I have nearly twice the number of rosetta samples so I updated the list. WARNING it's getting big. I plan to not post this again until it doubles in size again. This may take a while since I'm now sharing cpu time with SetiBeta.

I've deleted the ralph section of the comparison as it's the same at the earlier one I posted (see above/below for ralph). I've only updated Setibeta and rosetta on this new comparison.

I've made the SRAMANs which claimed too much in red and this data is not seriously affecting the average due to the number of samples, but it does increase the average a little.



Here's the wus included in the above.

Here's what the data displayed earlier looks like in a table.



I've heard Einstein is lowering there credits and I've been preparing this for Eric K (as others have) so he can see where his new "load store adjustment" is taking seti beta.

...I've heard Einstein is lowering there credits ...

The new credits about 65% of the old values per WU, they did that in 2 steps.

Long Einsteins used to give you ~177, then ~145 and now ~122 credits.

They improved the client speed in the last step so it will not drop to 65%, I guess it will be more like 80% - I cannot tell the exact value as I already used the beta client for a while and no WUs from the old client are in my list anymore.

I have nearly twice the number of rosetta samples so I updated the list. WARNING it's getting big. I plan to not post this again until it doubles in size again. This may take a while since I'm now sharing cpu time with SetiBeta.

I've deleted the ralph section of the comparison as it's the same at the earlier one I posted (see above/below for ralph). I've only updated Setibeta and rosetta on this new comparison.

I've made the SRAMANs which claimed too much in red and this data is not seriously affecting the average due to the number of samples, but it does increase the average a little.



Here's the wus included in the above.

Hi mmciastro,
I really appreciate the efforts you put in to research different topics, and have wondered about some of the methods you use, for example, to get results into your spreadsheets. Before I asked about that, you kindly described that. I was hoping for a "magic bullet" method, but we use the same technique. Perhaps the potential "results server" that has been mentioned will make that easier. However, it just goes to show how much effort you really put in on behalf of the project and all us crunchers. Thanks again!

I am wondering also about how you get the spreadsheet tables into graphics. I have been using "PrintScreen" but that's somewhat limited the way I do it: If I try to get some of my "lerger" spreadsheets as a screen capture, I need to zoom out so much there is not enough resolution in the resulting JPEG. I am experimenting with a "Print to JPG" print driver on WinXP, but I've had some problems with that approach, too--so far.

Any info on methods you use or know of to get larger speadsheets into a graphics file I would very much appreciate! Thanks, & keep up the great work!

Can you tell from just this data if a certain architechture is "better" with a certain type of WU? i.e. it gets signifigantly more credit that expected (as opposed to the opposite, amd/intel or different generation), w/o it being a bug :P

---

Can you tell from just this data if a certain architechture is "better" with a certain type of WU? i.e. it gets signifigantly more credit that expected (as opposed to the opposite, amd/intel or different generation), w/o it being a bug :P

@RL Casey, I do it the same way you do. I use print screen and then edit them so they are as close to the same size pics as possible. No magic trick from me I'm afraid.

@AVI, I only have these puters, so No, I can't make any determination based upon data from my own puters. I could if I chose to, to go back through my list of active posters and manually get data that way, but that's alot of work, I'm not quite up to yet.

Can you tell from just this data if a certain architechture is "better" with a certain type of WU? i.e. it gets signifigantly more credit that expected (as opposed to the opposite, amd/intel or different generation), w/o it being a bug :P

@AVI, I only have these puters, so No, I can't make any determination based upon data from my own puters. I could if I chose to, to go back through my list of active posters and manually get data that way, but that's alot of work, I'm not quite up to yet.

Hmm. Im very good at parsing pages with PHP/reg-exp. Is This data available on viewable pages? I can write smthing to gather it for you and dump to excel if you know how to analyze it. Email me: kLdd 16 gmail if you are interested.

---

@RL Casey, I do it the same way you do. I use print screen and then edit them so they are as close to the same size pics as possible. No magic trick from me I'm afraid.

mmciastro, thanks for the info! Also, ('Eureka!') I have downloaded a trial version of Zan Image Printer and it seems to work well. Hmmm, but it's $40 US after the 30-day full-function trial period. I suppose magic has its price. :-)
Here's a ZAN Print sample from WinXP Excel 2002 on some analysis work I have underway (coming soon to a forum near you!). Could use cropping and resizing, but gives the idea.
Thanks again.

@RL Casey, I do it the same way you do. I use print screen and then edit them so they are as close to the same size pics as possible. No magic trick from me I'm afraid.

mmciastro, thanks for the info! Also, ('Eureka!') I have downloaded a trial version of Zan Image Printer and it seems to work well. Hmmm, but it's $40 US after the 30-day full-function trial period. I suppose magic has its price. :-)
Here's a ZAN Print sample from WinXP Excel 2002 on some analysis work I have underway (coming soon to a forum near you!). Could use cropping and resizing, but gives the idea.
Thanks again.

SAve the Charts as images OR save it as a web page and grab the image that way.

Much easier and cheaper.

---

Team mauisun.org

and you may want to try out the 'get data from a web page' feature of Excel too.

