AMD vs Intel

Message boards : Number crunching : AMD vs Intel
Message board moderation

To post messages, you must log in.

AuthorMessage
Grandpa

Send message
Joined: 11 Jun 14
Posts: 11
Credit: 45,376,032
RAC: 0
Message 1097 - Posted: 21 Aug 2014, 4:08:04 UTC

I was just curious as to why a WU that is run on an Intel CPU that is slower gets more credit than A WU that is run on a faster AMD that uses the same amount of CPU time. Is the AMD given smaller WU's by default or something along that line.

This WU was run on an Opteron 63xx CPU @ 3.8Ghz and received 333.98 points

8271791 	7647561 	20 Aug 2014, 19:20:43 UTC 	21 Aug 2014, 2:58:06 UTC 	Completed and validated 	18,512.48 	18,290.26 	333.98 	Get Decic Fields v1.02 


This was run on an Intel 4650L @ 3.1Ghz it took slightly less time and received more credit

8271417 	7647187 	20 Aug 2014, 15:30:02 UTC 	20 Aug 2014, 22:45:45 UTC 	Completed and validated 	18,474.27 	18,441.73 	365.63 	Get Decic Fields v1.02 


This was ran on the same Intel machine and took slightly more time and received more credit

8272432 	7648202 	20 Aug 2014, 19:17:33 UTC 	21 Aug 2014, 3:32:04 UTC 	Completed and validated 	18,764.50 	18,717.99 	371.37 	Get Decic Fields v1.02 



Now some will say the Intel is actually faster than the AMD but I do not think that is correct. Both of these machines were started on the project ar the same time and the AMD has completed more WU's In the same amount of time and the AMD only has 48 cores vs the Intel having 64 cores.


63xx @ 3813 / 48 core
State: All (1018) · In progress (82) · Validation pending (0) · Validation inconclusive (0) · Valid (936) · Invalid (0) · Error (0)
Application: All (1018) · Get Decic Fields (1005) · Get Decics with Bounded Discriminant (13) 


4650L @ 3100 / 64 core
State: All (873) · In progress (100) · Validation pending (0) · Validation inconclusive (0) · Valid (773) · Invalid (0) · Error (0)
Application: All (873) · Get Decic Fields (858) · Get Decics with Bounded Discriminant (15)
[/code]
ID: 1097 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Greg Tucker
Project administrator
Project developer
Project tester

Send message
Joined: 8 Jul 11
Posts: 46
Credit: 7,144,042
RAC: 0
Message 1098 - Posted: 21 Aug 2014, 14:27:03 UTC - in response to Message 1097.  

It is just an affect of the BOINC credit formula that is based on benchmark tests run periodically. There is no logic in the scheduler, app or anywhere else that is looking if the WU will run on Intel or AMD and changing the WU size or adjusting credit. Other users have claimed that AMD is faster. It may just depend on the models you are comparing.
ID: 1098 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Eric Driver
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 8 Jul 11
Posts: 1341
Credit: 509,702,626
RAC: 554,083
Message 1099 - Posted: 22 Aug 2014, 1:18:31 UTC - in response to Message 1098.  

You bring up a very good point. Based on the number of results returned, your AMD is clearly faster. So why does it receive less credit per WU?

As Greg mentions, the credit is assigned based on runtime and on the BOINC benchmarks. So I think the question becomes, why does BOINC think your AMD is slower? Maybe you were doing something computationally intensive at the time BOINC ran it's benchmarks?

If I'm not mistaken, the benchmark results are stored in a file somewhere. Finding that could shed some light on this. If I get a chance later, I will try to dig a little further to see where that information is stored. Also, as you are probably aware, you can tell BOINC to compute new benchmarks.
ID: 1099 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Grandpa

Send message
Joined: 11 Jun 14
Posts: 11
Credit: 45,376,032
RAC: 0
Message 1100 - Posted: 22 Aug 2014, 1:34:13 UTC - in response to Message 1099.  
Last modified: 22 Aug 2014, 1:42:44 UTC

You bring up a very good point. Based on the number of results returned, your AMD is clearly faster. So why does it receive less credit per WU?

