Maximum 24/7 Folding

[matdem1]

I've asked before and stil have not gotten the answers:

System Config:

Intel Core 2 Quad Q9300 @ 2.53ghz
4 gigs @ 800 mhz
Gtx 260 Core 216 (BFG Max core oc'd)

This will fold 24/7 Can I utilize all four cores seperately to fold using the CPU client Than have the GPU CLient working simulatneously... Does one core need to be free to feed the GPU? HOW DO I DO THIS?? I have downloaded and run the nvid gpu advanced client and am running at about 6000 ppd... Last night I downloaded and ran the CPU X86 client and simply ran it twice so it is appearing twice in my tool bar. Today I ran a third....

For example I've seen Point per day break downs for Quad cores ... Does this assume all four cores are folding seperately 24/7... and if so do they just add up so that if a quad core can average 3000 points per day and a GTX 260 can average 6000 ppd should my totals be 9000 ppd????????

Am I going about this the rite way.. Is it as simple as downloading and running the cpu client for as many CPU cores as are available or do I have to do more... Please be specific.. with a step by step if possible. I want to Max out this rig and get two more up and running by the end of next week.

My second rig is ready to go:
Core 2 E7300 @ 2.66ghz
2 gigs 800 mhz
8800 Gt

Again a 24/7 folder.... suggestions?

[ReAp3r-G]

before i had problems with the SMP client, i ran both SMP and GPU clients

short answer is no, you don't need one core to be free to run the GPU client, the CPU will allocate enough for all your processes to run (folding clients)

only thing you cannot do is to run multiple instances of SMP clients unless you're on a VM

hope that helps

[matdem1]

Thanks for the fast reply...

How do I run both clients... Do I simply double click on the X86 Folding CLient and assume it will be smart enough to pick a core on it's own?

Are you saying that in my case I can only run one instance of the GPU and ONe instance of the SMP (x86 cpu) client simultaneously?

[matdem1]

I just noticed you're using an HD 4670 ... How many PPD does it score?

[BoBzeBuilder]

I'm sure the experts will fill you in with the detail, but from my undrestanding, you need to run one instance of the GPU client and one SMP client. The SMP client should recognize and utilize the four cores available.

You seem eager to start folding, may I ask for what team?

[Patonb]

You'll need to setup the smp properly, including either deino or mc-whatever... If you're x64 you'lkl need deino.

Go the the folding d/l site and read the install, as it's not just double click.

Then jus run it with the proper flags.

[JPinTO]

The SMP client generates more PPD than 4 CPU clients, so if you are able to get that running that is preferrable.

As to whether you need to dedicate a core for the GPU, it depends on a lot of variables. Best to try it out sharing a core between the SMP and GPU clients... if your GPU PPD stays at 6000PPD then you are good to go. Otherwise, you can try playing with the priorities, and see if that helps before you have to dedicate a core to the GPU exclusively. Regardless, even on 3 cores, run the SMP client.

- JP

[matdem1]

Does the speed of the internet connection effect your PPD and/or ns/day .......Here's why I ask... When hard connected I seem to get a higher ppd/ns vs wifi connected???? I've moved my quad core/ GTX 260 From Upstairs in my house with a wifi connection getting about 250 ns but when hard connected downstairs it gets over 500 ns ... I've done this about four times and each time it folds faster when hard connected???

[Flying Fox]

Connection speed has nothing to do with points. Remember points per day is an average. And if you use one of the monitoring tools such as FahMon before the current WU is finished, you are looking at a projected average. What is all that? Well, for the benefit of future newbies here is an introduction.

A work unit (WU) is a basic chunk of work that a client needs to complete before getting awarded points (let's put aside partial credit for the moment for EUE'ed WUs). Each WU belongs to one of the many projects. A project is usually a particular protein molecule that needs simulation work done on it and WUs within a single project perform similarly and are assigned a point value. You can think of projects as the "type" of any WU. For now, single core CPU, SMP CPU, GPU and PS3 clients all work on separate projects so there is no apples-and-apples comparison across the platform. The only common thing is "points" (right now GPU and SMP projects garner a disproportionately amount of points).