---

Team mauisun.org

OK, I'm officially done collecting samples for Rosetta. I think I have enough to say that for my computers rosetta grants about the right amount of credit with the new credit scheme. It's a bit high, but nothing that I can't live with. I am not attaching a pic of all the rosetta WUs involved as it would be huge. If you really want to see them layed out, then I can. I am including (at the bottom) graphs of the wus used which shows the variations in granted credit.

The middle graphs of how each of my 4 puters are doing per projects as it pertains to cross project parity.

Green text is the "benchmark calculated" claimed credit, or what I would/should be getting. Red text indicates data from "non current" credit schemes (I.E old data).

So Einstein is still granting very high credit/hour compared to other projects. CPDN/BBC is not present in your comparison, but they grant also more than the benchmark would indicate.

CPDN should grant some more because running a CPDN job to the (too often bitter) end eats up a lot of nerves and can cause quite some stress. They need a lot of HD space too but in this case the stress counts more I think. A crash after weeks of crunching is quite hard to take, even if you got your credits for it, a full run is the target.

CPDN is optimized btw., if their reference computer does not support SSEx, the granted credits per trickle on a modern computer might even be correct by SSE-benchmark.


Afaik. the SSE thing applies to Einstein too, the server grants about the same amount of credits that an SSE benchmark without any further optimisations would ask for.


SIMAP does use SSE but as they don't have fixed credits, their average is always a bit behind those SSE projects with fixed credits.

OK, I'm officially done collecting samples for Rosetta. I think I have enough to say that for my computers rosetta grants about the right amount of credit with the new credit scheme. It's a bit high, but nothing that I can't live with.

For the most part, I'm in agreement with you Tony, 100% of the way.

If you look at my systems, all the Windows ones (with one exception) match your statement, the exception being an old notebook that took a pasting in the new credit system.

On a more positive note, the two Linux systems are finally getting fair share. So overall I'd rate it thumbs up.

Something that is worth noting is that I run a 12 hour WU duration, which according to several posts I've seen, does alot to help smooth out the differences.

---

Hi. Could somebody check the numbers for my laptop (id#199825)? It was turned off for a couple of weeks, so I'm sure the RAC doesn't mean anything but it has been granted less credit than it has been claiming. This surprised me, because I was under the impression that Macs usually got a little less than they deserved in the old system. The client is standard, so I won't try to blame anything on the new credit system, but I was wondering if what I'm seeing is what I should expect to see.

---

Hi. Could somebody check the numbers for my laptop (id#199825)? It was turned off for a couple of weeks, so I'm sure the RAC doesn't mean anything but it has been granted less credit than it has been claiming. This surprised me, because I was under the impression that Macs usually got a little less than they deserved in the old system. The client is standard, so I won't try to blame anything on the new credit system, but I was wondering if what I'm seeing is what I should expect to see.

What are the specs exactly? I've never been able to completely decipher the BOINC code for macs.

One of mine (185730) looks similar, but not exactly the same. But your credits seem too small to me. At first glance, I would point the finger at your memory, only 256mb. What does MenuMeters say? Is your memory maxed out when running BOINC?

---

Reno, NV
Team: SETI.USA

seconds old cred new cred new cred/old cred
103,608.30 98.39 33.37 0.339160484
With a little over a day's worth of crunching, you've got about the same ratio of new/old credits that I've seen with the standard client on the non ppc macs.

Take a look at David Kim's comments from today in this thread: Discussion of New Credit System2
It's also mentioned here.
as well as the comparisons I did here.

---

What are the specs exactly? I've never been able to completely decipher the BOINC code for macs.

One of mine (185730) looks similar, but not exactly the same. But your credits seem too small to me. At first glance, I would point the finger at your memory, only 256mb. What does MenuMeters say? Is your memory maxed out when running BOINC?

CPU is 800MHz G4 and 256MB memory. I'm not sure what MenuMeters is but no, memory is not maxed out when running BOINC.

@Benny: So lower credit than claimed is about the same across the board for macs? Or just certain models? I don't understand the numbers you posted.

Thanks for your help all.

[edited: orthographic mistake]

---

The PPC Mac I quoted is getting around 0.545 of the credits it got from the standard client in the old credit system. The older, non PPC Macs are getting around 0.30-0.35 of the credits they got from the standard client in the old credit system. (The last post with Mac quotes was with a non standard client.)
It shows that there's a need for Mac optimization experts to take a look at the older Rosetta code and offer suggestions as to how to improve the the performance of the Mac applications.

Right now, the PPC folks have to work about twice as long/hard, while the older non PPC Macs are having to work about 3 times as long/hard as you used to for the same amount of credits you got in the past.

---

Hmm. Well mine is a PPC but only getting about a third of claimed credit. Could it be an older cpu (G4) and older OS (panther)?

In the end, I don't really care - I don't do this for rank or anything. As long as I see SOME credits coming in, it doesn't much matter how many. Mostly, I'm just curious.

Oh, one more question. So was I wrong that macs generally got less than they deserved under the old credit system?

---