As Greg mentions, the credit is assigned based on runtime and on the BOINC benchmarks. So I think the question becomes, why does BOINC think your AMD is slower? Maybe you were doing something computationally intensive at the time BOINC ran it's benchmarks?

If I'm not mistaken, the benchmark results are stored in a file somewhere. Finding that could shed some light on this. If I get a chance later, I will try to dig a little further to see where that information is stored. Also, as you are probably aware, you can tell BOINC to compute new benchmarks.


LOL Boinc points system is predigest against AMD, here is some very good proof of that all of these rigs were started at relatively the same time by BAM. As you can see the even the 6276's clock for clock are faster than Intel at Numberfields and the 63xx ES chips 48 core are much faster but yet both the AMD rigs make significantly less PPD than the Intels and they are doing as much or more work per day.

My machines stats are not hidden and are available for all to see here you can clearly see that the 63xx rig has the fastest completion time per WU of all the rigs and the 6276 rig is pretty much = to all the Intel’s yet it gets around 30% less credit than the Intel’s. I think Boinc needs to go back to the drawing board with their credit system.


Grandma AMD 6276 @ 3042 / 64 core	All tasks for computer 19134		
State: All (1431) · In progress (64) · Validation pending (0) · Validation inconclusive (0) · Valid (1367) · Invalid (0) · Error (0)			
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 @ 3813 / 48 core	All tasks for computer 19787		
State: All (1696) · In progress (84) · Validation pending (0) · Validation inconclusive (0) · Valid (1612) · Invalid (0) · Error (0)			
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 @ 3134 /64 core	All tasks for computer 19181		
State: All (1429) · In progress (64) · Validation pending (0) · Validation inconclusive (0) · Valid (1363) · Invalid (0) · Error (2)			
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 @ 3134 64 core	All tasks for computer 19133		
State: All (1407) · In progress (64) · Validation pending (0) · Validation inconclusive (0) · Valid (1343) · Invalid (0) · Error (0)			
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 @ 3175 / 64 core	All tasks for computer 19127		
State: All (1583) · In progress (88) · Validation pending (0) · Validation inconclusive (0) · Valid (1431) · Invalid (0) · Error (64)			
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 @ 3100 / 64 core	All tasks for computer 19220		
State: All (1436) · In progress (87) · Validation pending (0) · Validation inconclusive (0) · Valid (1349) · Invalid (0) · Error (0)			
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 @ 3134 / 64 core	All tasks for computer 19121		
State: All (1491) · In progress (143) · Validation pending (0) · Validation inconclusive (0) · Valid (1348) · Invalid (0) · Error (0)			
Application: All (1491) · Get Decic Fields (1476) · Get Decics with Bounded Discriminant (15)			
ID 	Today 	Yesterday 	2 Days Ago 
19121	117970	111557	106929

http://numberfields.asu.edu/NumberFields/hosts_user.php?userid=19345
ID: 1100 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Eric Driver
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 8 Jul 11
Posts: 1341
Credit: 509,702,626
RAC: 554,083
Message 1101 - Posted: 22 Aug 2014, 1:41:46 UTC - in response to Message 1100.  

I just recalled where I saw the benchmark results. They are in the server database in the host table. I just looked at yours and here is what I found:

AMD: GFLOPS = 2.08, GIOPS = 9.23
Intel: GFLOPS = 3.01, GIOPS = 11.44

So the problem is definitely with the benchmarks. When you get a chance, can you have the client recompute them, just to see if that fixes the problem.
ID: 1101 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Grandpa

Send message
Joined: 11 Jun 14
Posts: 11
Credit: 45,376,032
RAC: 0
Message 1102 - Posted: 22 Aug 2014, 1:43:48 UTC - in response to Message 1101.  
Last modified: 22 Aug 2014, 1:46:51 UTC

I just recalled where I saw the benchmark results. They are in the server database in the host table. I just looked at yours and here is what I found:

AMD: GFLOPS = 2.08, GIOPS = 9.23
Intel: GFLOPS = 3.01, GIOPS = 11.44

So the problem is definitely with the benchmarks. When you get a chance, can you have the client recompute them, just to see if that fixes the problem.

