Akos Optimizations require installing Calibrating Clients 2 !!!

Erik
Erik
Joined: 14 Feb 06
Posts: 2,815
Credit: 2,645,600
RAC: 0

RE: I will modify the 41.xx

Message 32121 in response to message 32120

Quote:
I will modify the 41.xx codes for this 10 days... :)

I noticed on the release page that there is a D41.08 & D41.15. The D41.08 downloads alb_S41.08.zip.

Akos Fekete
Akos Fekete
Joined: 13 Nov 05
Posts: 561
Credit: 4,527,270
RAC: 0

RE: RE: I will modify the

Message 32122 in response to message 32121

Quote:
Quote:
I will modify the 41.xx codes for this 10 days... :)
I noticed on the release page that there is a D41.08 & D41.15. The D41.08 downloads alb_S41.08.zip.

Thanks. It's corrected.

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5,851
Credit: 110,854,351,096
RAC: 33,938,212

RE: RE: RE: I just

Message 32123 in response to message 32120

Quote:
Quote:
Quote:
I just noticed that your XP1700+ box that you commented on in the Milestones thread has just had 5 unhandled exception errors, all from the same data file z1_1212.5, quite recently. Any further thoughts on what might be causing these errors? I haven't seen any more of these errors on any of my boxes although I would only notice if there were a whole string of them like the ones I reported previously. I haven't seen that behaviour occur since that time.
I don't know its ground, but i will look after.
Probably i found the reason of this fault. Really stupid and simple thing.

That's really good news!! The main reason I drew it to your attention was that I figured that the problem happening on your own box would spur you on to solve it. That little matter of pride in one's work.... :).

Quote:
I will modify the 41.xx codes for this 10 days... :)

Thanks very much for doing that. I have a sneaking suspicion that your current optimised versions may be useful for a while yet. Bernd did say at one point that nothing was set in stone and that the new S5 work might be able to be done with your apps. In any case, it's really good to have the most stable versions available, no matter how short the time.

Cheers,

Cheers,
Gary.

Akos Fekete
Akos Fekete
Joined: 13 Nov 05
Posts: 561
Credit: 4,527,270
RAC: 0

RE: RE: I will modify the

Message 32124 in response to message 32123

Quote:
Quote:
I will modify the 41.xx codes for this 10 days... :)
Thanks very much for doing that. I have a sneaking suspicion that your current optimised versions may be useful for a while yet. Bernd did say at one point that nothing was set in stone and that the new S5 work might be able to be done with your apps. In any case, it's really good to have the most stable versions available, no matter how short the time.

The S5 app will use the number of FPOPs to calculate the claimed credit value, this is different from the current solutions, so these optimized apps will not compatible with the ones.

But, don't forget the new apps consist the most part of the improvements and have better solutions ( eg. communication between BOINC and the app, etc. ).

DanNeely
DanNeely
Joined: 4 Sep 05
Posts: 1,364
Credit: 3,562,358,667
RAC: 9

Will any of the changes made

Will any of the changes made make bringing the existing optimizations to the s5 app difficult? ALso, what ever happened to the 3dnow/sse app?

Akos Fekete
Akos Fekete
Joined: 13 Nov 05
Posts: 561
Credit: 4,527,270
RAC: 0

RE: Will any of the changes

Message 32126 in response to message 32125

Quote:
Will any of the changes made make bringing the existing optimizations to the s5 app difficult?

The main calculation will be the same, except some constant (eg. cycle number).
Bernd implemented the most important improvements.

Quote:
ALso, what ever happened to the 3dnow/sse app?

As far as i know, Bernd is planning different versions ( 3DNow!, 64bit, etc... ), but he has lot of work, so some things have priority ( new WUs, new credit system, most popular platforms, etc... ).

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5,851
Credit: 110,854,351,096
RAC: 33,938,212

RE: RE: RE: I will

Message 32127 in response to message 32124

Quote:
Quote:
Quote:
I will modify the 41.xx codes for this 10 days... :)
Thanks very much for doing that. I have a sneaking suspicion that your current optimised versions may be useful for a while yet. Bernd did say at one point that nothing was set in stone and that the new S5 work might be able to be done with your apps. In any case, it's really good to have the most stable versions available, no matter how short the time.
The S5 app will use the number of FPOPs to calculate the claimed credit value, this is different from the current solutions, so these optimized apps will not compatible with the ones.

Thanks for the response but I don't think that the new app is going to use FLOP counting. Here is a direct quote of what Bernd said about this. It comes from the Windows Beta Test thread:-

Quote:

The credit is claimed by the Core Client (and the scheduler), and granted by the validator, so primarily this doesn't have much to do with the App.

There is a way to tell the Client the amount of FLOPs that's actually used from the App, instead of using the client's benchmark result, but it's hard to implement a counting that works on all CPUs and it would just balance for the different benchmark results of different Clients, but does not much about the different efficiencies of the Apps on different CPUs.

So in the next run, starting in about two weeks, we will grant a rather fixed credit for each (valid) Task done. But that, again, is rather related to the scheduler and the validator, not to the App.

I took this to mean that they weren't going to use FLOP counting but rather a home grown system of fixed credit. Also note the bit in the first paragraph which is saying that the app has little involvement with credit calculations anyway. Also, Bernd said the following in response to a direct question about whether we could continue using your optimised apps:-

Quote:
Quote:
Does this mean that those of us currently using a CPU specific akos app will be able to continue using it for S5?

I simply can't say that right now. It is covered in internal tests of the new setup which are not finished yet.

