| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: Suggestions / Feedback
Thread: Quadcores, Hexacores etc... An opportunity for better science |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 11
|
|
| Author |
|
|
Hypernova
Master Cruncher Audaces Fortuna Juvat ! Vaud - Switzerland Joined: Dec 16, 2008 Post Count: 1908 Status: Offline Project Badges: 
|
There has been here and there comments about the "problem" of all these new CPU's becoming more and more common with crunchers.
----------------------------------------It means that the average computing power is or will increase. Some have commented that this may have a negative impact on WU's availability to reach high scores as there would be less WU's available and so projects may finish to early the big crunchers gobbling up everything. In fact there are two growth opportunities. New members and new CPU's. These two elements fortunately add up. I see this improvement in the overall computing power as a great opportunity to, increase the resolution (granulation, precision etc.), enlarge the scope in terms of number of items to be crossed, pattern matched, of the scientific analysis that is being done. So I hope that the scientists will integrate this CPU power evolution to take maximum profit of it. There will be never enough CPU power if the size of the problem is well tailored. I suppose that when WCG started 6 years go the global available crunching power was much lower. I wonder if projects over the time did already take this into account and how. One day maybe quantum computers will completely push us to rethink the whole thing as they are estimated to be more then 10'000 times faster on certain computations compared to the fastest computation unit. But this is another story.  |
|||
|
||||
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Problem is, the lowest common denominator determines how well optimized the science applications run... the more optimized, the more the not so new
----------------------------------------
WCG
----------------------------------------Please help to make the Forums an enjoyable experience for All! [Edit 1 times, last edit by Sekerob at Jan 9, 2010 10:36:40 AM] |
||
|
|
KerSamson
Master Cruncher Switzerland Joined: Jan 29, 2007 Post Count: 1684 Status: Offline Project Badges: 
|
Hello Hypernova,
----------------------------------------you raise several aspects which are impacting significantly the WCG projects.
Taking a look on technology evolutions (e.g. hexcore), I would enjoy to not have to know any budget limitation and to be able to setup such powerful systems for grid computing purposes ! ... (who would not do this ?). Have a nice week-end, Yves |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Since no device is the same v.v. a multicore and what resources it has available for crunching and what else it's doing during use, it's nearly impossible to give advises on what produces best results in terms of efficiency (several prior threads touch on this). It's though above a shadow of doubt that running a quad on just a single project, (heavy duty in particular) is not as efficient as mixing heavy with light or one using the SSE features and the other the SSE2. E.g. running a mix of RICE and HCMD2 or RICE and HFCC/FAAH does better, on my setup, than just 4 RICE or 4 HFCC or FAAH... a question of demands and bottlenecks. A quad may have 4 physical cores but the rest is most likely not. You'll have to do the hands on to find what's best for you (to include combinations with projects outside WCG).
----------------------------------------As to devices profiles, officially they'll remain 4, unsupported, you can make as many as you like, at WCG, and name them anything you like.
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
Hypernova
Master Cruncher Audaces Fortuna Juvat ! Vaud - Switzerland Joined: Dec 16, 2008 Post Count: 1908 Status: Offline Project Badges:
|
Yves just a small remark concerning your last comment on budgetary limitations. After all we may not be in reality so much limited. It is again a matter of priorities.
----------------------------------------Here is a small exercise for smokers but which can be applied to other types of regular small spending, like alcohol, gambling etc. If I take a 5 US$ price per pack of 20 cigarettes. Prices vary a lot across the world but an average European price would be around 8 US$. Anyway let's have 5 US$&pack. A two packs per day smoker will spend 10 US$ per day. In one year he spends 3'650 US$. On an average lifetime of a PC of three years it's over  10'000 US$ In Europe this budget would stand at 17'500 US$ and in some european countries you would top easily 20'000 US$. I never smoked, so imagine in my 50 years of life with added interest I spared way over 1'000'000 US$. Out of this budget a few hexacores is peanuts.  Do not take the numbers as a scientific precise calculation, but simply as a way to illustrate a concept that small drops do make rivers, seas and oceans. |
|||
|
||||
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hyper ... that is exactly what I did to justify buying my second i7. I certainly am much healthier and the added electric expense to run the second unit is easily absorbed by my non-smoking savings. I have a formula that is based on my prior smoking habits that do not quite add up the same as you calcs but is still enough that I think I will replace my i7-920s with hexcores assuming they will have low-end skus available for about 500-600 sometime around March - April. I look at it this way, for ~ $1000 USD I will be adding the crunching equivalent of another i7-920 but using the same power envelope. It would cost almost as much if I bought all the components and built a nerw system myself. Besides not incurring any added electic costs I also will not have to try and figure out how to recycle anything except the two CPUs. I am sure I will be able to find another cruncher that will be looking for a bargain on chips :-)
|
||
|
|
Richard Mitnick
Veteran Cruncher USA Joined: Feb 28, 2007 Post Count: 583 Status: Offline Project Badges: 
|
I am sure that I am technically way out of my league here, so I will not comment on any of the technical aspects of this discussion of some of the new hot machinery.
