Intel stuffs Nehalem chips with joy Intel has tick-tocked itself into a frenzy about its upcoming line of multi-core fancies now being discussed under the "Nehalem" code-name. Server chief Pat Gelsinger took the stage this week at IDF to speak of Nehalem's wonders, telling the audience that the re-architected chip line will change their businesses and perhaps …
"n the first "part" of 2009, Intel will also dish out Tukwila - the beastliest version of Itanium yet at 2bn transistors." Blah, blah, blah, blah. The real question is how well does it perform while playing Crysis???
Paris, cus I'll be she's got some "interesting" toys to play with.
Mines the one with the liquid nitrogen cooling system sewn into the lining.
Can I now please also get a compiler (linux compatible, I don't do windows) that will allow me to magically transform all my "one-thread" softwares to miraculously multithreaded apps ?
(Can someone also definetly forbid this technology to all Gentoo fanboys ? Yes, I want to make faster divx encodes and the like, but I'm not beyond seeing Geentoo's EmErgE fanboyz suffer needlessy...)
>Steve, because he would be proud of me<
We need comparisons, how fast and how expensive are these chips? How does the price/performance compare with current Core Duo chips?
Like do you buy a CPU now or do you wait months for these new CPUs only to find they're too expensive at first and the faster models won't even be out for a further 6 months?!
More cores != better performance.
More cores are only a benefit if the software that are running on them is properly multi threaded. For single threaded software there is usually a performance hit as multi core CPUs usually run at a slower clock speed to keep power use and heat production down. In addition, the more cores there are, the longer the memory latency can become as the memory buses get clogged which leads to a core stalling for longer.
And Just to really add some confusion to the mix, some OSes don't scale as well over multiple cores.
Most software is already using multiple threads, from desktops to servers.
The last single-threaded OS was DOS with Windows 3.1... but there were even TSR's back then.
With the advent of Windows 95, spawn windows spawned new threads. Dozens of services require CPU time with NT. Menus are now being checked in the background for usage statistics & being being rebuilt reformed. Indexing is always happening in the background. Speech recognition leverages threading and lookups. People are playing music & video in the background on their PC's all the time. Video & Audio chatting is normal now, where those incoming and outgoing streams are constantly being compressed and uncompressed in the background. People use their computers while rip'ing CD's and burning CD's. People are using their computers while they are recording videos from their TV, as well!
People are always running with multiple applications on the screen at the same time, and those applications are normally running more than one thread.
Most servers heavily use multiple threads and processes. Often, there are hundreds of processes and thousands of threads (my 1 processor 6 core 24 threaded lightly loaded server runs 611 processes and 1786 threads, while my heavily loaded 8 processor 16 core sever runs 252 processes and 2568 threads.) Web services, databases, and applications leveraged by multiple users are heavily thread dependent.
More cores and threads allow these interactions to be more seamless, increase throughput, reduce overhead of context switches, better manage performance peaks during high user counts,
The single threaded operating system and application was a myth promoted by chip manufacturers who were behind-the-curve technologically... If it was not a myth, these chip-manufacturers would not have been investing trillions into multi-core and multi-threaded silicon.
Laurent_Z wrote:
Can I now please also get a compiler (linux compatible, I don't do windows) that will allow me to magically transform all my "one-thread" softwares to miraculously multithreaded apps?"
icc -parallel
Intel wrote:
"Enable the auto-parallelizer to generate multi-threaded code for loops that can be safely executed in parallel.
If you're all OO then:
icpc -parallel
or if you're an old beardy:
ifort -parallel
Performance: YMMV unless you think about code structure, memory affinity, etc.
Just when Hyper visor was sounding Great under Find Host. Vt-d, No Simple interconnecting, hopefully it will self configure.
Turbo isn't use of second memory here, as turbo game cards or turbo power with boost psu, this is ALL New use of term TURBO. More power can be managed especially with so many sq wave generators, that power can be regulated or even turned off in poor areas, is great. certain events like extracting files, decoding or opeing, can be handled individually by need, while unused time, much more conservatively.
drashek.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
