mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Hardware

Reply
 
Thread Tools
Old 2006-02-06, 23:50   #1
jasong
 
jasong's Avatar
 
"Jason Goatcher"
Mar 2005

5·701 Posts
Default My prediction about multi-core chips

My mind doesn't work like other people's, so when I have an opinion I have no idea how others will feel about it. Anyway, I've been thinking about this whole multi-core thing, and I'd like to make a betrig prediction. Maybe others have come to the same conclusion and haven't posted about it, but...

I predict that within the next 1-4 years, chips will continue to increase in number of cores, but the cores will no longer simply be copies of each other, but instead will be developed for special tasks, like video editing and encoding for instance.

Opinions? Flames? Comments that have a tenuous connection to processed meat?
jasong is offline   Reply With Quote
Old 2006-02-07, 02:21   #2
moo
 
moo's Avatar
 
Jul 2004
Nowhere

809 Posts
Default

hate to burst your bub but that would be bad for us. We would be back to square one because if each core was different odds are only one would be sutiable for are needs.
moo is offline   Reply With Quote
Old 2006-02-07, 03:32   #3
thechickenman
 
thechickenman's Avatar
 
Nov 2005
South Carolina

4D16 Posts
Default

Quote:
Originally Posted by moo
hate to burst your bub but that would be bad for us. We would be back to square one because if each core was different odds are only one would be sutiable for are needs.
For GIMPS and every other DC project... yeah..

But we're not Intel's or AMD's target consumer here ;)

Don't know about the technical aspects, but it sounds reasonable to me.
thechickenman is offline   Reply With Quote
Old 2006-02-07, 05:41   #4
moo
 
moo's Avatar
 
Jul 2004
Nowhere

32916 Posts
Default

im not sure that would be a waste of cores because it doesnt make a good sales pitch we have 2 cores one does video the other does everything else when they could put a lot of cores that do everything...
moo is offline   Reply With Quote
Old 2006-02-07, 10:00   #5
Peter Nelson
 
Peter Nelson's Avatar
 
Oct 2004

232 Posts
Default

I think this is entirely likely.

eg we already have specialisation like part of some cpus is normal processor, part is FPU for maths, part is on-die memory controller.

I don't think it unreasonable that some specialisation like eg physics processor core might get there one day but is more likely to get put into the gpu.

For cpu, I think a dedicated FFT core would be nice (would help Gimps a lot).

Some specialisation is the key to good performance, otherwise you just have chips which are mediocre at everything.

Take for example Xilinx spartan fpga chip has general logic, but also fixed multipliers for that purpose.

Xilinx Virtex 4 can have say four cores of powerpc, plus some reconfigurable logic, maybe DSP blocks, fast RAM blocks, high speed networking like multiple 2.5 or 10 gigabit links. That's in one chip, so why not?

Moving the networking into the processor could work (some have already done that for ethernet MAC just needing external PHY and connector). Networking can be accellerated by eg checksum calculation and generation/checking, TCP segmentation. This can all be done in hardware without burdoning general purpose processors.

In a way, hypertransport does this already, taking highspeed interconnect right onto the processor.

Some things could be done with a complementary co-processor, but for things needing optimised bandwidth to the main cpu, moving them onboard the main chip will make it fast, and is less headache to route to limited number of package pins, and noise by motherboard trace/distance.

I think Jason has shown why IBM Cell processor does not have 9 identical cores, but one big core and 8 synergistic little cores around it. They are making more efficient use of the available logic gates than say a second big core would make to performance.

So Jason, your idea already exists in the real world!

Personally I'd like to see the big processors include some programmable logic gate array which they could license from Xilinx, Altera etc. That way it would not suffer from eg blocks for video editing being useless for spreadsheets and vice versa. With a programmable logic you can load in the functionality suitable to accelerate your current application, without being locked into that.

Last fiddled with by Peter Nelson on 2006-02-07 at 10:01
Peter Nelson is offline   Reply With Quote
Old 2006-02-07, 16:45   #6
jinydu
 
jinydu's Avatar
 
Dec 2003
Hopefully Near M48

2·3·293 Posts
Default

I found this news article (admittedly, it is not really on-topic...)

http://news.yahoo.com/s/ap/20060207/...MwBHNlYwM3Mzg-
jinydu is offline   Reply With Quote
Old 2006-02-14, 16:07   #7
Greenbank
 
Greenbank's Avatar
 
Jul 2005

2×193 Posts
Default

Sun have been producing 8-core processors for 3 months now. The T1 is a complete rethink of processor structure with more emphasis on multiple simple cores that will boost the speed of heavily threaded processes.

This is bad for GIMPS as there is only 1 FP unit per die and the amount of cache is much smaller than standard processors.

http://en.wikipedia.org/wiki/UltraSPARC_T1 has more info.
Greenbank is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Intel announces multi-core enhancements for Haswell chips ixfd64 Hardware 8 2012-02-10 20:32
multi-core TF dbaugh Information & Answers 3 2011-09-06 01:32
GMP-Fermat and multi-core processors ixfd64 Factoring 1 2011-03-16 16:07
Prime95 and future multi-core chips ixfd64 Software 4 2011-01-02 18:34
Multi-Core / Multi-CPU Assignments (missing) worknplay Software 3 2008-11-05 17:26

All times are UTC. The time now is 21:41.

Thu Oct 22 21:41:45 UTC 2020 up 42 days, 18:52, 0 users, load averages: 1.98, 2.08, 2.00

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.