mersenneforum.org  

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

Reply
 
Thread Tools
Old 2005-09-04, 15:51   #12
moo
 
moo's Avatar
 
Jul 2004
Nowhere

809 Posts
Default

Pentium ds are droping in price george. http://www.tigerdirect.com/applicati...?EdpNo=1293094
99 dollars for 2.93 celeron D.
moo is offline   Reply With Quote
Old 2005-09-04, 16:58   #13
PhilF
 
PhilF's Avatar
 
"6800 descendent"
Feb 2005
Colorado

2·11·31 Posts
Default

Pentium D's and Celeron D's are two very different animals. The Pentium D is dual core, the Celeron D is not.
PhilF is offline   Reply With Quote
Old 2005-09-05, 08:10   #14
T.Rex
 
T.Rex's Avatar
 
Feb 2004
France

2×461 Posts
Default Cost of multi-threading

Quote:
Originally Posted by Prime95
However, it is probably the case that testing two different exponents will have more throughput.
Sure. Managing the parallelism between 2 or more threads of computation has a cost. Nevertheless, in certain circonstances (like trying to quickly check an exponent), it may be acceptable to have a lower global throughput. Also, running the FFT of 1 exponent on 2 cores on the same CPU may be a good idea for optimizing the common caches between the 2 cores. Is there around experts that may provide their opinion if this could reduce the cost to something acceptable ?
GLucas uses POSIX threads (NPTL on Linux): mutex, cond-var and thread objects Since mprime is written mainly in asm, it may be worth to not use the Native POSIX library (which implies costly context-switches in the kernel). Using a mprime proprietary user-land management of locks and threads may really reduce the cost. But it should require different skills on Windows and Linux. I'm aware of such tools but have no real skills.
Regards,
Tony
T.Rex is offline   Reply With Quote
Old 2005-09-05, 15:18   #15
moo
 
moo's Avatar
 
Jul 2004
Nowhere

14518 Posts
Default

opps sorry bought that. Forgot they changed cores again. Shakes fist at all the core changeing going on.
moo is offline   Reply With Quote
Old 2005-09-10, 22:41   #16
georgekh
 
georgekh's Avatar
 
Oct 2004

2×33 Posts
Default

it will be interesting to see what the new 65nm will bring but as of now i'm really liking what Intel is doing with there multi-core lineup vs AMDs garbage
georgekh is offline   Reply With Quote
Old 2005-09-11, 12:24   #17
Dresdenboy
 
Dresdenboy's Avatar
 
Apr 2003
Berlin, Germany

192 Posts
Default

Quote:
Originally Posted by georgekh
it will be interesting to see what the new 65nm will bring but as of now i'm really liking what Intel is doing with there multi-core lineup vs AMDs garbage
What did you read about AMDs cores, which will show up in 2006 and 2007 (the same timeframe as for Intel's lineup) that you think it is garbage? I'm interested in any news.
Dresdenboy is offline   Reply With Quote
Old 2005-09-11, 19:16   #18
Dresdenboy
 
Dresdenboy's Avatar
 
Apr 2003
Berlin, Germany

36110 Posts
Default

There is a new article on Conroe and the changes it may bring to us:
http://www.digit-life.com/articles2/cpu/p6-nexgen.html
Dresdenboy is offline   Reply With Quote
Old 2005-09-13, 04:02   #19
georgekh
 
georgekh's Avatar
 
Oct 2004

2×33 Posts
Default

Overall i think Intel chips are better in the name of science and math whereas AMD is good for gaming so I like to think of the future and i think Intel is a much better option then AMD

PS: i'm also a clock speed freak so that also makes Intel a better choice for me
georgekh is offline   Reply With Quote
Old 2005-09-13, 13:28   #20
Fusion_power
 
Fusion_power's Avatar
 
Aug 2003
Snicker, AL

7×137 Posts
Default

A man growing pecans on an acre of ground does not care one whit how much an individual tree yields. What he cares about is how many pounds of pecans are produced on the entire acre. If the trees overbear, he loses the crop because they are not filled out properly. If they underbear, the quality is great, but production cost exceeds the value of the crop. At about 2200 pounds per acre, there is a sweet spot where production is consistent year to year, quality is maximum, and market value yields the maximum profit.

