| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Community
Forum: Teams
Thread:  Join "MyOnlineTeam" Today - Chapter 45 |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 580
|
|
| Author |
|
|
keithhenry
Ace Cruncher Senile old farts of the world ....uh.....uh..... nevermind Joined: Nov 18, 2004 Post Count: 18665 Status: Offline Project Badges: 
|
Keith, Gerald and RT See my Sig  Throw in my netbook and I's a cooking  Just 3 machines still in the game Keith - for 6.25KWh/day !! Well, I am impressed! Those last 200K+ points seemed to come awfully fast. It's good to have you back guy! |
|||
|
||||
|
keithhenry
Ace Cruncher Senile old farts of the world ....uh.....uh..... nevermind Joined: Nov 18, 2004 Post Count: 18665 Status: Offline Project Badges:
|
Man, I wish WCG would send out newsletters more often! We have pramodp back in our active ranks again! Would love to hear from them and imdioxin both! Looks like we're at least two out of three on this last newsletter!
---------------------------------------- |
|||
|
||||
|
keithhenry
Ace Cruncher Senile old farts of the world ....uh.....uh..... nevermind Joined: Nov 18, 2004 Post Count: 18665 Status: Offline Project Badges:
|
in memory of Delores, thanks for browsing our thread. Many of us here have our own "Delores". If you're looking for a team, we'd love to have you here.
---------------------------------------- |
|||
|
||||
|
keithhenry
Ace Cruncher Senile old farts of the world ....uh.....uh..... nevermind Joined: Nov 18, 2004 Post Count: 18665 Status: Offline Project Badges:
|
Good Morning MOT, and all visitors. This discussion about processors and which are the fastest for WCG kinds of problems as well as which are the most efficient per unit of power used..has me confused again. Some years back it was generally thought that the AMD processors gave you more compute power (FLOPS at least) (Floating Point Operations per Second). Then Intel seemed to come off the faster clock speed kick and make more of the Floating point operations, and some scalar operations, take fewer clock cycles thereby increasing the throughput of their chips without increases in clock cycles. Further they increased the on-chip cache, added Hyper-Threading and turbo-boost and of course lowered the line width and the resultant power consumption. Now at least for some time, Intel seemed to have the upper hand when it came to FLOPS (which the WCG workload has a high percentage of). Now folks are again talking about how fast the AMD machines are and their multi-core etc. Things are as confused as ever for me. I am an old throughput kind of a guy. I like to listen to the discussions on memory fetches and translate look-aside buffers etc. etc. But at the end of the day, I want to watch the contending technologies run batches of work that are reasonably typical and then make my choice. I just wish there were a way we could do that and remove the FUD (Fear Uncertainty and Doubt). I started this post in the morning but have in between paragraphs fixed two different computers - neither of them mine. So I guess it is Good Afternoon MOT. Edit> RT, I can relate to wanting to get the most crunching out of what machines I have but trying to figure out the best way to squeeze every last drop of virtual blood from the turnip tends to make my head hurt. Since every project here has its own science app, it seems like each one could behave differently and so the "perfect" cruncher could change with each new project. I think doing what crunching you can is what matters the most. If I get an opportunity to get a new machine, I'll try to get the best* I can get at the time but I'll leave the "pink-ing" to the experts. * for me, a fast CPU with as many affordable cores as possible, a respectable hard drive and, these days, at least 3MB if not 4MB of memory. |
|||
|
||||
|
keithhenry
Ace Cruncher Senile old farts of the world ....uh.....uh..... nevermind Joined: Nov 18, 2004 Post Count: 18665 Status: Offline Project Badges:
|
It figures ... I manage to set a personal WCG points record today and all of the sudden the forecast is for highs around 90F - 100F for the next few days. Some of the machines can't handle 100% in the heat, so my point production will fall off for a few days  unless all those PV WUs magically validate. Crunch on! - D P.S. @DM - thanks for stopping by. One for the no good deed goes unpunished book it seems. We're getting ready for some real fall weather here at long last. It'll make it there in time too. You've had a lot of nice days lately. Well done! |
|||
|
||||
|
darth_vader
Veteran Cruncher A galaxy far, far away... Joined: Jul 13, 2005 Post Count: 514 Status: Offline Project Badges: 
|
It figures ... I manage to set a personal WCG points record today and all of the sudden the forecast is for highs around 90F - 100F for the next few days. Some of the machines can't handle 100% in the heat, so my point production will fall off for a few days unless all those PV WUs magically validate. Crunch on! - D P.S. @DM - thanks for stopping by. One for the no good deed goes unpunished book it seems. We're getting ready for some real fall weather here at long last. It'll make it there in time too. You've had a lot of nice days lately. Well done! - D |
|||
|
||||
|
RT
Master Cruncher USA - Texas - DFW Joined: Dec 22, 2004 Post Count: 2636 Status: Offline Project Badges: 
|
