Boinc Points / Credit AMD vs Intel

Message boards : Questions and problems : Boinc Points / Credit AMD vs Intel
Message board moderation

To post messages, you must log in.

AuthorMessage
Grandpa

Send message
Joined: 20 Jun 14
Posts: 11
United States
Message 55514 - Posted: 22 Aug 2014, 4:12:58 UTC
Last modified: 22 Aug 2014, 5:04:33 UTC

It appears that the boinc benchmark / credit system has a bit of a problem when it comes to NumberFields I have seen this problem before on other projects also but really did not pay much attention to it since I figured I was just imagining things. That is until I started all of the rigs at relatively the same time using Bam on Numberfields, I noticed that the AMD 63xx rig was completing more WU's than the Intel was and was receiving less credit for WU's preformed in the same amount of time as the Intel’s were completing them in.

I then posted in the NumberFields forum asking if there was a difference in assignments for Intel and AMD and the scientist reply was there were no differences. He then checked my stats and did confirm that the AMD rigs were completing the WU's faster than the Intel’s and should be receiving the same credit for CPU time as the Intel’s but were not they actually recieve 30% less than the Intel’s which are slower at NuberFields Clock for Clock.

He then asked me to re bencmark my system to see if that would help and it actually lowered my benchmark scores whic I assume is going to make an even greater difference in the credit spread. I do believe the Boinc benchmark and credit system has a problem that might need to be looked at in really is not a fair system when it comes to some projects when it come to AMD vs Intel.

As you can see looking at the data below the AMD 63xx rig is much faster than the Intel’s and the 6276 rig is faster clock for clock than the Intel’s but both received much less credit for more completed WU's in the same time period.

http://numberfields.asu.edu/NumberFields/forum_thread.php?id=209

Grandma AMD 6276 @ 3042Mhz / 64 core	All tasks for computer 19134	Linux	3.2.0-47-generic	BOINC version	7.0.65	AMD Opteron(TM) Processor 6276 [Family 21 Model 1 Stepping 2]
State: All (1431) · In progress (64) · Validation pending (0) · Validation inconclusive (0) · Valid (1367) · Invalid (0) · Error (0)						(64 processors) 
Application: All (1431) · Get Decic Fields (1411) · Get Decics with Bounded Discriminant (20) 						
ID 	Today 	Yesterday 	2 Days Ago 			
19134	81030	79662	79588			
						
Grandpa AMD 63xx @ 3813Mhz / 48 core	All tasks for computer 19787	Linux	3.2.0-47-generic	BOINC version	7.0.65	AMD Eng Sample, ZS288145TCG54_34/28/20_2/16 [Family 21 Model 2 Stepping 0]
State: All (1696) · In progress (84) · Validation pending (0) · Validation inconclusive (0) · Valid (1612) · Invalid (0) · Error (0)						(48 processors) 
Application: All (1696) · Get Decic Fields (1676) · Get Decics with Bounded Discriminant (20) 						
ID 	Today 	Yesterday 	2 Days Ago 			
19787	75189	75350	73633			
						
Musky Intel 4650 @ 3134Mhz /64 core	All tasks for computer 19181	Linux	3.2.0-55-generic	BOINC version	7.0.65	Genuine Intel(R) CPU @ 2.70GHz [Family 6 Model 45 Stepping 5]
State: All (1429) · In progress (64) · Validation pending (0) · Validation inconclusive (0) · Valid (1363) · Invalid (0) · Error (2)						(64 processors) 
Application: All (1429) · Get Decic Fields (1407) · Get Decics with Bounded Discriminant (22) 						
ID 	Today 	Yesterday 	2 Days Ago 			
19181	115422	112214	105445			
						
Patriot Intel 4650 @ 3134Mhz / 64 core	All tasks for computer 19133	Linux	3.8.0-39-generic	BOINC version	7.0.65	Genuine Intel(R) CPU @ 2.70GHz [Family 6 Model 45 Stepping 5]
State: All (1407) · In progress (64) · Validation pending (0) · Validation inconclusive (0) · Valid (1343) · Invalid (0) · Error (0)						(64 processors) 
Application: All (1407) · Get Decic Fields (1387) · Get Decics with Bounded Discriminant (20) 						
ID 	Today 	Yesterday 	2 Days Ago 			
19133	110536	109153	110064			
						
