Folding Forum

Greetings to Bruce, PantherX, bollix47, 7im, Jesse_V, Zagen30, and all the other cool people who've responded to my past requests for advice/assistance.

I have a Samsung N145+ netbook with an Intel Atom N455 CPU at 1.66Ghz and memory upgraded to 2Gb running Windows 7 starter edition.

From the time I acquired it ( early 2011 ) I've been running the 6.23 console only client.

It currently is working on project 6897 ( 392, 0, 78 ) with a TPF of 34-35 min.

Task Manager shows it as a dual core processor.

With the advent of 7.1.52 I am wondering if I can possibly get more points per day from this unit by going with 7.1.52 SMP or 7.1.52 with Uniprocessor and then adding a second Uniprocessor client for the second core.

Please advise whether I should just stay with 6.23 or, after completing the current job go with 7.1.52 SMP , or 7.1.52 Uniprocessor and then add the second processor as a client.

If the recommendation is to go 7.1.52 Uniprocessor and then add the second processor as a client please explain how to add the second processor --- I can see on one of my rigs running 7.1.52 the "add client" button, but am unsure of how to proceed from there.

If a screenshot of Task Manager showing the "Performance" tab running the current 6.23 ( or any other info/data )is needed , please advise and I'll try to provide. ( If needed and you know how to get a screenshot in W7 starter edit. please advise procedure -- this is the only rig I have running W7 and I use Hoversnap to get screen shots on my XP rigs-- I don't know if it is W7 compatible --have never needed to get a W7 screenshot )

Regards to all and looking forward to hearing from you, barebear

Both the 6.23 client and the 7.1.52 client run the same cores, so choice of client does not affect ppd on a particular WU.

The main advantage of the new client is that it should detect what processors your system has and configure appropriately. If you have 2 cores, the v7.1.52 client should offer SMP, but you have the option of running 2 uniprocessor clients instead.

The question is whether the Intel Atom N455 processor is powerful enough to complete SMP WUs by the deadline.

See this thread: keep missing deadline after finishing 95-99% which suggests that SMP will be a problem, and that 2 uniprocessor clients may not produce much more ppd than a single uniprocessor client, due to the hyperthreaded cores sharing resources.

For derrickmcc.....

.........due to the hyperthreaded cores sharing resources.

I've been watching the Performance tab in Task Manager and 1 core seems to be running in the high 80's area while the other is in the high 50's.

2 uniprocessor clients may not produce much more ppd than a single uniprocessor client, due to the hyperthreaded cores sharing resources.

Based on this comment, reading through the link you sent, and seeing my current core performances, it would appear that I'm just as well off staying w/ 6.23 ?

BTW --- this system, like the other 6 here at home, is on 24/7 w/ no power saving mode or night shutdown

One of the benefits of the v7 client is that you can change the configuration easily and thereby you can test what configuration works best for you. Changing slots from 2x uniprocessor to 1 SMP and back again is not a problem. As mentioned before, v6 and v7 clients use the same cores so there is no inherent performance difference between the clients. So I vote switching to v7 and play around to determine what works best for you.

For P5-133XL.....

Thanks for your input !

Perhaps what I'll do is let the current V6 job finish, turn off the program ( it can just sit there since it's the console only version ? ), and play w/ V7.

From your comments and what derrickmcc suggests previously, and reading through the link he posted, it appears that there are suitable alternatives in terms of either 1 or 2 uniprocessor clients.

I get the feeling from reading through his reference that I shouldn't bother w/ SMP and am better off trying the 2 uniprocessor clients ?

Can I switch between SMP and 1 or 2 uniprocessor clients without hurting the project being worked on ?

For derrickmcc......

Any guess as to what improvement the 2 uniprocessor clients might yield considering the hyperthreading resource utilization ?

barebear wrote:Can I switch between SMP and 1 or 2 uniprocessor clients without hurting the project being worked on ?

For derrickmcc......

Any guess as to what improvement the 2 uniprocessor clients might yield considering the hyperthreading resource utilization ?

There is a difference between N uniprocessor clients running on N processors and 1 SMP client running on N processors. They use different techniques, and you can't switch between them in the middle of a project. If you would like to change it as P5-133XL suggests, then you have to do that in between WUs otherwise those uniprocessor WUs will be dropped.

I have both a netbook and an older laptopl The netbook is an Atom -- one similar but not identical to yours, with two HyperThreaded virtual cores sharing the resources of a single CPU. The Notebook is a true dual-core CPU.

With hyperthreading, Windows reports that it's only using half the system when it's running one uniprocessor WU. Iif you run two WUs, Windows reports that it's doing twice as much work but because of resource sharing it's really not, Look at the frame times wihen one client/slot has stabilized (at least three frames) and compare the frame times FOR THE SAME PROJECT when running two slots/clients. You'll find that each WU will run significantly slower when two are running than when only one is running. It will be somewhat less than half as fast . . . perhaps 180% of the single WU frame times.

