| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Community
Forum: Chat Room
Thread: K10 barcelona multicore float performance |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 6
|
|
| Author |
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
hi!
----------------------------------------I know that the K10 chip is comparatively slow while using just one core, but faster using all 4. Now i wondered what meets the reality of wcg most, single core performance (i always see 1 wu per core ^^) or overall-performance. Can the wcg apps make use of barcelonas' multi-core performance boost? [Edit 1 times, last edit by Former Member at Sep 30, 2007 12:21:05 AM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Welcome to the forum
---------------------------------------- The K10 chip is comparatively slow to what? 4 cores = 4 work units at once; it really is that simple. Even if there was only one core, and it was four times as fast, it would still get the same amount of work done. Or are you referring to the mythical "reverse hyperthreading"...? Although, even if you were the result would be the same. Or maybe I'm totally misunderstanding what you are talking about. [Edit 2 times, last edit by Former Member at Sep 30, 2007 12:45:05 AM] |
||
|
|
twilyth
Master Cruncher US Joined: Mar 30, 2007 Post Count: 2130 Status: Offline Project Badges: 
|
boinc can use all 4 cores of quad core processors, so one core of a quad core will do the same amount of work as a single core running at the same speed AND with the same architecture. But of course the architecture of the quads is almost by definition different from single cores.
----------------------------------------If you compare a quad opteron 2350 running at 2 ghz to an intel xenon 5335 running at the same speed, the opteron will do about a third more floating point operations per second - see Xbit SPEC comparisons. BUT - overall, the intel chips seem to be faster on most other measures of performance. So it's still an open question as to how well the barcelona chips would do on real WCG apps. One very detailed review originally posted by Dave Autumns compares Barcelona to Intel quad cores - see Techreport comparison. Page 7 does a comparison using several Folding@Home work units. From that, However you also have to look at the cost. Opteron is a server chip and commands a premium. You can get an Intel Q6600 rated for 2.4 ghz for about $270. The Opteron 2347 running at 1.9 ghz goes for about $330. There will be a lot more new chips from both companies in November. AMD will have the desktop version of the quad Opteron - Phenom. There are some intel launches too, but I'm still trying to sort out the different code names - Harpertown, Penryn, etc. I would wait until there are some benchmarks for WCG on the Barcelona. I haven't seen any yet, but I know they're coming. Check out XtremeSystems.org and look for the AMD thread.   [Edit 2 times, last edit by twilyth at Sep 30, 2007 2:10:22 AM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
thank you!
that was exactely what i meant, twilyth^^ i'm going to watch out for benchmarks using wcg apps...although it seems that they depend on single-core performance. maybe it could be an alternative to make one app crunching one wu with four threats simultaneously...but i don't think it'll happen. |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Think Didactylos mentioned somewhere that BOINC developers were working on enabling SMP in a future release. Here's a brief discussion with some other links going from there on the pro-con.
----------------------------------------Not sure I'd prefer on a quad for 1 job to go bad and 3 not being effected or jobs sequentially. It would also be interesting to see if a client is attached to a project mix of classic single core and SMP processes. Looking for heavy work-outs, crunching it is. Think either the AC@H or HPF2 are the most demanding at present.
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
twilyth
Master Cruncher US Joined: Mar 30, 2007 Post Count: 2130 Status: Offline Project Badges:
|
So you're saying you would want to process the same work unit on all 4 cores at the same time? If so, that's not necessary because in order to use the grid effectively, the work has already been broken down into parallel threads - that's what the work units are for. It's like recoding a video stream. You break the stream into 4 independent blocks and assign one block to each cpu core. There's no need to break each block down any further - you've already parallelized the computational process.
----------------------------------------What is your primary concern? Doing more work units or doing each WU as quickly as possible? Also, it's worth pointing out that there are architectural changes with the quad cores that make them inherently faster per core. For example, in the review I mentioned above, compare the performance of 2P (2 socket) Opteron 22xx systems with 23xx systems running at the same speed - there is better performance per core because the 23xx's have better logic, more cache, wider pipes, etc. If you're looking to build a new system and don't want to wait until Nov-Dec. I would get a Q6600 and a very good air cooler (I'm using a Zalman 9700) and overclock it. I've got mine running at 3.2 ghz (with help from olympic of team XtremeSystems) and it produces about 13,000 points and 35 results per day. You can build a complete system with 2gig ram, case, psu, m/b, hdd for about $800 - $900. I paid almost twice that a month ago for the same components - except I got 4 rather than 2gig of ram and reused an old atx case I had. |
||
|
|
|