Basically, you measure while the WU is being processed and you get 2 numbers: how much % of the WU was completed during your measurement period, and the measurement period in units of time:

points per day = (% done) / (time spent for that % in seconds) * points for the WU * 86400
(if you measure in minutes then adjust accordingly)

Now here is the kicker: the points system is supposed to level the playing field and provide an approximate scaling, but the current point assignment scheme is not as fair as one would hope. You can have WUs that are quick to finish and generate a lot of points, or you can get WUs that are slow to finish yet still not get much. And any client types have varying projects like that. So you will see mention of "WU fishing" where a user may trash the less attractive (ppd or total points wise) WUs and hope for the next assignment will give him a better WU.

When you quote ppd number, it is more useful to quote it with the project/WU type as the qualifier. That way comparison can be done easier. For yourself though, you probably need to gather points data from the various WU types that your client has processed over the few days, total up the points and divide by the number of days to get your average ppd on the particular machine. Arguably it is not very useful in forum discussions as no one's WU mix will be the same for any degree of sane comparison. So it is better to just quote ppd numbers of a particular WU type, with your folding client type. For example, "on my E2160 @2.7GHz I can do project 2605 with the SMP client for about 1100ppd". On the GPU side, I have seen 500-point WUs taking forever resulting in a much lower ppd than 300-point WUs that take a shorter amount of time to finish.

As for mixing GPU and CPU folding clients, like others have said you need to determine for yourself to see if it is worth it for you. For a mid-to-top end GPU, they get more points so definitely start with the GPU client first. Measure its ppd with a particular project and then install+fire up the SMP client. There should be a drop-off with the GPU client but you have to total up the ppd values for both clients to see it result in a net gain. New Nvidia drivers seem to reduce the CPU usage by a lot but I am not sure about the situation on AMD GPUs. You may also need to fiddle with affinity if things do quite work out with mixing clients. Worse case scenario if you have a GPU and a dual core CPU? You can run the GPU client taking up one core, and then a single core CPU client on another.

[matdem1]

My team number is 158474 and my team name is Value Technology.

My Name is Matthew Longstreet, My son is Matthew Longstreet Jr, My Mother is Emma Longstreet (has alzheimers), Denise Hughes

I have purchased three new Desktop Pc's and have purposely set them up for GPU folding with NVIDIA CArds (GTX 260,8800 Gt and 9800GT) but after adding my Mothers name, and my Friend Denise I do not see them on my team statistics page. Denise Hughes and Emma Longstreets names and Points per day do not appear on my team statistics page. This is very frustrating as I've invested a LOT of time and money.

I simply put my mothers name in the configuration where it asks for a name (it automatically said underscores are used etc) than inputted my team number in the appropriate box. Why aren't their numbers showing on my team?

Two are missing:
Emma Longstreet
Denise Hughes

What am I doing wrong and where are their points going?... Please help as two more machines are being built for folding and I want to make sure I set them up correctly for my points to count on my team.

[just brew it!]

Have the machines folding under those names turned in any work units yet? They don't show up in the stats until they've turned in at least one work unit.

[ReAp3r-G]

I just noticed you're using an HD 4670 ... How many PPD does it score?

Well using the 4670 alone i think my average would be somewhere between 1.5k to 1.8k per day...when i had the SMP client running with my PS3 as well i was averaging over 3k PPD...i'm not using the SMP client anymore because there is no definite answer for the problem i've been having with it so it's just my GPU client and the PS3 running, both of which are doing just fine...my contribution is steady and that's all that matters

hope you're having a good time figuring this out, it's not that hard really...MPICH SMP client is the easier of the two to use (compared to DEINO), but that's just my personal preference...i don't know if either one makes more points than the other, but IMO and IIRC i think both of them should have the same turnout

as for running it, just install the clients, follow the instructions carefully...and once you're done just use the shortcut (be it on the desktop or the start menu) and you're good to go! cheers!