----------------------------------------As Sek and some others know, I was just recently involved with setting up my new Win7 64bit I7-920 hyper-threaded CPU machine. It took about a month. With tremendous help from WCG CA's, esp Sek, I tried many different pref settings. I have this machine on all WCG projects, weighted at 200, and I have four other non-WCG projects with std 100 weights using BOINC software. I have GPU processing not available. At standard prefs, I was burning through stuff so fast that the machine would run out of work. I got happy finally at Connect every =0.30 and Additional work = 2.0. So, that is the history. My point here is that these machines are becoming rather ordinary. My machine is not some US$10000 super box. I am not in this for points or credits. I simply find it very satisfying to use my machines for some good besides work, music, video and the like. I am also running two Core 2 Duo machines and a little netbook with an Atom N270 CPU. Before the new I7 machine, I was averaging maybe 1000 credits per day. Now, I am averaging 3400 credits per day. This is the new reality. We can only hope that the scientists are able to recognize the new power, GPU along with CPU, that is becoming available,and make use of it. |
||
|
|
KerSamson
Master Cruncher Switzerland Joined: Jan 29, 2007 Post Count: 1684 Status: Offline Project Badges:
|
Hi,
----------------------------------------Thank you for your remarks regarding possible resources ! ... But I did never smoke Cheers, Yves |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Since no device is the same v.v. a multicore and what resources it has available for crunching and what else it's doing during use, it's nearly impossible to give advises on what produces best results in terms of efficiency (several prior threads touch on this). It's though above a shadow of doubt that running a quad on just a single project, (heavy duty in particular) is not as efficient as mixing heavy with light or one using the SSE features and the other the SSE2. E.g. running a mix of RICE and HCMD2 or RICE and HFCC/FAAH does better, on my setup, than just 4 RICE or 4 HFCC or FAAH... a question of demands and bottlenecks. A quad may have 4 physical cores but the rest is most likely not. You'll have to do the hands on to find what's best for you (to include combinations with projects outside WCG). As to devices profiles, officially they'll remain 4, unsupported, you can make as many as you like, at WCG, and name them anything you like. Here's an illustration of RICE. You may derive which ran with 4 concurrent RICE jobs for near full time and which ran as 3 concurrent with a 4th job (HCMD2), micromanaged in (manually suspend/resume). Tip is the top one ran overnight with 4 concurrent RICE for the full run time as did some others further down. R00504_ 7e6a6b45f23cae92cb6ffaf20634407b_ 02_ 004_ 7-- Valid 1/8/10 01:46:58 1/12/10 07:02:23 7.01 129.6 / 115.5 R00504_ 7e6a6b45f23cae92cb6ffaf20634407b_ 03_ 004_ 13-- Valid 1/8/10 01:46:58 1/12/10 03:02:27 7.00 129.5 / 148.6 R00504_ 7fcd5396cab3281fcb972ee4508a67ee_ 00_ 003_ 16-- Valid 1/8/10 01:46:58 1/12/10 02:09:58 7.01 129.6 / 138.8 R00504_ 7fcd5396cab3281fcb972ee4508a67ee_ 00_ 004_ 6-- Valid 1/8/10 01:46:58 1/12/10 01:34:33 7.03 130.0 / 128.4 R00504_ 7fcd5396cab3281fcb972ee4508a67ee_ 00_ 005_ 7-- Valid 1/8/10 01:46:58 1/11/10 20:00:00 7.02 129.8 / 120.0 R00504_ 8d1fcde105357752e7ff214b3c60e021_ 01_ 002_ 0-- Valid 1/8/10 01:46:58 1/11/10 19:07:19 7.01 129.6 / 142.7 R00504_ 7244b40129ac6487a8ddb1bba864bf39_ 01_ 005_ 17-- Valid 1/8/10 01:46:42 1/11/10 19:01:21 7.02 129.7 / 114.4 R00504_ 7244b40129ac6487a8ddb1bba864bf39_ 02_ 000_ 5-- Valid 1/8/10 01:46:42 1/11/10 15:14:23 7.01 129.5 / 119.1 R00504_ 72552a56d91e0f1e432682d64172a710_ 03_ 000_ 17-- Valid 1/8/10 01:46:42 1/11/10 12:35:05 7.00 129.4 / 127.0 R00504_ 7a9b4dc297d952a47fa91d984e35527f_ 03_ 000_ 13-- Valid 1/8/10 01:46:42 1/11/10 11:33:53 7.01 129.5 / 134.3 R00504_ 7e4199b2cb1aabf54fcd2cd122ee25ba_ 03_ 000_ 14-- Valid 1/8/10 01:46:42 1/11/10 09:36:00 7.01 129.6 / 123.4 R00504_ 7e627d555ff4247ac2e10a294d90f160_ 02_ 003_ 17-- Valid 1/8/10 01:46:42 1/11/10 05:55:26 7.01 129.5 / 122.5 R00504_ 7e6a6b45f23cae92cb6ffaf20634407b_ 02_ 001_ 6-- Valid 1/8/10 01:46:42 1/11/10 05:15:17 7.02 129.7 / 115.2 R00504_ 7e6a6b45f23cae92cb6ffaf20634407b_ 03_ 000_ 4-- Valid 1/8/10 01:46:42 1/11/10 02:26:09 7.00 129.5 / 143.4 R00504_ 7fcd5396cab3281fcb972ee4508a67ee_ 00_ 009_ 15-- Valid 1/8/10 01:46:40 1/10/10 22:52:08 7.00 129.5 / 135.3 Those with low award actually have less seeds completed on average per the result logs. Anyway, I just make sure to run a mix at all times on the quad.
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
Chief Gonzo
Cruncher Joined: Jul 3, 2008 Post Count: 15 Status: Offline Project Badges: 
|
Since the badges are awarded on the basis of computing time (hours, days, years) I don't see that, if getting badges is the goal, faster CPUs are the only consideration. I'm running an old laptop with a slow CPU and it generates the same hours as a faster CPU. Of course, dual and quad processors rack up the hours faster! And, as far as getting the research done, the faster CPUs do return more results and faster.
|
||
|
|
|