Scotty Intel 4650 @ 3175Mhz / 64 core	All tasks for computer 19127	Linux	3.2.0-47-generic	BOINC version	7.0.27	Genuine Intel(R) CPU @ 2.70GHz [Family 6 Model 45 Stepping 5]
State: All (1583) · In progress (88) · Validation pending (0) · Validation inconclusive (0) · Valid (1431) · Invalid (0) · Error (64)						(64 processors) 
Application: All (1583) · Get Decic Fields (1564) · Get Decics with Bounded Discriminant (19) 						
ID 	Today 	Yesterday 	2 Days Ago 			
19127	113613	128981	124152			
						
Core32 Intel 4650L @ 3100Mhz / 64 core	All tasks for computer 19220	Linux	3.2.0-47-generic	BOINC version	7.0.65	Genuine Intel(R) CPU @ 2.60GHz [Family 6 Model 45 Stepping 5]
State: All (1436) · In progress (87) · Validation pending (0) · Validation inconclusive (0) · Valid (1349) · Invalid (0) · Error (0)						(64 processors) 
Application: All (1436) · Get Decic Fields (1411) · Get Decics with Bounded Discriminant (25) 						
ID 	Today 	Yesterday 	2 Days Ago 			
19220	107368	108940	109216			
						
Tear Intel 4650 @ 3134Mhz / 64 core	All tasks for computer 19121	Linux	3.2.0-47-generic	7.0.65	7.0.65	Genuine Intel(R) CPU @ 2.70GHz [Family 6 Model 45 Stepping 5]
State: All (1491) · In progress (143) · Validation pending (0) · Validation inconclusive (0) · Valid (1348) · Invalid (0) · Error (0)						(64 processors) 
Application: All (1491) · Get Decic Fields (1476) · Get Decics with Bounded Discriminant (15)						
ID 	Today 	Yesterday 	2 Days Ago 			
19121	117970	111557	106929			
ID: 55514 · Report as offensive
Grandpa

Send message
Joined: 20 Jun 14
Posts: 11
United States
Message 55524 - Posted: 22 Aug 2014, 16:49:05 UTC - in response to Message 55514.  
Last modified: 22 Aug 2014, 16:50:44 UTC

Just an update I did a calculation of the last 100 Numberfields WU's ran per machine and it shows a 30% discrepancy between AMD and Intel for an approximatively equal amount of work done by the computers in an equal time period.

https://www.dropbox.com/s/x1vz7ohl5za0vrh/AMD%20vs%20Intel.ods


AMD Opteron(TM) Processor 6276 [Family 21 Model 1 Stepping 2]				
(64 processors) 				
Total runtime 100 WU's	1810010.65	Total points 100 WU's	26128.8	Grandma 6276 AMD
Avg runtime per WU sec.	18100.1065	avg points per WU	261.288	3042Mhz
Avg runtime per WU min.	301.6684416667			
Avg runtime per WU hr.	[color=red]5.0278073611[/color]	avg points per hr runtime	[color=red]51.9685781959[/color]	
				
AMD Eng Sample, ZS288145TCG54_34/28/20_2/16 [Family 21 Model 2 Stepping 0]				
(48 processors) 				
Total runtime 100 WU's	1203388.68	Total points 100 WU's	21917.72	Grandpa AMD 63xx
Avg runtime per WU sec.	12033.8868	avg points per WU	219.1772	3813Mhz
Avg runtime per WU min.	200.56478			
Avg runtime per WU hr.	[color=red]3.3427463333[/color]	avg points per hr runtime	[color=red]65.5680025177[/color]	
				
Genuine Intel(R) CPU @ 2.70GHz [Family 6 Model 45 Stepping 5]				
(64 processors) 				
Total runtime 100 WU's	1839477.54	Total points 100 WU's	38003.75	Musky 4650 Intel
Avg runtime per WU sec.	18394.7754	avg points per WU	380.0375	3134Mhz
Avg runtime per WU min.	306.57959			
Avg runtime per WU hr.	[color=red]5.1096598333[/color]	avg points per hr runtime	[color=red]74.3762818653[/color]	
				
Genuine Intel(R) CPU @ 2.70GHz [Family 6 Model 45 Stepping 5]				
(64 processors) 				
Total runtime 100 WU's	1839477.54	Total points 100 WU's	38003.75	Patriot 4650 Intel
Avg runtime per WU sec.	18394.7754	avg points per WU	380.0375	3134Mhz
Avg runtime per WU min.	306.57959			
Avg runtime per WU hr.	[color=red]5.1096598333[/color]	avg points per hr runtime	[color=red]74.3762818653[/color]	
				
