I've been keeping an eye on ClearSpeed for awhile, ever since the original TomsHardware article ages ago, but none of the software I use would benefit from it. ... the future is looking far better for them now.
Platform breakdown: (The starting blocks for the race to 2010):
Time to get into coding for next generation platforms for 2008 - 2012.
:arrow: IBM - Cell / G5 and beyond (Former Apple / Mac provider)
:arrow: Intel - x86/x64 EMT64 / IA-64 (Itanium on 65nm / 45nm or 32nm would rock) (now providing to Apple / Mac)
:arrow: AMD - Opteron x86/x64 AMD64 + ClearSpeed
:arrow: Via - Low power processors, you could run 100's of them using only 500 watts
:arrow: NSC - Sold designs for Geode to AMD (New AMD Geode LX released recently btw, similar to Via)
:arrow: Cyrix - NSC brought them out, then sold Geode to AMD, and rest to Via (There was a Via/Cyrix transition period)
:arrow: Transmeta - Selling patents to Sony, no longer making processors.
:arrow: Sun Microsystems - Still in the game 'CoolThreads' - 32 threads per chip is their flagship model / design currently.
Yeah, but dude, last time it was raised was in a negative light.
I've included a (very) small summary at the end indicating where companies are going.
AMD stand out to me as doing something similar, but to process large amounts of math related operations.... a move that interests me, because I have a interest in ClearSpeed.
ClearSpeed, is the Weitek of the 21st Century... and back then I had an interest in Weitek Co-Processors.
Heck, go google 'Weitek'.... I've never asked you to do something on google before, but you've sure tried to 'tell' me to (Although it appears you have zero power over me, and we both share a common interest in computers and IT, I even enjoyed your thread on the RAID flash media, and contributed information on the 8 GB LaCie Carte to it so perhaps more people will try it out).
This thread is to discuss the future of accelerated math processors / cores, a future Weitek never had. We could be seeing massive gains in performance using a similar amount of power, and the same manufacturing processes. (eg: 90nm SOI, 65nm, 45nm, 32nm, etc)
Comments such as the above are not ..... appreciated.
I'll just activate my Firefox plugin to autoclick [Ignore] on all your posts for another hour or two.... [sigh].
I don't understand you bringing up add-in cards; they have always existed for special functions and have nothing to do, at all, with the CPU, system or even platform. (Your video card is no different than the Clearspeed board in this respect.)
All the excitement is coming from the fact that AMD's coherent HT would allow
something like this to be integrated almost as another 'core' if you will: you can hook-up the two processors directly, without any latency-adding slow middle parts. And since deticated hardware is always better than general-purpose hardware (again, look at your video card) for special applications (3d rendering) the excitement is easy to understand among professionals.
As for your 'Platform breakdown'...whatever :roll:.
If AMD can license it's CHT out fast enough, and get enough (exclusive)partners on-board before Intel's CSI gets here (something I doubt will happen - we'll see), their interconnect technology will win and Intel will have little choice but to license it as well.
A more probable scenario is a fight not only on the CPU front but on the 'who has what specialty hardware partners' front as well.
I'm just wondering why it took (is taking) AMD so long to license its CHT for such uses. Maybe because they had little real competition in the x86 server market, but it's a shame to hold back on innovation just because your competition is floundering.
I'm waiting for a CPU that makes an actual difference in gaming, and it's not going to come from a faster general-purpose one, despite all the fanboy screaming. We 'need' the CPU to somehow accelerate graphics and a co-processor is something worth trying in that direction.
The math coprocessor for the 80286, 80386 and 80486 are single core and mostly are helping the main processor to do the math(ex. they have tabled trigonometric functions).
ClearSpeed have a lot of single core processing units that are sharing and processing the data(floating point operations) by them selves.
Becouse of the physics of the silicon made MOS transistors(the limit of signal acceptions and transitions on apropriate voltage), higher voltage is needed to achieve higher freqfency. Both, the voltage and the power consumation(read generated heat) are growing exponentional to the linear growth of freqfency.
If a job can be devided in two or more that can be done in parallel at same time, more processors runing at lower freqfency and lower voltage(using the power more effective) will do the job faster and will disipate less heat. There are jobs that must be processed sequentionaly and have no use of parallelism, but multimedia SIMD operations are a clear example where parallel processing can be implemented. Multimedia are becoming more general-purpose and in future general-purpose will be multimedia. While the heat and power consumation are limiting the further performance of singlecore processors, multicore processors are the future.
The current dualcore(in future with 4 or more cores) processors and the revolutioner asymetric 9-core architecture "Cell" are the begining of the multi-core multi-cpu era of desktop computing. In future we will use asymetric processors with many same and different types of builtin cores(like ClearSpeed) depending on the purpose.