Slashdot Mirror
Two Approaches to the Next-Generation Desktop
puppetman writes: "Tom's Hardware has a review up of a pre-production P4/2666 using 533 mhz Rambus memory (and shows it stomping the competition). The Pentium 4 needs memory bandwidth, and DDR doesn't supply it. Or does it? Anandtech, ironically, has a preview of the E7500 chipset from Intel - dual channel DDR with support for up to 16 gig of RAM. With a new bus architecture, this looks perfect for high-load databases that need wide pipes to hard-drives, memory, and ethernet. Both of these technologies look great for mid-range database servers.
Anandtech claims that dual DDR200 will provide 3.2 gig/second bandwidth, where Tom claims that DDR266 (single channel) offers only 2.1 gig/second. Intel is sure hedging their bets. I wonder what AMD has up their sleeves."
the link to the 3500 amd the 3675 p4 can be found here http://slashdot.org/articles/02/01/17/1823233.shtm l
Saying Java is nice because it works on all OS's is like saying that anal sex is nice because it works on all genders.
Newbies like me using JBuilder Enterprise really like the automatic code recompilation and all the other shiny widgets and other nifty graphical gismos. The thing runs like crap with less than 512 mb of RAM on a PIII-800, although it does feel reasonably snappy on my 1.4 ghz Tbird with a gig of ram.
To access data in a Rambus module, the request must pass through all modules in sequence up to the module that has the data and then must pass back through those modules to deliver the data to the northbridge. This is, BTW, why continuity RIMMs are required.
As one can derive, this greatly increases latency as the number of modules increases. Servers, being systems that generally have lots of RAM, often have at least 8 modules available.
Due to this increased latency as a function of the number of modules (and other factors), Rambus is therefore poor memory for servers.
Note that this is per channel, meaning a dual channel Rambus system with eight modules has the memory latency of a four module system because the modules are split between the Rambus channels.
Computer Science is no more about computers than astronomy is about telescopes. --E. W. Dijkstra
I'm running radiative transfer codes overnight, I wish I could get them to finish in about 20 minutes. If I had a 5 gig. system, that still wouldn't be fast enough. If it were, then I would crank up the resolution, and run it overnight. It won't be fast enough until I can run a weather model at a reasonbly high resoulution on my home PC in near real time. It will never be fast enough. You double the processor speed, I'll halve my grid sizes.
With all of the posts saying that our 1GHz's are fast enough, I say until Quake n looks like Final Fantasy (the movie!) we don't have fast enough CPU,RAM,Video,[Insert Bottleneck Here].
There is no longer anything that can be done with computers that is nontrivial and clearly legal. -- Paul Phillips
DDR 1600(200) does indeed provide 1.6GB/s of memory bandwidth, just as DDR 2100(266) provides 2.1GB/s, and DDR 2700(333) provides 2.6GB/s. The current P4 line of processors use a quad pumped 100Mhz pipeline capable of handling 3.2GB/s of memory bandwidth. This can be accomplished by a dual channel PC800 rambus memory controller or by a dual DDR 1600 memory controller (Which nobody currently has). The future specs for the quad pumped 133Mhz pipeline uses the new PC1066 rambus in a dual channel configuration. This same memory bandwidth can be acheived using a dual DDR2100 bus. However the new DDR 2700 can provide a 166Mhz quad pumped memory bus which would, in theory, be the fastest solution. If intel wanted to increase their lead in the market, they would be smart to experiment with a dual DDR 2700(333) configuration with their P4 platforms. Personally I prefer DDR as it doesn't have a proprietary intellectual property licensing scheme that rdram has. Just my $0.02
It is important for a variety of reasons not to let up upon the current technological pace occurring today. There are so many factors to consider economically, scientifically, and sociologically. If we allowed a slow down of the current pace of technological advancement it could have a devastating impact on our society at large.
First off, it is naive to think that current users wouldn't use or enjoy more powerful computers. It is the software industries fault that end users are unable to fully utilize the more powerful machines being built. Already plenty of comments have suggested a variety of applications from facial recognition to video editing that all would benefit from faster more powerful computers.
It is actually important to me that regardless of the 'need' the average user has for more powerful computers, that the software industry does its job to drive the users to want more power.
Only by nurturing and then feeding the publics appetite for technology does the industry continue to push us forward technologically. If millions of people and companies didn't demand the upgrades and new features that are available with more powerful systems we risk losing all the potential gains for the future that these desires produce.
Not all geeks want to concern themselves with the minutia of building their own box. Sometimes they just want to plug it in and get some work done.
pencil-dick said:
"You bought an IBM computer.
Sorry, waste of money, obviously bought by somebody with a wallet larger than their brain. You are not in a position to talk about the internals of a computer, because if you were, you would have a custombuilt beigebox like the rest of us."
Out in the real world, where make our wallets larger than you can imagine, employers expect us to get work done on computers, without interruption. To those ends, they often insist on brands such as Compaq and IBM. This is because they have large contracts with these companies that stipulate when a machine breaks there will always be an identical spare on hand to swap out the broken one.
You see, punk-ass hardware monkeys like yourself aren't generally interested in a computer that will work non-stop, error-free. Some people are. Those interested parties call people like you when something breaks and we just demand you bring us a new one. We know it'll take you 30 minutes to roll a new one in on your magic card, or 4 hours under our desk to repair it, and we also know that those 3.5 extra hours are worth more in time lost to the company than you make in a week. (That's why we demand a new one. And get it.) What's funny is that we laugh at you when you leave, because you think your A+ certification and ability to plug in an ide cable means something, and because you're probably planning on going back to your bench to troubleshoot the damn thing.
Anyway, get your head out of your ass and learn what getting paid real money to work on a computer is all about. The price difference between a Sys Athlon beige box and a similar IBM or Compaq bought on contract is about $800. And that's roughly what it costs a company when they lose a few hours of time from a SysAdmin, DBA, Java Programmer, or other similar keyboard jockey.