[lou15214]

Matt, just checked F@H site and your team has 5 members. Go to F@H site, click on stats, then team stats, put team number in search box. then click on team name and it will show team members.

BTW I have 2 8800gt's folding with a ppd between 3000-5000 for each, depending on the WU. I'm using the advanced GPU systray client. fahcore11 I think.

[farmpuma]

I've asked before and stil have not gotten the answers:

Since this topic has been addressed numerous times in this forum your responses will likely be few as most regular readers know the information you are seeking is already available in other threads. We all try to maximize our ppd and our contribution to the F@H research. In fact, during the course of most threads this topic will likely be addressed. For instance, the 'UGN rise to the top' thread has a lot of good information.

I was very sorry to hear of your mothers condition and I wish you the best of luck with her care. The best advice I ever received was, "Hire a Medicaid lawyer when it comes time for the nursing home." Most Medicaid caseworkers seem to be quite good at finding ways to deny your claim.

My father died of alzheimer's on 15 January of this year after many years of my care and even better care in the nursing home. He had forgotten how to swallow properly and was aspirating food and liquids into his lungs. He did not respond well to the diet of thickened liquids and I am told his passing was pain free.

Does the speed of the internet connection effect your PPD and/or ns/day .......Here's why I ask... When hard connected I seem to get a higher ppd/ns vs wifi connected???? I've moved my quad core/ GTX 260 From Upstairs in my house with a wifi connection getting about 250 ns but when hard connected downstairs it gets over 500 ns ... I've done this about four times and each time it folds faster when hard connected???

What the heck is "ns"? While I would expect a wifi connection to use more system resources than a hard wired nic, I still don't know how much it is impacting your points production. Install FahMon or do the math to get valid numbers on your ppd impact.

So you will see mention of "WU fishing" where a user may trash the less attractive (ppd or total points wise) WUs and hope for the next assignment will give him a better WU.

I do my very best to not delete any work unit and I do not encourage any "catch and release' fishing practices. But, assuming the servers are aware of work units not being fully downloaded, there are many ways to avoid work units which are poorly matched to your system resources or internet connection. The most obvious being the server ip address which along with the project summary page will tell you what work unit you are downloading. On servers with good and bad work units one can use the work unit 'tells' to determine which is being downloaded. For instance, the good 2653 SMP work units are just under three megabytes while the bad 2665 work units (which take me two hours to upload) are about five megabytes and the good 353 point GPU work units download in about twenty to thirty seconds on my dial-up connection while the bad 511 point work units take about thirty-five to forty-five seconds. Good luck with your work unit fishing as it can be as frustrating if not more so than fishing for fish.

[matdem1]

Thanks to all of your input I have both of these machines "straightened out"... I will add two more folding machines in the next week or so. I would like to dedicate a machine to each person in my family who is or was affected by these dreadful diseases for which this project seeks to find a cure..

Thank you for all of your help!

[ReAp3r-G]

you're most certainly welcome Mat...cheers buddy!

[HurgyMcGurgyGurg]

What the heck is "ns"?

Its the performance indication given in the GPU viewer for the system tray client.

ns is nanoseconds a day. The problem is its only useful for a single WU, and is not related to ppd.

So while a big wu gets more ppd it may only be doing 100 ns a day. While a small easy wu may be doing 500 ns a day and still get less ppd.

My guess was he was on a slower wu when he was on the wifi and when he checked on the hard connection he had gone to another wu.

By the way, don't run the viewer while folding it takes up a lot of resources.

[farmpuma]

Thank you for the "ns" info. I haven't run a systray client since the early part of my time with fah4 and win98.

"If it isn't doing the folding, it's making the folding slower."

[kiaas]

I might be ordering a GPU sometime next month, so I can get a GPU folder up. CUrrently My Dual-core 2.8ghz Wolfdale is my highest contributor on my team(if counting only ppd..my iMac has been running long enough to hold onto #1 for a while)
I have a vested interest in adding resources for Folding@Home to use myself, as my current life expectancy is ~40years. (I have cystic fibrosis.)