Intel's CPU future through 2012 leaked online
With Intel's Developer Forum just days away, an apparently leaked presentation for the event indicates where the chipmaker's processors are headed over the course of the next four years.
The slides obtained by French tech site CanardPlus start off by recapping the imminent launch of Nehalem, which is now officially labeled Core i7 and is the first big break from Intel's traditional architecture. As is increasingly well-known in tech circles, i7 will switch to a new point-to-point bus architecture and return the Pentium 4's Hyperthreading feature, which can sometimes mimic a second core by running more than one code thread at the same time. The technology is already set to be discussed in-depth at the Developer Forum and will launch in the fall with new Core and Xeon desktop processors.
It's here, however, that the presentation veers into largely unfamiliar territory. Apart from planning a chip die shrink to 32 nanometers for i7 due later in 2009, known as Westmere, Intel's next big change in architecture is now set to take place in 2010 with a technology known as Sandy Bridge.
While lightly discussed in the past, Sandy Bridge is now said to focus heavily on vector math -- an important component to certain 3D and movie operations and once the strongest selling point of PowerPC-based Macs. The processor design will introduce support for new programming features known as Advanced Vector Extensions, or AVX, which will not only be much more complex with 256 bits of data versus 128 for today's SSE equivalents but will support as many as three or four calculations in one instruction depending on the task at hand.
The overhaul of Intel's chip design will also be built with the capability to handle at least eight cores on a single chip and will have much less Level 2 memory cache than today, at just 512KB per core, in return for 16MB of Level 3 to be distributed among all the cores. This architecture will be shrunk sometime in 2011 when it's known as Ivy Bridge, according to Intel.
For future processors, the Santa Clara, Calif.-based company has less detail but vows a breakthrough at least as significant as Sandy Bridge. Nicknamed Haswell, it will have "revolutionary" power management, an all-new approach to caching, and the option of dedicated vector coprocessors in a package separate from the main processor. It will also be the first Intel chip to support Fused Multiply-Add instructions that, as the name suggests, include math with both additions and multiplications in a lone instruction.
Intel isn't expected to confirm at least some of these details for either Sandy Bridge or Haswell ahead of the Developer Forum, which starts August 19th in San Francisco. With both processor generations not due for at least another two years, though, the presentations made at the event will, for now, be the best look at where mainstream computers will be in the future.
The slides obtained by French tech site CanardPlus start off by recapping the imminent launch of Nehalem, which is now officially labeled Core i7 and is the first big break from Intel's traditional architecture. As is increasingly well-known in tech circles, i7 will switch to a new point-to-point bus architecture and return the Pentium 4's Hyperthreading feature, which can sometimes mimic a second core by running more than one code thread at the same time. The technology is already set to be discussed in-depth at the Developer Forum and will launch in the fall with new Core and Xeon desktop processors.
It's here, however, that the presentation veers into largely unfamiliar territory. Apart from planning a chip die shrink to 32 nanometers for i7 due later in 2009, known as Westmere, Intel's next big change in architecture is now set to take place in 2010 with a technology known as Sandy Bridge.
While lightly discussed in the past, Sandy Bridge is now said to focus heavily on vector math -- an important component to certain 3D and movie operations and once the strongest selling point of PowerPC-based Macs. The processor design will introduce support for new programming features known as Advanced Vector Extensions, or AVX, which will not only be much more complex with 256 bits of data versus 128 for today's SSE equivalents but will support as many as three or four calculations in one instruction depending on the task at hand.
The overhaul of Intel's chip design will also be built with the capability to handle at least eight cores on a single chip and will have much less Level 2 memory cache than today, at just 512KB per core, in return for 16MB of Level 3 to be distributed among all the cores. This architecture will be shrunk sometime in 2011 when it's known as Ivy Bridge, according to Intel.
For future processors, the Santa Clara, Calif.-based company has less detail but vows a breakthrough at least as significant as Sandy Bridge. Nicknamed Haswell, it will have "revolutionary" power management, an all-new approach to caching, and the option of dedicated vector coprocessors in a package separate from the main processor. It will also be the first Intel chip to support Fused Multiply-Add instructions that, as the name suggests, include math with both additions and multiplications in a lone instruction.
Intel isn't expected to confirm at least some of these details for either Sandy Bridge or Haswell ahead of the Developer Forum, which starts August 19th in San Francisco. With both processor generations not due for at least another two years, though, the presentations made at the event will, for now, be the best look at where mainstream computers will be in the future.
Comments
I should really stop keeping up with all of this future hardware information. Since i am always thinking of what is coming next, I feel bad buying the current products.
I could care less about Sandy Bridge or Ivy Bridge right now. I've got myself on a schedule. I bought a Core2Duo 17" MacBook Pro (which was nice because of the 64-bit processor, but I have a 32-bit chipset) and my next 17" MacBook Pro will be next year with Nehalem. It'll be time for an upgrade and it'll be a good upgrade at that. After that, my next 17" MBP won't probably be until 2012, so I'm more interested in Haswell.
