| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Community
Forum: Chat Room
Thread: Which Linux is best? |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 54
|
|
| Author |
|
|
krakatuk
Advanced Cruncher Germany Joined: Oct 3, 2008 Post Count: 141 Status: Offline Project Badges: 
|
Nanoprobe, there must be really something wrong with performance here. My Q6600@3,4Ghz makes HCC WUs in 40-45min running on Dotsch 64bit from a USB stick.
----------------------------------------Here are my results: X0000105140084200811031034_ 1-- dotschux0022154 FF640 Valid 13.11.10 21:37:05 14.11.10 00:36:04 0.69 17.7 / 17.5 X0000107571401200902271007_ 0-- dotschux0022154 FF640 Valid 13.11.10 21:34:42 14.11.10 00:33:14 0.70 17.9 / 17.9 X0000107570494200902271020_ 0-- dotschux0022154 FF640 Valid 13.11.10 21:23:33 14.11.10 00:25:29 0.70 17.9 / 18.9 krakatuk  |
||
|
|
nanoprobe
Master Cruncher Classified Joined: Aug 29, 2008 Post Count: 2998 Status: Offline Project Badges: 
|
nano, what level of boinc do you have running? 16 pts per hr is kinda low for a 4.1 GHz processor. If you are running an older level of boinc (5.xxx) then that is the cause, there was some sort of bug where the benchmark would give only about 1/2 the points. I'm running 6.10.17 which comes as part of the install package with DotschUx. Are you running Linux 32-bit or Linux 64-bit on your Intel I7? Linux 64-bit is much more efficient, due to a new 64-bit compiler that is not restricted to the basic x86 instruction set but uses all the added SSE instructions that have been added over the last 2 decades. The old GCC 32-bit compiler was embarrassingly inefficient compared with many Windows compilers. The new Linux 64-bit compiler appears to be very efficient. DotschUx version 1.2 is 64 bit. I have tested Win7 Pro 64 on this particular i7 also and Linux always awards less points per unit of runtime verses Windows. The efficiency is there because Linux crunches faster than windows. Up to twice as fast on HCC and 50-60% faster on C4CW, it just doesn't award as many points which I don't really care about. # of WUs done is the bottom line for me. Nanoprobe, there must be really something wrong with performance here. My Q6600@3,4Ghz makes HCC WUs in 40-45min running on Dotsch 64bit from a USB stick. Here are my results: 40-45 minutes per WU looks pretty good to me if you're running @ 3.4 GHz.
In 1969 I took an oath to defend and protect the U S Constitution against all enemies, both foreign and Domestic. There was no expiration date.
  |
||
|
|
krakatuk
Advanced Cruncher Germany Joined: Oct 3, 2008 Post Count: 141 Status: Offline Project Badges:
|
I think I know what is the reason!
----------------------------------------You have HT and I have none, so every your real core does 2 WUs in less than an hour. It's faster then one WU in 40 min in my case!  |
||
|
|
anhhai
Veteran Cruncher Joined: Mar 22, 2005 Post Count: 839 Status: Offline Project Badges:
|
krakatuk, yeah nano has HT, that is the cause of the diff in performance.
----------------------------------------nano, my linux box also usually awards less pts per hr then when it was under windows. However, when I updated to the latest boinc (from 5.xxx to 6.10.56) the pts claim is within reason. I am claiming about 15-16 pts per hr and I only have a 2.6GHz intel processor. It is not over claiming since c4cw usually awards me more pts then I ask for. I mostly crunch c4cw on my linux box. It is about 50% faster on linux (3.2 hrs vs 4.7). Not sure how you and many others are getting 100% better performance on HCC, I only get about 50% (1.7hrs vs 2.6hrs) better performance there too. But like you said results are what matters, thats why I now have a few machines on linux instead of windows and they are crunching c4cw. I don't do c4cw on my windows machines since I know linux is faster.  |
||
|
|
nanoprobe
Master Cruncher Classified Joined: Aug 29, 2008 Post Count: 2998 Status: Offline Project Badges:
|
krakatuk, yeah nano has HT, that is the cause of the diff in performance. Yes, HT is enabled. nano, my linux box also usually awards less pts per hr then when it was under windows. However, when I updated to the latest boinc (from 5.xxx to 6.10.56) the pts claim is within reason. I am claiming about 15-16 pts per hr and I only have a 2.6GHz intel processor. It is not over claiming since c4cw usually awards me more pts then I ask for. I mostly crunch c4cw on my linux box. It is about 50% faster on linux (3.2 hrs vs 4.7). Not sure how you and many others are getting 100% better performance on HCC, I only get about 50% (1.7hrs vs 2.6hrs) better performance there too. But like you said results are what matters, thats why I now have a few machines on linux instead of windows and they are crunching c4cw. I don't do c4cw on my windows machines since I know linux is faster. The BOINC version that comes as part of the DotschUx install is 6.10.17. I'm not sure updating would make any difference and since I'm pretty much a Linux noob I would have no idea how to do that anyway. My average crunch times for HCC on windows were from 1 hour 55 minutes to 2 hours and 5 minutes. With Linux they are 57 minutes to 1 hour 5 minutes. Pretty much a 100% gain is because I'm running @ 4.1 GHz @1600HGz memory speed with the timings @ 7-8-7-24. You're actually getting a 65% gain @ 2.6 GHz which is very good IMHO. I have switched all of my dedicated crunchers to Linux and turned them loose on HCC just to see how many WUs could be finished. 805 yesterday and 5000+ for the week.  Even better than I imagined. I will add more projects to the mix in a few days after I'm done playing. 