Yeah I will give it a try

(EDIT)
I have now re benchmarked them so we shall see
ID: 1102 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Eric Driver
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 8 Jul 11
Posts: 1341
Credit: 509,702,626
RAC: 554,083
Message 1103 - Posted: 22 Aug 2014, 1:57:54 UTC - in response to Message 1102.  

I just recalled where I saw the benchmark results. They are in the server database in the host table. I just looked at yours and here is what I found:

AMD: GFLOPS = 2.08, GIOPS = 9.23
Intel: GFLOPS = 3.01, GIOPS = 11.44

So the problem is definitely with the benchmarks. When you get a chance, can you have the client recompute them, just to see if that fixes the problem.

Yeah I will give it a try

(EDIT)
I have now re benchmarked them so we shall see


It actually got a little worse. GFLOPS unchanged but GIOPS went from 9.23 to 9.00

Not sure what else we can do at this point.
ID: 1103 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Grandpa

Send message
Joined: 11 Jun 14
Posts: 11
Credit: 45,376,032
RAC: 0
Message 1104 - Posted: 22 Aug 2014, 2:15:41 UTC - in response to Message 1103.  
Last modified: 22 Aug 2014, 2:16:15 UTC

I just recalled where I saw the benchmark results. They are in the server database in the host table. I just looked at yours and here is what I found:

AMD: GFLOPS = 2.08, GIOPS = 9.23
Intel: GFLOPS = 3.01, GIOPS = 11.44

So the problem is definitely with the benchmarks. When you get a chance, can you have the client recompute them, just to see if that fixes the problem.

Yeah I will give it a try

(EDIT)
I have now re benchmarked them so we shall see


It actually got a little worse. GFLOPS unchanged but GIOPS went from 9.23 to 9.00

Not sure what else we can do at this point.


Yeah I noticed that Boinc credit system is messed up when it comes to AMD processors and some projects. Maybe all projects
ID: 1104 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Grandpa

Send message
Joined: 11 Jun 14
Posts: 11
Credit: 45,376,032
RAC: 0
Message 1105 - Posted: 22 Aug 2014, 4:49:15 UTC - in response to Message 1104.  

Well I posted over at the boinc forums so we will see if they can shed any light on the subject.

http://boinc.berkeley.edu/dev/forum_thread.php?id=9569#55514
ID: 1105 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Eric Driver
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 8 Jul 11
Posts: 1341
Credit: 509,702,626
RAC: 554,083
Message 1106 - Posted: 28 Aug 2014, 22:38:00 UTC - in response to Message 1105.  

Well I posted over at the boinc forums so we will see if they can shed any light on the subject.

http://boinc.berkeley.edu/dev/forum_thread.php?id=9569#55514


I just read the above forum, and something came to mind. Due to all the problems we were having with "Credit New", NumberFields started using the validator option to give credit based on runtime (Note: this is an option that's built into the BOINC server that not too many people know about).

For that reason, the credit granted by this project is highly dependent on the benchmarks. So that is the root cause of the problem here, and I would ask why your AMD benchmarks are 30% lower than your Intel benchmarks. Could there be something wrong with the BOINC benchmark tests, or at least something that biases the results towards Intel?
ID: 1106 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Grandpa

Send message
Joined: 11 Jun 14
Posts: 11
Credit: 45,376,032
RAC: 0
Message 1107 - Posted: 30 Aug 2014, 3:52:09 UTC - in response to Message 1106.  

Well I posted over at the boinc forums so we will see if they can shed any light on the subject.

http://boinc.berkeley.edu/dev/forum_thread.php?id=9569#55514


I just read the above forum, and something came to mind. Due to all the problems we were having with "Credit New", NumberFields started using the validator option to give credit based on runtime (Note: this is an option that's built into the BOINC server that not too many people know about).

For that reason, the credit granted by this project is highly dependent on the benchmarks. So that is the root cause of the problem here, and I would ask why your AMD benchmarks are 30% lower than your Intel benchmarks. Could there be something wrong with the BOINC benchmark tests, or at least something that biases the results towards Intel?


I believe so since they are apparently using FLOPS as there benchmark basis and Intel will out produce AMD by quite a bit when it comes to FLOPS.
ID: 1107 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote

Message boards : Number crunching : AMD vs Intel


Main page · Your account · Message boards


Copyright © 2024 Arizona State University