Old timer ramping up a bit

I've been folding for many years. But it's been a long time since I spent any time on it.

I have four dual core opterons rigs (4 cores each) that I was running ten 5.04 clients on.
Today, I converted all of then to smp clients. Three are windows boxes, one is linux.
None have cards capable of running GPU clients.

I also have one 8 core socket F board on the burner (providing I can get it stable) coming up.

I have some questions since I haven't followed the projects for awhile.

1. Is it worth adding a GPU client to any of the smp rigs? If so, what is a good bang for the buck card?

2. What are you all using to monitor your clients these days (cross platform please)?

3. Any improvements I could make to my setup?

TIA. Feels good to pay attention again. I've had a run of friends/family run into medical issues lately. Some survived, some didn't. I guess that happens and only gets worse as you get older.

1: yes
2:hfm.net
3: That depends how deep you want to get into it and how much you want to spend.

Fold on!

Good solid answers there.....

Add a few more details to what rigs you are running, and how much you are willing to spend, and we will gladly chime in with ways to improve on what your current hardware is doing, and where your money is best spent

make sure you get a pass key for your bonus points.

1. Is it worth adding a GPU client to any of the smp rigs?

Click to expand...

In terms of production, definitely, you'll boost each system's PPD substantially if you add at least one GPU client to each. More importantly, you'll be greatly diversifying and increasing the research being accomplished. Spread the cards across your machines if you add more than one to reduce SMP client cannibalizing.

If so, what is a good bang for the buck card?

Click to expand...

GTS 450 or GTX 460 seem to be good bets.

2. What are you all using to monitor your clients these days (cross platform please)?

Click to expand...

Everyone except me is using HFM.net.

3. Any improvements I could make to my setup?

Click to expand...

What frequency are you running the Socket F machine?

TIA. Feels good to pay attention again. I've had a run of friends/family run into medical issues lately. Some survived, some didn't. I guess that happens and only gets worse as you get older.

Click to expand...

Sorry to read about this. Best of luck.

Okay, some interesting answers

I'll definitely give hfm.net a shot.

The four machines are all K8N-DL's with 280 cores.
Will the GPU client hurt their production much?

Passkey is in for all four.

The socket F is two 2350's, so 2Ghz. I gave up on the current board and just ordered another.
I have two dual core hanging around that I might try in this board.

What am I willing to spend? Haven't really thought much about it.
From the answers here, it appears I should get at least 1 GPU's per box.
I don't really have enough information to decide if I should more than one on all four.

An SR-2 is out of the question, SWMBO would not approve.
I have been eyeing a couple C32 rigs though....

To give you some idea of how long it's been, I have several Athlon XP's in my cabinet with scratched traces and conductive paint on them. Sadly all the old MP boards have blown caps and are no longer worth goofing with.

dwdawg said:

The four machines are all K8N-DL's with 280 cores.
Will the GPU client hurt their production much?

Click to expand...

That depends which client and which WUs. They vary tremendously. Stick with the GPU3 client whatever series card(s) you get. There are some WUs received by the GPU2 client that affect the SMP client quite dramatically, and I'm trying to migrate all my systems to the GPU3 client. Unfortunately some cards like the 200-series cards do best with the GPU2 client.

The socket F is two 2350's, so 2Ghz. I gave up on the current board and just ordered another. I have two dual core hanging around that I might try in this board.

Click to expand...

What's wrong with the board and which one do you have?

What am I willing to spend? Haven't really thought much about it.
From the answers here, it appears I should get at least 1 GPU's per box.
I don't really have enough information to decide if I should more than one on all four.

Click to expand...

I think your budget will determine how many cards you'll ultimately get. Basically, the more GPU clients you add to a box with an SMP client, the slower the SMP client will run because GPU clients utilize a portion of a CPU core for their own WUs. So, you can see where this leads with additional cards.

To give you some idea of how long it's been, I have several Athlon XP's in my cabinet with scratched traces and conductive paint on them. Sadly all the old MP boards have blown caps and are no longer worth goofing with.

Click to expand...

I sill have several of the same processors in storage, and all my boards for them died. The last one was a year ago but it ran for 8 years.

Looks like Apollo and the others got you covered on the info.

dwdawg said:

What am I willing to spend? Haven't really thought much about it.From the answers here, it appears I should get at least 1 GPU's per box.
I don't really have enough information to decide if I should more than one on all four.

Click to expand...

Why not just start out with singles ? Then after you're off and running and your output is going as you'd like, decide which if not all you'll go dual GPUs with.

dwdawg said:

To give you some idea of how long it's been, I have several Athlon XP's in my cabinet with scratched traces and conductive paint on them. Sadly all the old MP boards have blown caps and are no longer worth goofing with.

Click to expand...

