:-)
Slashdot Mirror
The 90nm Cell has a die size of 221 mm2. The 90nm G5 is only 66 mm2. 'Nuff said.
The 90nm chip is only a prototype. The final Cell will be 65nm, and will be more practical at that size. If they can solve the transistor leakage problems that have been troubling Intel...
Having an OS that can do this does indeed make clustering easier, but it doesn't help as much for rendering farming (I wrote the render farm distribution code for a major AV processing package). Managing & synchronising multiple copies of a process across multiple CPUs and/or machines (for parallel processing, as with farms) requires application support, not just OS or hardware, at least for the great majority of applications.
None of this is particularly reliant on the CPU per se. Cell is great for number crunching, is well designed for multiple CPU systems, and IBM will no doubt churn out some killer (and cheap) supercomputers with it, but it'll take far more than just the hardware design presented to even approach the levels of hype ("Cell revolution" indeed) that Sony have been emitting....
The primary advantage of consoles (to developers) is that they are a uniform environment, no hardware variations to complicate your software flow. I doubt any console developer would bother writing complex clustering code for the rare case that a player has an external device that can assist a little with the calculations, even if a high-bandwidth/low-latency two-way link existed between them.
It's true that Mac users treasure a boost in a Photoshop benchmark, even those users that rarely use Photoshop :-) AV processing does involve more than just number crunching, however - you need a lot of system and memory bandwidth to keep all those processors fed with enough data. Photoshop, After Effects et al are frequently limited more by bandwidth than by the CPU (which is why you see rather less than 2x increases in a dual system).
Right now, it has 4x as many transistors as a G5, runs at twice the clock speed, and likely puts out a hell of a lot more heat than a G5 does.
That's a theoretical performance boost. Few apps will be able to take full advantage of 9 simultaneous processors, even after being coded with specific support for it. Still, it'd give a nice speedup to a couple of specific Photoshop tasks, and that's all you need to feed the Jobs RDF. If a Pentium III can speed up the Internet, then why not?
wouldn't it be great to run PS3 games on your Mac?
Like than all those Xbox games I run on my PC. Much better than playing it on my silly large-screen surround home theatre setup.
your PS3 could use your Mac to make its games run faster!
Yeah! The PS3 can run the physics code, and then copy all its internal state over the GigE link to a Mac every frame, which can render it! That's gotta result in a speed increase, and having to code for two different systems wouldn't be that much harder, would it?
It'll have to be shrunk to 65nm before it can hope to be competitive.
Sony may be able to do that with the 65nm final design, when it arrives some time in 2006. Then we'll see.
Even then, there are other considerations that may make it a less-than-ideal fit for a general purpose computer - all those vector units are great for number crunching, but how much of that do you do each day? And when you're not, that's 3/4 of the cost of your chip sitting around idle. There are more cost-effective alternatives.
64-bit PPC on it has VMX. That's Altivec, baby. Sure, the SPE's don't have the full functionality of VMX but so what.
Read Part II of the article - it's not a full implementation of VMX (the SPEs don't have VMX at all - they have a different instruction set altogether). Hannibal believes the weak VMX implementation will be a major downside for Apple. Then there's the lack of out-of-order execution etc.
The biggest issue I see is that the Cell's design requires the programmer to have full control of the machine.
Not so. That's what operating systems are for. SPEs would be treated as a shared resource - you ask the OS to loan you one, and if you get it, you run your code on it. Or, you ask the OS to run your code, and it schedules it onto an available SPE when it can.
e.g. 234 M transistors (!) That's why I don't think this will be replacing the G5 any time soon. The die size (at the current prototype's 90nm) is over 200 mm2.
It'll have to get a fair bit smaller/cheaper before the PS3 can use it without major subsidies, and I don't know why they think general consumer devices will want it. God knows how much power it dissipates with all 8 SPEs clocking over at 4 GHz...
This might sound a little callous, but why not just let its orbit decay naturally (helped along with whatever its gyros can do), and put the some of the money into a global Hubble-damage insurance policy instead?
Chances are it'll just land in the water anyway, and even if it does hit the ground, it's still extremely unlikely to hit anything (or anyone) important. Skylab got away with it just fine.
Bad PR, you say? Announce a global Hubble Impact Lottery! Put $100M into the fund, and divide it up proportionately among those who are lucky enough to have to have any flaming wreckage plunging into their backyard. Highest allocations reserved for property damage or actual, provable injury. People will be praying for a "windfall" like that.
Now, where did I put that giant electromagnet anyway...
