Archive for the Intel Category
I’m a Python developer, dabble in Ruby, run FreeBSD on over 20 machines, plus I like to boot Ubuntu once after Saturday in order to still prove that Linux trumps Windows any given ‘Sunday’.
When I see articles like this on parallelism, I wonder if they really know why there can’t be more split work on threading. Intel given compilers aside, there is only so much extra work that can be divided among threads before there is too much overhead involved in the processing of said work.
In the end, you can call octacores the end of all, or you can wonder what the world will be like when you have 200 terabytes on board, with your 80 core machine running on an 8 gigabit connection. We simply don’t know what to split all those tasks into, let alone the system being able to reach any efficiency about it.
That’s what I’m calling the future of computing, since most people consider the future to not even include small time computer makers. The changes going on in the Industry are new, but the same old thing all over again.
Your local computer store wants to make you pay up to 40% markup, and in some cases 100% markup to just cover warranties that, if they priced their machines lower, and sold more machines, would barely need to cover labor not parts. This type of behavior precludes the local shops from selling machines, at great insistence that the bigger guys make it impossible to sell a complete computer system.
Sun on the one hand would love to sell you a network computer, aka a thin client so that there is no need for you to have a hard drive or for that matter, even a multi-core machine. The reality is that any type of network computer such as that will require a fat pipe to allow multimedia or data intensive applications to run. The soothsayers would love to tell you that this is where computing is headed, but in reality it’s going in a different direction all together. (I simply cannot imagine a screen being updated 180 frames per second accros 8 monitors)
Intel is demonstrating that teraflop performance is at the desktop very soon, with technology, that they announced today would be going into new processors. Larrabee is the name of the technology that will enable processors to actually manage up to 80 cores, and do it with an amazing amount of power savings.
At the Intel Developer Forum today, the Chief of server chips, Pat Gelsinger confirmed the design, and saying that ” It will be many cores and you can expect that different versions of the processor will have varying numbers of cores.” He refused to elaborate on any of the new products, but simply stated that they will provide at least a teraflop in processing power.
Pat went further to actually demonstrate an 80 core unit, and then wowing at least me with the ability to reach 1 teraflop with 46 watts, which is no small feat. Pat also noted that not many coders have the required skill set to craft an application that would take advantage of so many cores.
I for one welcome the impending 80 core chip, as such, having the chip in sufficient quantities will allow coders, such as myself, the ability to dive deep into multi threaded design of any application. Applications such as Pygame could no doubt spawn threads that would run across multiple cpu’s and then allow many many more sprites to run at the same time.