Genuine Intel(R) CPU @ 2.70GHz [Family 6 Model 45 Stepping 5]				
(64 processors) 				
Total runtime 100 WU's	1725430.97	Total points 100 WU's	36059.14	Scotty 4650 Intel
Avg runtime per WU sec.	17254.3097	avg points per WU	360.5914	3175Mhz
Avg runtime per WU min.	287.5718283333			
Avg runtime per WU hr.	[color=red]4.7928638056[/color]	avg points per hr runtime	[color=red]75.2350608382[/color]	
				
Genuine Intel(R) CPU @ 2.60GHz [Family 6 Model 45 Stepping 5]				
(64 processors) 				
Total runtime 100 WU's	1766059.59	Total points 100 WU's	34952.56	Core32 4650L Intel
Avg runtime per WU sec.	17660.5959	avg points per WU	349.5256	3100Mhz
Avg runtime per WU min.	294.343265			
Avg runtime per WU hr.	[color=red]4.9057210833[/color]	avg points per hr runtime	[color=red]71.2485675526[/color]	
				
Genuine Intel(R) CPU @ 2.70GHz [Family 6 Model 45 Stepping 5]				
(64 processors) 				
Total runtime 100 WU's	1696143.06	Total points 100 WU's	35126.23	Tear 4650 Intel
Avg runtime per WU sec.	16961.4306	avg points per WU	351.2623	3134Mhz
Avg runtime per WU min.	282.69051			
Avg runtime per WU hr.	[color=red]4.7115085[/color]	avg points per hr runtime	[color=red]74.5541051237[/color]	
ID: 55524 · Report as offensive
Profile Jord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15477
Netherlands
Message 55557 - Posted: 24 Aug 2014, 13:24:03 UTC

Credits aren't calculated based on benchmarks times unit of time anymore, but are based on how many floating point operations tasks take. What I understand is that AMD CPUs are not as good at doing FLOPs as Intel CPUs are, which counts for a lot of the discrepancies.
ID: 55557 · Report as offensive
Grandpa

Send message
Joined: 20 Jun 14
Posts: 11
United States
Message 55567 - Posted: 24 Aug 2014, 17:42:11 UTC - in response to Message 55557.  

Credits aren't calculated based on benchmarks times unit of time anymore, but are based on how many floating point operations tasks take. What I understand is that AMD CPUs are not as good at doing FLOPs as Intel CPUs are, which counts for a lot of the discrepancies.


Which we all know is true, but in the case of NumberFields my AMD CPU's are faster than the Intel's core fore core clock for clock, so they should be getting more credit per hr. of runtime per core. The credit system has a bug or error in it when it comes to some types of work preformed by the boinc project. I do believe that the developers of the credit system should try to fix this discrepancy. It is not really fair to use a credit system that favours one over the other, work done is work done. Why should one computer receive 30% more credit for less work done than the another computer doing more work.

I am not a AMD or Intel fanboy I own both and tend to use whatever does best at the time. But I would like to receive = credit for = work. If one CPU is better than another at a given project it should be reflected in the credit received for doing the given work.

This does seem to be an area that nobody seems to want to address for some reason, I posted ample information to show the discrepancy yet the developers of the credit system have not even acknowledged there there is a problem with the system, I do not pretend to know how much work it would take to fix it, but at least it should be acknowledged and looked at. If they can not fix it then make it public knowledge that brand X may be faster at a given project but brand Y is going to receive more credit for doing less work because ..........
ID: 55567 · Report as offensive
SuperSluether

Send message
Joined: 6 Jul 14
Posts: 94
United States
Message 55639 - Posted: 27 Aug 2014, 0:50:28 UTC - in response to Message 55557.  

Credits aren't calculated based on benchmarks times unit of time anymore, but are based on how many floating point operations tasks take. What I understand is that AMD CPUs are not as good at doing FLOPs as Intel CPUs are, which counts for a lot of the discrepancies.


Wait, so a faster processor would get more credit for a task than a slower processor doing the same task? That doesn't seem fair, because then older computers wouldn't get nearly as much!
ID: 55639 · Report as offensive
Profile Jord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15477
Netherlands
Message 55647 - Posted: 27 Aug 2014, 14:30:42 UTC - in response to Message 55639.  

Wait, so a faster processor would get more credit for a task than a slower processor doing the same task?

No, since the calculation of credits is done based on the amount of floating point operations (flops) that a task takes. Speed doesn't matter in this.