Akos' Apps are still based on the 4.37 App, which is pretty old code, especially the BOINC part. If it turns out we need e.g. fixes or features from newer BOINC, then you won't be able to continue using these Apps.

I took this to mean that we might be able to use your existing apps if the new app didn't need fixes or features from the newer BOINC. He seemed at least to be leaving the door open just a smidgen :). However, I'm interpreting your next statement below to mean that fixes and/or new features are going to be needed which will render the old apps invalid.

Quote:
But, don't forget the new apps consist the most part of the improvements and have better solutions ( eg. communication between BOINC and the app, etc. ).

I guess the only question left to ask you (and this is very much a dumb layman's question since I have no real concept of how you worked your magic) is this:-

If your old optimised apps run about twice as fast as the new Windows Beta, doesn't that leave you some room to make an "Optimised Beta" that will outperform the current Beta? Bernd has stated that it is likely that the current Beta will become the new app if all goes well with the testing. Obviously it isn't that simple otherwise you would have probably already done it :). I was just curious, that's all :).

Cheers,

Cheers,
Gary.

Franz Bauer
Franz Bauer
Joined: 22 May 06
Posts: 5
Credit: 20,077
RAC: 0

RE: Probably 2nd and 3rd

Message 32128 in response to message 32116

Quote:
Probably 2nd and 3rd run with HT.


Hi Aksof:

Thanks for your response 8 Jun 2006 5:17:21 UTC to my earlier questions.

And Dan Neely you were right. I was running 50:50 Einstein and Seti.

I’ve posted the results using both your suggestions:

Intel Pentium 4 CPU 3.40 GHz, 1 GB RAM, ? MB L2, 800 MHz FSB With Hyper Threading ON
5.4.6 core client and Standard Ver. 4.37 (average of 59 work units)
CPU time = 13,923.19 sec., Claimed credit = 8.65 CS/hr., Granted credit = 8.69 CS/hr.

Pentium 4, 650 3.4 GHz, 2 GB RAM, 2 MB L2, 800 MHz FSB With Hyper Threading ON
5.3.12.tx36 core client with Akosf Ver U41.04 (average of 102 work units)
CPU time = 6,466.87 sec., Claimed credit = 17.08 CS/hr., Granted credit = 20.46 CS/hr.

Improvement over standard 4.37 app.
CPU time = 2.15x, Claimed credit = 1.98x, Granted credit = 2.35x

Pentium 4, 650 3.4 GHz, 2 GB RAM, 2 MB L2, 800 MHz FSB With Hyper Threading OFF
5.3.12.tx36 core client with Akosf Ver U41.04 (average of 119 work units)
CPU time = 3,336.49 sec., Claimed credit = 38.82 CS/hr., Granted credit = 44.24 CS/hr.

Improvement over standard 4.37 app.
CPU time = 2.09x, Claimed credit = 2.25x, Granted credit = 2.54x

Improvement over Akosf Ver U41.04 app.
CPU time = 0.97x, Claimed credit = 1.14x, Granted credit = 1.08x

Aksof, I noticed you were over in Seti asking questions. So, I have posted some of results I obtained with the regular Seti and the enhanced Seti.

Pentium 4, 650 3.4 GHz, 2 GB RAM, 2 MB L2, 800 MHz FSB With Hyper Threading ON
5.3.12.tx36 BOINC client and standard Seti V4.18 (average of 23 work units)
CPU time = 2,072.61 sec., Claimed credit = 16.81 CS/hr., Granted credit = 23.30 CS/hr.

5.3.12.tx36 BOINC client and Crunch3r’s Seti SSE3 V4.11 (average of 97 work units)
CPU time = 3,110.58 sec., Claimed credit = 34.90 CS/hr., Granted credit = 29.81 CS/hr.

5.3.12.tx36 BOINC client and Crunch3r’s Seti SSE3 V5.11 (average of 86 work units)
CPU time = 10,213.65 sec., Claimed credit = 46.27 CS/hr., Granted credit = 18.03 CS/hr.

5.3.12.tx36 BOINC client and Crunch3r’s Seti SSE3 V5.12 (average of 31 work units)
CPU time = 13,751.11 sec., Claimed credit = 14.38 CS/hr., Granted credit = 14.13 CS/hr.

The claimed credit for V5.11 is the reason poor Crunch3r was crucified and burned at the stack by the trailer park boys at Seti and threatened by Eric to be hunted down by UCB’s lawyers.

Hopefully Einstein doesn’t do the same mistake when they switch to S5.

I think I will try BoincStudio next. It seems to have some improved advantages over the 5.3.12.tx36 BOINC client.

Regards
Franz

archae86
archae86
Joined: 6 Dec 05
Posts: 3,146
Credit: 7,093,164,931
RAC: 1,376,841

I suggest to the moderators

I suggest to the moderators it is probably time to "unsticky" this thread, as not much S4 work remains, and it is far off-focus for S5 work, so might distract a newcomer.

I won't feel offended if you delete this post whichever way you decide, as it is otherwise unrelated to the thread topic.

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5,851
Credit: 110,854,351,096
RAC: 33,938,212

RE: I suggest to the

Message 32130 in response to message 32129

Quote:

I suggest to the moderators it is probably time to "unsticky" this thread, as not much S4 work remains, and it is far off-focus for S5 work, so might distract a newcomer.

I won't feel offended if you delete this post whichever way you decide, as it is otherwise unrelated to the thread topic.

The process was already in train. If you hadn't bumped it, it would have dropped out of sight :).

This thread should now be regarded as closed. It remains visible for historical purposes.

Cheers,
Gary.

Comment viewing options

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