CPU Time versus Credit

Zap
Zap
Joined: 12 Feb 06
Posts: 15
Credit: 3900434
RAC: 0

Below my completed wu's as

Message 97451 in response to message 97450

Below my completed wu's as shown for the moment. It looks like I got two GC's against 1 ABP and it looks like the GC takes on avarege about 2.5 times in cpu seconds to complete. Don't know if that is also the case with clock time.It looks also like the GC is quite underpaying or the ABP overpaying credits. I don't realy mind because I left SAH quite a while ago because I couldn't do Astropulse only there. It was not for the credits I left otherwise I wouldn't be doining EAH :-). But I can understand why some might abort the GC's If they see results like these in there stats or leave the project. I will not. Not for this reason that is.

174436465 74077170 9 May 2010 21:04:52 UTC 11 May 2010 17:13:14 UTC Over Success Done 28,807.61 74.83 250.67
174421390 74072706 9 May 2010 18:23:21 UTC 11 May 2010 14:07:04 UTC Over Success Done 29,100.10 74.83 250.67
174323725 73101651 9 May 2010 12:56:54 UTC 9 May 2010 21:04:52 UTC Over Success Done 9,701.16 65.25 160.00
174247493 73121522 8 May 2010 20:31:59 UTC 10 May 2010 21:49:26 UTC Over Success Done 22,884.36 74.66 250.10
174204090 73992983 8 May 2010 15:28:53 UTC 9 May 2010 12:56:54 UTC Over Success Done 9,969.92 67.06 160.00
174201007 73991785 8 May 2010 10:30:09 UTC 9 May 2010 15:56:35 UTC Over Success Done 28,248.83 74.83 250.67
174058979 73057301 7 May 2010 15:41:13 UTC 8 May 2010 20:31:58 UTC Over Success Done 22,931.36 74.66 250.10
173962695 72829246 6 May 2010 15:13:17 UTC 7 May 2010 15:41:13 UTC Over Success Done 9,739.56 65.51 160.00
173888455 73869840 6 May 2010 15:06:49 UTC 8 May 2010 10:30:08 UTC Over Success Done 22,423.74 74.85 250.75
173761621 73816205 5 May 2010 20:47:53 UTC 6 May 2010 19:08:51 UTC Over Success Done 9,957.28 66.97 160.00
173732290 73804345 5 May 2010 14:57:08 UTC 6 May 2010 19:08:51 UTC Over Success Done 22,595.74 74.85 250.75
173687601 73786785 5 May 2010 10:33:49 UTC 5 May 2010 20:47:48 UTC Over Success Done 9,274.14 62.04 160.00
173541569 72879309 4 May 2010 13:04:25 UTC 5 May 2010 20:47:48 UTC Over Success Done 22,722.25 74.66 250.10
173437286 73683127 3 May 2010 20:00:08 UTC 5 May 2010 14:57:07 UTC Over Success Done 22,417.56 74.66 250.10
172799979 73417183 29 Apr 2010 18:19:57 UTC 1 May 2010 22:58:34 UTC Over Success Done 22,831.90 74.66 250.10
172650404 73352374 28 Apr 2010 19:21:01 UTC 30 Apr 2010 15:41:07 UTC Over Success Done 9,604.20 67.00 160.00

tolafoph
tolafoph
Joined: 14 Sep 07
Posts: 122
Credit: 74659937
RAC: 0

RE: Below my completed wu's

Quote:
Below my completed wu's as shown for the moment. It looks like I got two GC's against 1 ABP and it looks like the GC takes on avarege about 2.5 times in cpu seconds to complete. Don't know if that is also the case with clock time.It looks also like the GC is quite underpaying or the ABP overpaying credits. I don't realy mind because I left SAH quite a while ago because I couldn't do Astropulse only there. It was not for the credits I left otherwise I wouldn't be doining EAH :-). But I can understand why some might abort the GC's If they see results like these in there stats or leave the project. I will not. Not for this reason that is.

Same here.
On my Core2Duo E6750 @3.2 GHz the ABP WUs take around 10k sec and get 160 credits. The S5GCE WUs took around 22k sec and got 250 credits. The new S5GC1 WUs take around 26.5 k sec and also got 250 credits.

I´m a physicist from Hannover and only interested in the science. I don´t care if my RAC is 800 or 1200, but some other crunchers would probably beg to differ.