How fast a processor runs is not directly proportional to total throughput. At one time, it was. Today, multiprocessor cores and the various reduced cycle cores mean that a slower clock speed may yield significantly more throughput. Before you get on the clock speed bandwagon, consider carefully if more clock speed actually translates to more computational throughput.

The theoretical limit at which a processor can function is about 6 ghz with technology available today. This would cause serious heat problems that could not be addressed by using fans alone. Consider that a 3 ghz chip can perform x amount of work, then a 6 ghz chip does 2x. What if you put 2 cores in the 3ghz chip and achieved @ the same 2x but without having to install a water cooling system? The cost just went down! What about doing the same computation with fewer clock cycles on a single core? Again, the cost for total throughput goes down.

I am NOT hung up on clock speed. I'm more interested in total computational throughput.

Fusion
Fusion_power is offline   Reply With Quote
Old 2005-09-13, 17:26   #21
ColdFury
 
ColdFury's Avatar
 
Aug 2002

14016 Posts
Default

Quote:
What if you put 2 cores in the 3ghz chip and achieved @ the same 2x but without having to install a water cooling system? The cost just went down! What about doing the same computation with fewer clock cycles on a single core? Again, the cost for total throughput goes down.
Using two cores will almost never give you a 2x increase in speed except in the most embarassingly parallel applications.

Adding clock speed is often the easier way to increase speed, since adding more functional units does not guarantee any speed increase. There's inherent limits to the amount of parallel computation possible in a program, so all the added functional units end up getting unused most of the time. This is one of the reasons for HyperThreading.
ColdFury is offline   Reply With Quote
Old 2005-09-13, 20:30   #22
Dresdenboy
 
Dresdenboy's Avatar
 
Apr 2003
Berlin, Germany

1011010012 Posts
Default

Quote:
Originally Posted by georgekh
Overall i think Intel chips are better in the name of science and math whereas AMD is good for gaming so I like to think of the future and i think Intel is a much better option then AMD

PS: i'm also a clock speed freak so that also makes Intel a better choice for me
It differs. Intel chip != Intel chip. IA64 is good for science, that is true. Xeon is ok for that, but using the latest core (Nocona) it doesn't deliver much for the power it consumes. That's why there exist Blades using Banias or Dothan, which are not on par (FPU performance wise) with Xeon (IA64 is a different league), but deliver much more per watt than Xeon even then. That's why Intel is now going this way for future server CPUs. That's the reason why IBM build Blue gene super computers with 700 MHz CPUs - because they have a good performance/watt ratio.

Except for the dual core opterons the current AMD offers for HPC are indeed not that interesting for many system buyers. The standard server business is a different story. However, I'd like to read at least somewhat objective comments The world is not black/white.

The next good x86 CPU we'll see, I think, is Merom with it's increased and improved capabilities compared to Dothan. And the next good x86 CPU following it could be the formerly called "K10". Nobody knows, except for a few patents, and a few CPU guys, who switched between different CPU manufacturers.
Dresdenboy is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Aouessare-El Haddouchi-Essaaidi "test": "if Mp has no factor, it is prime!" wildrabbitt Miscellaneous Math 11 2015-03-06 08:17
"Sandy Bridge" is coming out today ixfd64 Hardware 8 2011-01-07 00:07
Intel i7 ("Nehalem") chips launched ixfd64 Hardware 34 2008-11-25 18:22
Would Minimizing "iterations between results file" may reveal "is not prime" earlier? nitai1999 Software 7 2004-08-26 18:12

All times are UTC. The time now is 05:28.


Mon Dec 6 05:28:40 UTC 2021 up 135 days, 23:57, 0 users, load averages: 1.10, 1.35, 1.54

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, 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.