Intel has a tick-tock format to their architecture, so there is one big upgrade (Sandy Bridge) and then a small upgrade (Ivy Bridge). Just upgrade when you need it and then build a timeframe after that, there is always something better coming out.
I should really stop keeping up with all of this future hardware information. Since i am always thinking of what is coming next, I feel bad buying the current products.
This is a pretty stupid attitude. I like it when new things come out, even if I bought a new machine not long ago. It means when I buy a new one I'll be getting something better that has the bugs ironed out. I buy new machine every 18-24 months and sell the old one. It's a bit more expensive this way, but if you can't afford it, then you shouldn't be in the market or care. If you just want the latest and greatest (no matter how buggy or how long you have to wait in line for) then you should just go get your head checked.
I could care less about Sandy Bridge or Ivy Bridge right now. I've got myself on a schedule. I bought a Core2Duo 17" MacBook Pro (which was nice because of the 64-bit processor, but I have a 32-bit chipset) and my next 17" MacBook Pro will be next year with Nehalem. It'll be time for an upgrade and it'll be a good upgrade at that. After that, my next 17" MBP won't probably be until 2012, so I'm more interested in Haswell.
Intel has a tick-tock format to their architecture, so there is one big upgrade (Sandy Bridge) and then a small upgrade (Ivy Bridge). Just upgrade when you need it and then build a timeframe after that, there is always something better coming out.
Also it's all just vapourware. God knows what they will actually bring out and when. Wait until it's sitting on the shelf, or at least announced.
I should really stop keeping up with all of this future hardware information. Since i am always thinking of what is coming next, I feel bad buying the current products.
Buy an iMac every two to three years, perfect timing to keep you close to the latest and greatest, at least as far as what they put in the iMac. If you use another computer upgrading every three to four years could be good for the MacPro's. Assuming that you use your machine for work, if pleasure go the iMac route or MacBook route every two to three years.
http://www.intel.com/pressroom/archi...070204comp.htm
http://techresearch.intel.com/articl...Scale/1449.htm
http://news.cnet.com/Intel-shows-off...3-6158181.html
This is a pretty stupid attitude. I like it when new things come out, even if I bought a new machine not long ago. It means when I buy a new one I'll be getting something better that has the bugs ironed out. I buy new machine every 18-24 months and sell the old one. It's a bit more expensive this way, but if you can't afford it, then you shouldn't be in the market or care. If you just want the latest and greatest (no matter how buggy or how long you have to wait in line for) then you should just go get your head checked.
The point of my original post is that even if I kept buying the latest and greatest components, they would be obsolete in one year and their prices would be significantly lower. It sounds to me like you assume that I am still using a 128K mac because i can never bring myself to upgrade or that I hate the progression of technology.
In reality, I just buy products when the release features a significant performance update over the last generation. That is why I will purchase the next mac pro which should have at least a 40% faster cpu, a gpgpu, and numerous other upgrades.
For my pc, I just keep replacing each component individually every 6-12 months trying to anticipate the next release. Then I sell the old components while they are still the latest generation so that they retain most of their value. During the time after the sale and before the next product's release, I just use older components or another computer. It is actually cheaper than buying a new computer with similar specs every 2 years.
What happened to the 80-core processor first announced September 2006 and claimed to be on the market in five years (so 2011)? See:
It was only a research project for testing inter-core networking, never intended for production. Larrabee probably incorporates a few things Intel learned from it.
It is also a good thing this is off in the future as I'm out of the PC market for awhile
The point of my original post is that even if I kept buying the latest and greatest components, they would be obsolete in one year and their prices would be significantly lower.
Wow. Get a dictionary. Look up the word, "obsolete". I don't think it means what you think it means.
In reality, I just buy products when the release features a significant performance update over the last generation. That is why I will purchase the next mac pro which should have at least a 40% faster cpu, a gpgpu, and numerous other upgrades.
In reality, you should buy a product because you need or want it, and should use your current machine as a metric of determining the value, not the previous release that you didn't buy.
What CPU do you think will be in the next Mac Pro that will be "at least a 40% faster cpu."?
In reality, you should buy a product because you need or want it, and should use your current machine as a metric of determining the value, not the previous release that you didn't buy.
What CPU do you think will be in the next Mac Pro that will be "at least a 40% faster cpu."?
he's obviously referring to Nehalem, and he means in total throughput, not clockspeed. I don't think we'll see 40% improvement across the board, but on some multi-threaded, floating-point intensive applications 40% is probably realistic. Remember, Nehalem has the second half of the SSE4 instruction set which will also improve things in addition to all the other changes.
he's obviously referring to Nehalem, and he means in total throughput, not clockspeed. I don't think we'll see 40% improvement across the board, but on some multi-threaded, floating-point intensive applications 40% is probably realistic. Remember, Nehalem has the second half of the SSE4 instruction set which will also improve things in addition to all the other changes.
That is what I was getting at but I was given the OP the benefit of the doubt for the comment. SSE4 would be nice but I fear Apple won't even begin utilising it until mid to late 2009. Have they even utilised SSSE3 yet?