Can you say 100 core cpu

ShuttleLuv said:
Windows 7 support? LOL
Click to expand...

Doesn't windows 7 have provisions for massive parallelization? I remember reading about how its threading optimizations will become noticeable with something crazy like 50 threads.
 
wow is the only thing popping into my head... that and the OC numbers that people may be able to push outta thins thing if its ever actually released.... or compatible with windows.
 
I can imagine some may be able to get this up to 2.0GHz maybe even 2.5GHz with water cooling. With that it may be able to boot windows 7 in under 2 minutes.

Again this is not a general purpose cpu. It will be pretty bad at that.

However it would be great at running in a blade server running linux with 100 websites running each in their own virtual container.
 
Last edited:
100 cores is nice but then your programs would need to take advantage of it as well, not just the core os
 
I do not see this being feasible any time soon. Being a programmer as a profession I know that most computations can not be divided into 100 parts. That's absolutely nuts. We have hard enough time efficiently dividing a problem in 4 parts.
 
Last edited:
drescherjm said:
I do not see this being feasible any time soon. Being a programmer as a profession I know that most computations can not be divided into 100 parts. That's absolutely nuts. We have hard enough time efficiently dividing a problem in 4 parts.
Click to expand...

That depends entirely on the application you're writing; consider a task like rendering which is separable into millions of independent calculations.

Also note that there are different methods of parallelizing a task. I spent some time last year programming on a 336 core processor. It uses a pipelined architecture which is unlike conventional CPUs.
 
jimmyb said:
That depends entirely on the application you're writing; consider a task like rendering which is separable into millions of independent calculations.

Also note that there are different methods of parallelizing a task. I spent some time last year programming on a 336 core processor. It uses a pipelined architecture which is unlike conventional CPUs.
Click to expand...

Rendering, network processing, cloud computing. There are lots of applications that something like this can be great for. It isn't designed for desktop usage (just ask for a quote on one of their current 64 core chips ... I am willing to bet it will run you thousands of dollars).
 
I have to ask the question then that if this cpu is going to cost thousands of dollars is it really going to be that much quicker then dropping all the money into a bunch of gulftown chips? This totally isn't my area to be dipping into, just curious what this thing would actually excel at.
 
SavageThrash said:
I have to ask the question then that if this cpu is going to cost thousands of dollars is it really going to be that much quicker then dropping all the money into a bunch of gulftown chips? This totally isn't my area to be dipping into, just curious what this thing would actually excel at.
Click to expand...

Performance / watt is probably stellar compared to running a large cluster of Intel boxes, for applications where this processor excels. It likely doesn't have much I/O capability, and probably requires fairly tight cache coherence to perform well, but I imagine for tasks where these limitations aren't relevant it can do very, very well. It should also be quite useful where you have a large number of streams that you'd like to process simultaneously, or where there are lots of realtime threads running at once. For example flow classification on an enterprise-scale router/content filter, or perhaps some large scale control systems and the like. It's also physically small, so you can pack lots of them in a small space.

I think for it to be competitive though you'd really have to design your application to the chip, but where it works it's probably considerably cheaper than the alternative you'd need for some of these things (standard CPU + ASIC, for example).
 
The issue is simple.. follow my advice precisely

1.) Get a flux capacitor
2.) open it up, cut the wire from the bottom of the "Y"
3.) follow the wire back to a battery block.
4.) Remove the 9 volt battery
5.) cut the black wire from the battery box
6.) find some 746h3l2-12 quick discharge batteries (google is your friend). Make sure you get the -12 version since the -11 version will only deliver 11 million volts. 12 million volts is required at the minimum

7.) Since this will be a one shot deal, you can just twist the positive wire from the 746h3l2-12 to the wire you snipped from the flux capacitor and then twist the ground wires (black is ground)
8.) Power that baby up, you will now be in [H] mode. If not, be sure you don't have your wires crossed.
9.) Program that baby to 2050, roughly from today's date and "peddle to the metal"

Once you hit 2050, reply to thread and let me know how funny this is.
 
drescherjm said:
I do not see this being feasible any time soon. Being a programmer as a profession I know that most computations can not be divided into 100 parts. That's absolutely nuts. We have hard enough time efficiently dividing a problem in 4 parts.
Click to expand...

But for computations that can be divided into multiple parts WITHOUT using locks, it could massively speed those computations up, even if you have to add some extra calculations in order to divide the computation into multiple parts.

And for large computations, it is not quite as difficult as you think. You just have to start thinking differently. Try things that are not generally accepted practice.

This is why the program I am currently working on supports pretty much whatever number of threads and work units you want to use. The only real limiting factor is the size of those variables. If I wanted, I could simply change those two variables to 64-bit unsinged ints and it would support 18,446,744,073,709,551,615 threads and work units.
 
Obviously this isn't relevant to this CPU, but I just realized that ~88 Nehalem cores (@2.66ghz) equal a GTX 280 in raw Gflops. That's pretty cool! Normalized for clock speed Nehalem provides 1.36x Gflops/core. That doesn't sound like a lot, but consider that half of each core is L2 + L3 and it provides OOO execution. I'd be interested in seeing Intel develop a GPU from the ground up. I wonder what kind of density gains we would see, and what kind of clocks we could get with High-k + metal gates! AFAIK SOI incurs a density cost due to the extra insulator layer, and Intel avoids that. Could be wrong on this.
 
Back when Intel was churning out P4s, AMDs SOI (with the aide of IBM) was a superior process however since Intel introduced the core2 that all changed. Intel drastically improved its transistors.
 
I thought the highest was now 6 cores, I really need to get back on track with CPU's. Thanks for the article it was a very enjoyable read.
 
You must log in or register to reply here.
Back
Top