For instance, a Seti Astropulse task that gets sent out by the project to take 42,293,901,863.366425 flops, will do so equally on a 3.5GHz and 1.0GHz computer. The 1.0GHz CPU may take longer than the 3.5GHz CPU on doing the calculations, but in the end it doesn't matter, as both the computers will run the task for 42,293,901,863.366425 flops and then finish the task. The credits are based on the flops value, not on the time it took both CPUs to finish the task.

Note: we're talking FLOPS, not FLOPs. The latter abbreviation is a measure for floating point operations per second.
ID: 55647 · Report as offensive
noderaser
Avatar

Send message
Joined: 2 Jan 14
Posts: 276
United States
Message 55660 - Posted: 28 Aug 2014, 3:45:04 UTC - in response to Message 55639.  

Wait, so a faster processor would get more credit for a task than a slower processor doing the same task? That doesn't seem fair, because then older computers wouldn't get nearly as much!

World Community Gird (and perhaps some other projects that I don't know about) have two metrics for credit, points (based on BOINC cobblestones) and computation time--and their badges are based on computation time. That way, even someone with a slow computer can get recognized on an equal plane with a supercomputer. However, computers with multiple cores can rack up time much faster since each core is counted separately, and those people who have tons of computers on their account still get tons of hours computing time.
My Detailed BOINC Stats
ID: 55660 · Report as offensive
SuperSluether

Send message
Joined: 6 Jul 14
Posts: 94
United States
Message 55681 - Posted: 29 Aug 2014, 2:46:09 UTC - in response to Message 55647.  


Note: we're talking FLOPS, not FLOPs. The latter abbreviation is a measure for floating point operations per second.


Oh, that makes more sense. :)
ID: 55681 · Report as offensive
Grandpa

Send message
Joined: 20 Jun 14
Posts: 11
United States
Message 55702 - Posted: 29 Aug 2014, 14:22:38 UTC

No, since the calculation of credits is done based on the amount of floating point operations (flops) that a task takes. Speed doesn't matter in this.

For instance, a Seti Astropulse task that gets sent out by the project to take 42,293,901,863.366425 flops, will do so equally on a 3.5GHz and 1.0GHz computer. The 1.0GHz CPU may take longer than the 3.5GHz CPU on doing the calculations, but in the end it doesn't matter, as both the computers will run the task for 42,293,901,863.366425 flops and then finish the task. The credits are based on the flops value, not on the time it took both CPUs to finish the task.

Note: we're talking FLOPS, not FLOPs. The latter abbreviation is a measure for floating point operations per second.


AMD Opteron(TM) Processor 6276 [Family 21 Model 1 Stepping 2]
(64 processors)
Total runtime 100 WU's 1810010.65 Total points 100 WU's 26128.8 Grandma 6276 AMD
Avg runtime per WU sec. 18100.1065 avg points per WU 261.288 3042Mhz
Avg runtime per WU min. 301.6684416667
Avg runtime per WU hr. 5.0278073611 avg points per hr runtime 51.9685781959

Genuine Intel(R) CPU @ 2.70GHz [Family 6 Model 45 Stepping 5]
(64 processors)
Total runtime 100 WU's 1839477.54 Total points 100 WU's 38003.75 Musky 4650 Intel
Avg runtime per WU sec. 18394.7754 avg points per WU 380.0375 3134Mhz
Avg runtime per WU min. 306.57959
Avg runtime per WU hr. 5.1096598333 avg points per hr runtime 74.3762818653


OK but it still does not change the fact that this type of credit system has a pretty big flaw in it. According to the developer at Numberfields each of the machines above are doing the same amount of work and lets say that each WU is worth 1 FLOPS since the numbers were averaged over 100 WU's each and all things being equal those 2 machines should have received roughly the same amount of credit with a slight advantage going to the AMD, but in actuality AMD is receiving 30% less.

I do not really see this as being a fair credit system to the AMD users when it come to certain types of work.

AMD 100 WU's = 100 FLOPS = 502.78 hrs of runtime = 26706.79 points of credit for 100 theoretical FLOPS

Intel 100 WU's = 100 FLOPS = 510.96 hrs of runtime = 38003.74 points of credit for 100 theoretical FLOPS

So in plain simple terms AMD gets less credit for doing the same amount of work as Intel does. I am just pointing out a pretty big flaw in the current credit system, from everything I have read and been told this system was set up to promote equality between all work done and to try and discourage cheating, It appears to me that it may have missed it's mark when in the equality field and they may need to go back to the drawing board and try and fix the problem or at least let people know that the credit system has a problem with some projects.
ID: 55702 · Report as offensive

Message boards : Questions and problems : Boinc Points / Credit AMD vs Intel

Copyright © 2024 University of California.
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.