If your assigned WU is using FahCore_78 (not eligible for a bonus) there will be about a 20% gain in points per day. If your assigned WU is using FahCore_a4 (eligible for the Quick Return Bonus), you'll earn enough of a bigger bonus to make it worth running only one WU and completing it faster. Returning the results faster is important to the science and the Pande Group is gradually converting all the uniprocessor projects to bonus-eligible projects.

Once again Thank You ! Jesse_V

For Bruce.......

You'll find that each WU will run significantly slower when two are running than when only one is running. It will be somewhat less than half as fast . . . perhaps 180% of the single WU frame times.

Make my day ? There was a typo and you meant 80% ? If not, I'm beyond lost. ( which in many things is nothing new for me )

On the assumption that there was a typo and unless you advise otherwise, I'm going to let the current 6.23 unit finish, stop the program, install 7.1.52 ( leaving the 6.23 console version just sitting in Program Files ) and run 1 uniprocessor.

Please do get back to me on this and let me know if I've got it right ?, thanks !

barebear wrote:

You'll find that each WU will run significantly slower when two are running than when only one is running. It will be somewhat less than half as fast . . . perhaps 180% of the single WU frame times.

Make my day ? There was a typo and you meant 80% ? If not, I'm beyond lost. ( which in many things is nothing new for me )

I think what he meant is the TPF will be 180%, which means it's almost half as fast. 80% TPF = faster. 180% TPF = slower.

Thank you iceman1992 !

barebear wrote:Once again Thank You ! Jesse_V

For Bruce.......

You'll find that each WU will run significantly slower when two are running than when only one is running. It will be somewhat less than half as fast . . . perhaps 180% of the single WU frame times.

Make my day ? There was a typo and you meant 80% ? If not, I'm beyond lost. ( which in many things is nothing new for me )

On the assumption that there was a typo and unless you advise otherwise, I'm going to let the current 6.23 unit finish, stop the program, install 7.1.52 ( leaving the 6.23 console version just sitting in Program Files ) and run 1 uniprocessor.

Please do get back to me on this and let me know if I've got it right ?, thanks !

Calling the time per frame of one WU running alone 100%, and then forcing two WUs to share the same floating point processor will take LONGER for each one -- with each frame taking somewhat less than twice as long -- hence 180%. Each one runs slower, but you do finish a total of slightly more WUs per day.

For Bruce.....

Thanks for getting back to me.

I'll accordingly start w/ 2 7.1.52 uniprocessor clients and report back with results.

It'll be at least a couple of days though ---still have 17 6.23 frames to finish and they're taking about 43 minutes each.

For Bruce and everyone else who has responded so far........

The great adventure has started.......

6.23 finished its job and is just sitting in my Program Files.

7.1.52 is installed with my adding a second uniprocessor --- Client Control in expert mode shows 2 uniprocessor slots with id's of 00 and 01 .

00 is working on project 6896 ( 78, 1, 18 ) and 01 is working on project 6888 ( 288, 2, 189 ) with respective credits of 135 and 69 pts.

The netbook window is so small that I had to look at the config screen on my big monitor to be sure of what was in the 2 boxes at the very bottom of the configuration pane.

In the few minutes that it has taken to write this immediately after the install, the 6888 project is already at 1.16% and showing a TPF of 26 min 13 sec with ETA of 1.80 days ( I'm a 2 finger typist ).

I think the ETA might be wrong considering the TPF estimate and amount of work already completed ? Comments anyone ?

I'll report back with data on the 6896 project when it shows something re TPF.

If anyone reading the above description of what I've done thinks I might not have done it correctly, please advise ASAP -- I'm a little unsure because I've never added a client before.

barebear wrote:00 is working on project 6896 ( 78, 1, 18 ) and 01 is working on project 6888 ( 288, 2, 189 ) with respective credits of 135 and 69 pts.

The netbook window is so small that I had to look at the config screen on my big monitor to be sure of what was in the 2 boxes at the very bottom of the configuration pane.

In the few minutes that it has taken to write this immediately after the install, the 6888 project is already at 1.16% and showing a TPF of 26 min 13 sec with ETA of 1.80 days ( I'm a 2 finger typist ).

I think the ETA might be wrong considering the TPF estimate and amount of work already completed ? Comments anyone ?

TPF is based on 1% completed, 100% will be ~43.67 hours = 1.8 days. This is equivalent to 38ppd.

Referring back to your original post (regarding the 6.23 client):

It currently is working on project 6897 ( 392, 0, 78 ) with a TPF of 34-35 min.

This would equate to 55ppd

So we just need a TPF for your project 6896 WU to see what you have gained in ppd.

Early predictions of TPF (and PPD) are generally less accurate that after several frames have been completed in a steady-state condition.