Though I suppose they could keep their precious A and give me 1536/3072, I wouldn't mind.
However, transfers are still made at your upstream bandwidth limit - e.g. with 1536/256 ADSL like mine, I get around 25 KB/s to upload to others, whether on the same net or not.
Best link I had was in the Bad Old Days of early Telstra cable... up to 30 Mbit/s down and 768 Kbit/s up. This was rather drastically offset by the 100 MB (yes, megabyte) monthly volume cap, which I could exceed at the so-reasonable price of 35c/MB...
Time to create article - 3 days /. - 5 minutes
Time to submit it to
Time to get a new webserver...
I see no particular reason why they couldn't add dualcore support to their BIOS - the hardware is pin-compatible, their customers are certainly going to want it, and AMD would probably help them ahead of anyone to get it right.
But, as always, buy hardware for only what it does today, and you'll never be disappointed tomorrow.
TaskManager can do this under Windows - the user can force any app onto specific CPU(s), and it won't migrate the app at all.
Besides, if SLI gaming is your big thing, two full x16 slots is overkill, and won't affect your framerates more than 1% at best. Two x8 slots will be plenty for anything around or on the horizon.
Sounds to me like you want one of the existing nForce4 SLI boards from Gigabyte, with a drop-in dual-core Opteron to go with it. All you need, nothing you don't.
Pentium..........(original)
Pentium II.......(twice as good)
Pentium III......(3x as good)
Pentium IV......(4x as good)
Pentium M.......(1000x as good)
Pentium MMX.(2010x as good)
Apparently the Pentium III-M and IV-M were fairly good chips too.
RAM cache is fine, but it only partially hides latencies (and not for all applications), unless you're caching the entire disk. Best is a faster spinning drive and a RAM cache.
There is generally a tradeoff between speed and capacity, with the faster spindle drives (both ATA and SCSI) usually being smaller, but still delivering faster transfers and seeks despite the slightly greater density of the larger drives.
Hardly. All else being equal, if you double the rotational speed, you double the bulk transfer rate. Also, rotational delay is the major factor in the drive's average seek times.
Bulk transfer rate is more important in most applications
I would contend that seek time is more important for most applications (transaction servers, database access, web servers, booting PCs etc) than transfer rate (used mostly by media-intensive industries).
If it spins twice as fast but has half the density, it has the same bulk transfer rate.
Why half the density? Why not the same density but 50% higher spin (and transfer) rate?
6) interest in SCSI is outdated. SATA with one (competent) controller per disk has better characeristics.
Not according to this article, and others. It's improving, SATA NCQ/TCQ is a big help, but its only advantage is price, still. 10K and 15K SCSI drives have real advantages in both transfer rate and seek times.
Is there an up side, or did we just end up with the worst of both worlds?
Actually, most religions believe that too. Souls are the ineffable quality that sets humans apart from the rest of creation.
You can certainly take the view that anything humans do is as natural as anything else. We're not going to screw up the planet any worse than a ELE-asteroid slamming into the crust, and that's happened more than once already.
But it's going to cause some inconvenience for us (drastic climate change, flooding etc), and for the species that go extinct as a result. It makes sense to minimise the effect we're having on the environment simply to give us (and the rest of nature) more time to adapt to the changes. The cost of dealing with global climate shifts will be an awful lot higher than losing a few jobs.
If the DVD/HD-DVD hybrid disc is the same or very close to the DVD-only disc price, people will buy the HD version on the off-chance it will be useful to them in the future. But if it costs significantly more, it'll get completely ignored.
This could be a chance for Toshiba & partners to lock in a good chunk of the market before the market even exists - but they'll never give up their precious new-product premiums. No way will they let their fancy new HD content go for the same price as existing SD content (especially as the hybrid disc will no doubt cost a little more to manufacture).
Toss it over the side onto the Greenpeace zodiacs?
Their 64 bit architecture is Itanium, and that's not in desktops yet. Prescott & other EM64T-enabled chips are "32 bit chips with 64 bit extensions" - hence their terminology of IA-32e. It's just a normal 32 bit architecture with some 64 bit address & data registers bolted on and a few new instructions to access them, not a real 64 bit architecture.
So by that logic, I figure that Prescott is really a 4 bit CPU, with extensions...
So I went with Steam, where for US$60 (only AUD$73) I could get everything except the t-shirt. And I didn't have to wait until Wednesday either.
Oh, and rewarding Valve instead of Vivendi (manufacturer of fine buggy whips) was a nice bonus :-)