Archived

This topic is now archived and is closed to further replies.

h2egc

Intel Plans 32-core Processor By 2009/2010

Recommended Posts

Help AVSIM continue to serve you!
Please donate today!

First I've seen a name attached to it. Good find. This will continue Intel's Tick-Tock strategy. The Penryn's job was not to introduce a new, faster CPU. It was to prove the 45nm manufacturing process. Its only claim to fame was it higher O/C potential. Once 45nm was proven, a new archecture was inserted. Enter Nahelem. This still just a tweaked Core2, however the change is the expensive north bridge is now onboard the die. This change is more about platform cost reductions. Clock for clock performance increases of 30% will prove optimistic. This basic architecture will (next) be shrunk to 32nms. This will be the same non-cloclxclock performance event we saw with Penryn, but it may start to use all that extra real estate for more cores. I expect the second iteration (the "tock") of this 32nm platform will be about maximizing the number of cores onboard. ~ 40 cores was floating around, so 32 makes sense. From here on out, its Not about clock for clock performance increases. It will be about number of core on board. As this is the irrevocable case, these 10-30% increases pale in the light of what's coming . . . in only a couple of years. The fanbois will still go on and on. After all, Intel Needs to make a living in the mean time., but let the deep-pockets enthusiasts pay for that. If you have a Q . . . any Q, hang on. That 10X performance increase is near, software willing.

Share this post


Link to post
Share on other sites

Where is all this new info?Besides the rather limited amount of info from the former link above.

Share this post


Link to post
Share on other sites

What is promising is if Sony / IBM are able to expand on their 'Cell' technology. They have a chip that has 8 processors on it for the PS3 and it is amazingly fast. If they could modify this and get it to work in the PC industry it would be a significant advancement.

Share this post


Link to post
Share on other sites
What is promising is if Sony / IBM are able to expand on their 'Cell' technology. They have a chip that has 8 processors on it for the PS3 and it is amazingly fast. If they could modify this and get it to work in the PC industry it would be a significant advancement.
This is a misconception. Lots of techno jargon here, I don't expect everyone to understand but if you desire an answer, it lies within. Read on if you dare.The Cell processor is a very powerful streaming media processor. It handles floating point calculations with great speed, rivaling even modern x86 (Intel and AMD) processors. However, it does not support the x86 Instruction Set Architecture and therefore cannot run any program which Intel/AMD processors run. This x86 ISA was created by Intel, and is licensed to AMD at a cost (and by court order). STI (Sony/Toshiba/IBM) does not posess this license, and as anyone who's followed this industry over the years knows, the last thing Intel wants is more competition. The only other path of entry into this market is emulation, an option which drastically reduces performance. See Transmeta for a perfect example of this. Even Intel's own Itanium Processor Family, the most expensive processors Intel makes, emulates x86 instructions and at a great performance cost. A $10,000 Itanium processor will have its doors blown off by a $100 Core 2 processor when executing x86 instructions.Additionally, the design of the Cell Broadband Engine Architecture is also not well-suited to the execution of serial code with many dependencies. It is an in-order processor with relatively poor branch prediction, which is the complete opposite of modern x86 processors which are Out-of-Order (OoO) and have excellent branch prediction. Branch prediction is technology designed to predict the correct code path to follow when a branch is encountered (pseudo-code example: if condition x is true do y, else do z). OoO processors have the capability to re-arrange the stream of instructions on-the-fly in an attempt to avoid pipeline stalls (a condition where an instruction unit is unable to proceed with the task at hand because it is waiting on the result of another instruction at another point in the pipeline) so as to keep all of the available execution resources as busy as possible. This is a large contributor to one of the most important aspects of performance: Instructions Per Clock (IPC). Since in-order processors cannot do this, they have to run instructions in the order they are issued by the application. Essentially, in-order processors get less work done per clock than their OoO counterparts, and thus have to be clocked much higher to get the same amount of work done per second. Now you may ask yourself: why not just run these processors at a higher clockspeed then? There are multiple problems with this approach, power consumption/thermal output are not the least of these. There are also manufacturing barriers to overcome when attempting to run processors at such high clocks. This is the reason why Intel abandoned the Pentium 4 (Netburst Architecture) and adopted the Core Architecture instead. Originally the Netburst family of microprocessors was intended to run at speeds up to 7GHz, its predecessors slated to run at speeds in excess of 10GHz! They ran into a wall when attempting to clock their Pentium 4 chips to the sky (a thermal/power wall) and had to go back to the drawing board to design a chip which gets more work done with fewer clock cycles. In order to achieve high performance from an in-order processor design and always keep execution resources busy by avoiding pipeline stalls, as well as avoiding unnecessary work created by traversing all branches in a tree, instructions have to be re-ordered by the application compiler when the application is compiled, or even hand-tweaked by the application developer. The first option is the easiest but offers very little performance gains, the second takes a lot of time and is generally the most difficult aspect of modern programming. Now you may ask: why does the Cell (or it's tri-core cousin in the XBox 360 - Xenon, which uses the same IBM PowerPC ISA) manage to do so well with all these barriers? The reason is because game consoles are a closed environment. This means when an application developer creates an application for one of these systems they have a specific set of hardware and software to code for. Every PS3 has the same processor, same amount of memory, same graphics processor, same Operating System - it all adds up to the same performance capabilities and the same limitations. This is not true in the PC world, where there are millions of possible hardware AND software configurations which make optimization an extremely difficult process. PC application developers have to deal with different video cards, different processors, different chipsets, different operating systems, even different drivers. At best PC application developers can only hope to optimize for a small subset of the most commonly used hardware and software configurations. In the console world, application developers can spend significantly more time optimizing their application(s) than their PC counterparts because of this. Regards,-Max

Share this post


Link to post
Share on other sites