In 1969 I took an oath to defend and protect the U S Constitution against all enemies, both foreign and Domestic. There was no expiration date.
|
||
|
|
kateiacy
Veteran Cruncher USA Joined: Jan 23, 2010 Post Count: 1027 Status: Offline Project Badges: 
|
nano, my linux box also usually awards less pts per hr then when it was under windows. ... C4CW is about 50% faster on linux (3.2 hrs vs 4.7). Not sure how you and many others are getting 100% better performance on HCC, I only get about 50% (1.7hrs vs 2.6hrs) better performance there too. But like you said results are what matters, thats why I now have a few machines on linux instead of windows and they are crunching c4cw. I don't do c4cw on my windows machines since I know linux is faster. This is something that has baffled me ever since I first read about it. Why is BOINC set up to reward crunchers for crunching more slowly (Windows) instead of using their equipment more efficiently? I would think it would be in the interest of speeding up the scientific results to give crunchers as much incentive as possible to crunch as fast as possible. What am I missing?  |
||
|
|
nanoprobe
Master Cruncher Classified Joined: Aug 29, 2008 Post Count: 2998 Status: Offline Project Badges:
|
What am I missing? I have no answer for that. From my observations I get basically the same amount of points for the amount of time I crunch whether it be in Windows or Linux. Linux gives me more WUs finished for the same amount of time and getting the units done is what I care about. It would be nice if the points awarded were weighted more toward the number of WUs completed but for whatever reason it doesn't seem to work that way.
In 1969 I took an oath to defend and protect the U S Constitution against all enemies, both foreign and Domestic. There was no expiration date.
|
||
|
|
anhhai
Veteran Cruncher Joined: Mar 22, 2005 Post Count: 839 Status: Offline Project Badges:
|
nano, you are getting about 16 pts per hr under linux. Are you saying you getting the same under windows? That doesn't make sense, I also have Intel processors but they are running at 2.6 GHz and they are giving me 16-17 pts per hr. (not over claiming sense they run a lot of zero redunancy and I usually get more then I claim).
----------------------------------------But you are right, results are what matter. P.S. I remember Sekerob saying that boinc is working on some new credit scheme to factor in the linux factor. They were testing it at seti, so it should go official eventually. |
||
|
|
BobCat13
Senior Cruncher Joined: Oct 29, 2005 Post Count: 295 Status: Offline Project Badges: 
|
P.S. I remember Sekerob saying that boinc is working on some new credit scheme to factor in the linux factor. They were testing it at seti, so it should go official eventually. Malariacontrol.net is using the new credit system if you want to try it out. I have run a few under Windows, but haven't tried Linux yet. On Windows, the granted credit was 30-60% higher than what the client benchmarks would have requested. It seemed the longer the task, the higher the percentage over the benchmark numbers. Once it is official, it would be great to see WCG adopt it so low-claiming benchmark wingmen don't affect the granted credit. I have seen claims of less than 10 credits on HCC under linux. |
||
|
|
krakatuk
Advanced Cruncher Germany Joined: Oct 3, 2008 Post Count: 141 Status: Offline Project Badges:
|
Anhhai, do you have HT?
---------------------------------------- Maybe you get same amount of points/hr for a WU as nanoprobe, but your core is doing only 1 WU same time nanoprobe's core is doing 2 WUs? This HT is such a mess! I hope it really helps to do work faster and not just confuses boinc and cheats stats... |
||
|
|
|