Sascha

Mad_Max
Mad_Max
Joined: 2 Jan 10
Posts: 153
Credit: 2134792969
RAC: 440417

Same thing with my Core2Duo

Same thing with my Core2Duo (T7700 @ 2.40GHz )
~11k sec ABP WUs
~28k sec S5GCE WUs

I do not think it was a mistake. Just for some reason, core2duo is very effective in latest ABP application. In this type of job he overtakes my Athlon Ghz almost 2 times (despite the lower 25% clock speed). And often outstrips even the wingmans Core i5/i7 based computers.

Though a small general overvaluation(compare to GW tasks) of ABP tasks still exist...

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3522
Credit: 686043288
RAC: 582127

RE: Same thing with my

Message 97454 in response to message 97453

Quote:

Same thing with my Core2Duo (T7700 @ 2.40GHz )
~11k sec ABP WUs
~28k sec S5GCE WUs

I do not think it was a mistake. Just for some reason, core2duo is very effective in latest ABP application. In this type of job he overtakes my Athlon Ghz almost 2 times (despite the lower 25% clock speed). And often outstrips even the wingmans Core i5/i7 based computers.

Though a small general overvaluation(compare to GW tasks) of ABP tasks still exist...

You can't get it right for all platforms. E.g. on my Intel Mac mini (Core Duo), ABP2 takes about 20.5k sec and GC1 around 37k sec (!). Here the ratios runtime/credits for the two apps are much closer http://einsteinathome.org/host/932558/tasks.

Happy crunching
hbe

Stan Pope
Stan Pope
Joined: 22 Dec 05
Posts: 80
Credit: 426811575
RAC: 0

Bikeman said: RE: You

Message 97455 in response to message 97454

Bikeman said:

Quote:
You can't get it right for all platforms.

Yes! Here are some recent Ratios of Credits per Core Per Wall Clock Hour for einsteinbinary_ABP2 3.08 vs. einstein_S5GCE 3.04 on several platforms that I watch (closely):

[pre]
CPU Cores OS Ratio
ABP2:S5GCE
I7-860 8 Vista Home Basic 1.15
I7-860 8 Windows 7 X64 1.00
Q9550 4 Vista Home Prem X64 1.57
Q8400 4 Vista Home Prem X64 1.31
Q6600 4 Vista Home Prem X64 1.47
Q6600 4 XP Home Pro 1.44
FX 64 2 Vista Home Prem 0.89 (maybe low on RAM)
Pentium D 2 Vista Home Prem 1.69

[/pre]

Stan

Mad_Max
Mad_Max
Joined: 2 Jan 10
Posts: 153
Credit: 2134792969
RAC: 440417

Stan Pope Add my one: CPU =

Stan Pope
Add my one:
CPU = Core2Duo T7700
Cores = 2
OS = Win Xp Pro (32 bit)
ABP2:S5GCE Ratio 1.64 (based on overage numbers from 4 S5GCE WUs and 3 ABP2 WUs)

Thanks for numbers. Here is what I call the general(not concrete platform) overvaluation of ABP2
Almost all platforms Credit/Time ratio significantly higher on APB2 WUs.
While at the correct assessment about half of the platform should show results below 1.0, and only the other half above.

This is the reason for the prevalence of "Cherry picking" and frequently asked questions at a forum in the style of "howto opt out of GW search and crunch only APB2?" Receiving the reply that it is impossible, some humbles (obedient), others use a small "xml hack" (smart), somes manual aborts GW WUs in the hope of getting more APB WUs (not smart)

Mike Hewson
Mike Hewson
Moderator
Joined: 1 Dec 05
Posts: 6534
Credit: 284710168
RAC: 110462

RE: Yes! Here are some

Message 97457 in response to message 97455

Quote:
Yes! Here are some .....


Thanks for providing this! An excellent demonstration of how the task/algorithmic differences play out on the hardware ( see 'equivalent' Turing machines ...). That's quite a range of performance 'gearings' and so neatly explains the origin of distress of some contributors. As Mad Max mentions the ABP credits are relatively generous, thus favoring ABP crunching overall ( but not necessarily in the specific ).

How could I put the policy ? "You have to eat your greens, but there's a nice dessert on occasions". That leaves the issue of what to say/do to/about ( short list ):

- the one who dutifully eats all as given but doesn't often score desserts, and/or

- the one who claims intolerance to legumes, and/or

- the one who tosses any veggies onto the floor and will only eat desserts, and/or

- food fights ;-) :-)

Cheers, Mike.

( edit ) Kindly bear in mind that a key reason for introducing desserts onto the menu at all was to combat ennui from the possible blandness of all green consumption. :-)

I have made this letter longer than usual because I lack the time to make it shorter ...

... and my other CPU is a Ryzen 5950X :-) Blaise Pascal

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.