Good Morning MOT, and all visitors. This discussion about processors and which are the fastest for WCG kinds of problems as well as which are the most efficient per unit of power used..has me confused again. Some years back it was generally thought that the AMD processors gave you more compute power (FLOPS at least) (Floating Point Operations per Second). Then Intel seemed to come off the faster clock speed kick and make more of the Floating point operations, and some scalar operations, take fewer clock cycles thereby increasing the throughput of their chips without increases in clock cycles. Further they increased the on-chip cache, added Hyper-Threading and turbo-boost and of course lowered the line width and the resultant power consumption. Now at least for some time, Intel seemed to have the upper hand when it came to FLOPS (which the WCG workload has a high percentage of). Now folks are again talking about how fast the AMD machines are and their multi-core etc. Things are as confused as ever for me. I am an old throughput kind of a guy. I like to listen to the discussions on memory fetches and translate look-aside buffers etc. etc. But at the end of the day, I want to watch the contending technologies run batches of work that are reasonably typical and then make my choice. I just wish there were a way we could do that and remove the FUD (Fear Uncertainty and Doubt). I started this post in the morning but have in between paragraphs fixed two different computers - neither of them mine. So I guess it is Good Afternoon MOT. Edit> RT, I can relate to wanting to get the most crunching out of what machines I have but trying to figure out the best way to squeeze every last drop of virtual blood from the turnip tends to make my head hurt. Since every project here has its own science app, it seems like each one could behave differently and so the "perfect" cruncher could change with each new project. I think doing what crunching you can is what matters the most. If I get an opportunity to get a new machine, I'll try to get the best* I can get at the time but I'll leave the "pink-ing" to the experts. * for me, a fast CPU with as many affordable cores as possible, a respectable hard drive and, these days, at least 3MB if not 4MB of memory. I suppose.. But, the way I understand the nature of these science projects, they have a relatively high Floating Point component. That being the case, the CPUs that do floating point best seem to me to be the most productive. Now as to cost...I have a somewhat different perspective on that. It seems to me that it makes sense go up the price performance curve a bit beyond the "knee" in the curve due to the fact that we run these puppies 24x7 at 100%; so the cost difference per hour is infinitesimal. I am currently trying to justify a new build with a Gigabyte MB and an Intel I7/930. Of course there are faster Intels with more cores but they seem to be well beyond the aforementioned price/performance curve knee. It bothers me a bit that I don't know the relative FLOPS throughput of the AMD processors though and whether they perform along the same price/performance curve or ... while having higher clock speeds, consume them less efficiently thereby not producing the throughput of slower clocked I7s. Add to that David A's comments about the electrical consumption and it seems way too complicated to be confident in any decision that you make. So, Dubby...you asked if anyone has comments... I wish I had some sound advice...but I don't....Good Luck Old Friend |
|||
|
||||
|
darth_vader
Veteran Cruncher A galaxy far, far away... Joined: Jul 13, 2005 Post Count: 514 Status: Offline Project Badges:
|
* for me, a fast CPU with as many affordable cores as possible, a respectable hard drive and, these days, at least 3MB if not 4MB of memory. I think you mean 3GB if not 4GB. Windows can't even think about booting with 4MB. With the current versions of Windows, it seems 2GB is the practical minimum and for crunching you should add around .5 GB for each thread you're going to run. And finally, you need 64-bit Windows if you want to use more than 3GB plus a little (Windows 32-bit can't use the full 4GB). I suspect you know all this already. RT's discussion about what's best touches on several points. First of all a hypertreaded core is never going to produce the same amount of work as 2 cores. Remember that hyperthreading only exists because of the disparity between cache access and main memory access. It makes it difficult to do apples-to-apples comparisons even for the same vendor. There are all sorts of permutations of ALU / CPU / FPU between processors that means the best processor for one project may not be the best for others. Edit: Just to be clear, I mean that 2 CPU cores will out-perform a similar single hyperthreaded core running 2 threads. - D [Edit 1 times, last edit by darth_vader at Sep 26, 2010 4:08:41 AM] |
|||
|
||||
|
Somervillejudson@netscape.net
Veteran Cruncher USA Joined: May 16, 2008 Post Count: 1065 Status: Offline Project Badges:
|
Cost is certainly an issue and when you look at the high cost difference between the mid and high speed i7's there is quite a difference. Currently building a new machine but a top of the line i7 is at least 3-4 k. Not cheep. Servers with dual processors almost seems to be a better choice.
|
|||
|
||||
|
darth_vader
Veteran Cruncher A galaxy far, far away... Joined: Jul 13, 2005 Post Count: 514 Status: Offline Project Badges:
|
The cost of the "extreme edition" processors has never been worth it. They are way above the knee in the performance curve and are priced all out of proportion to the additional, relatively small, increase in processing power.
----------------------------------------AMD vs. Intel at the high end currently favors Intel. Intel processors tend to get more work done at equivalent processor speeds and the advantage that AMD used to have with their integrated memory controller is now gone as well. However, AMD's soon-to-be-announced processors, currently codenamed bobcat and bulldozer, may change things when available later this year or early next year. At the mid to low end, AMD is a bit ahead price / performance, but Intel will consume less power for equivalent performance. Add to this the fact that Intel intentionally cripples the low and mid range so as not to compete with their top end and thereby leaves more of an opening for AMD. AMD has been lagging ever since C2D came out. Maybe they will close the gap, or almost close the gap. Serious competition at the high end will be good for all of us, but you have to remember that the high end is where the manufacturers expect high profits. Thus, the knee in the curve exists..... Trying to sort all of this out is difficult and changes every couple of months. The marketing by the processor makers doesn't help matters. - D [Edit 1 times, last edit by darth_vader at Sep 26, 2010 4:29:21 AM] |
|||
|
||||
|
|