Yeah those were the days I gave away my last dual AthlonMP 2600 to a coworker who needed a PC last year. It had sat in the closet for about 4 years before that.

Sorry to hear of the issues with family. It is getting more frequent as the years tick by.

Name of the game is less rigs, but more power in those rigs.

i7s are a good choice. 920\930 is a good place to start.

Agree w/Mr Kendrak - incrementalism is bad. You will save money in the short term but end up paying for it in the electricity bills. And you will have less processing power to show for it.

thefreeaccount said:

Agree w/Mr Kendrak - incrementalism is bad. You will save money in the short term but end up paying for it in the electricity bills. And you will have less processing power to show for it.

Click to expand...

And let's not forget a fraction of the research when one eschews many clients for one or two. Although you people are correct in respects to overall production (points) and power consumption, let us not ignore _all_ the factors since this discussion is an open ended one. GPU clients, however much the hex-core camp may despise them with all the negatives perceived surrounding their role in this project, produce a TON of science.

APOLLO said:

And let's not forget a fraction of the research when one eschews many clients for one or two. Although you people are correct in respects to overall production (points) and power consumption, let us not ignore _all_ the factors since this discussion is an open ended one. GPU clients, however much the hex-core camp may despise them with all the negatives perceived surrounding their role in this project, produce a TON of science.

Click to expand...

If the size of the WU is any indication. The smp WU have a bunch more info in them than the GPU clients. The bigadv have well over 10x the information. And the standard isn't even in the same boat.

I'm on my phone or id look up the number of atoms in each type of WU. The SMP and even more so the -bigadv.

I know how you feel about it, but I think your counting all WU as =. And they are very different in the amount of information they have. Also, the more atoms that are together the more useful the information. They don't have to put together many WU and then send out new WU to make sure they fit well together (inspect the seams as it where).

Kendrak said:

If the size of the WU is any indication. The smp WU have a bunch more info in them than the GPU clients. The bigadv have well over 10x the information. And the standard isn't even in the same boat.

I'm on my phone or id look up the number of atoms in each type of WU. The SMP and even more so the -bigadv.

Click to expand...

True, I was not including the classic client WUs and emphasized GPU but we might as well admit standard SMP into this discussion on account of its size/complexity. I'm all for -bigadv research hence my 14 month travail with it, yet under no misconception what my machines are producing in amount of research. I have several -bigadv clients and currently a dozen GPU clients running. If one machine over 4GHz processing 24 threads is equivalent or superior in research, I will conclude my participation in the project forthwith. It belies my effort and 'raison d'être' to continue with the burden of futility that will inevitably weigh on my thoughts as I struggle to maintain a semblance of noble conviction what my hard fought efforts truly accomplish.

I know how you feel about it, but I think your counting all WU as =.

Click to expand...

It may seem that way and I understand how one might arrive to that conclusion about my posts, but I am being totally honest with you Kendrak, in no way, shape or form have I believed this to be true or intended to appear that way. I realize GPU WUs and classic WUs are incomparable to -bigadv on a one to one basis as MIBW correctly pointed out in the beta -bigadv thread weeks ago. My sticking point is the general misconception that WU value or client PPD = research value or research quantity. IMO, this impression is equally false. Otherwise, FLECOM's achievement was symbolic at best, nominal at worst. A single client no matter how fast or productive remains singular.

I would really like for someone knowledgeable in this debate, who is under the perception that one high powered -bigadv client can amount the same science (research quantity+import) as my small mixed farm (being of comparable production) that comprises a couple of slower -bigadv clients and 12 GPU clients, to demonstrate or prove their position incontrovertibly, so that I may be relieved of this 'falsity' once and for all.

APOLLO said:

...It may seem that way and I understand how one might arrive to that conclusion about my posts, but I am being totally honest with you Kendrak, in no way, shape or form have I believed this to be true or intended to appear that way. I realize GPU WUs and classic WUs are incomparable to -bigadv on a one to one basis as MIBW correctly pointed out in the beta -bigadv thread weeks ago. My sticking point is the general misconception that WU value or client PPD = research value or research quantity. IMO, this impression is equally false. Otherwise, FLECOM's achievement was symbolic at best, nominal at worst. A single client no matter how fast or productive remains singular.

I would really like for someone knowledgeable in this debate, who is under the perception that one high powered -bigadv client can amount the same science (research quantity+import) as my small mixed farm (being of comparable production) that comprises a couple of slower -bigadv clients and 12 GPU clients, to demonstrate or prove their position incontrovertibly, so that I may be relieved of this 'falsity' once and for all.

Click to expand...

I'd really like to see some insight into this as well. I've always held the belief that more WU's produced and returned contributes to the overall strength and diversity of the project, as opposed to one or two larger units. That is one of the reasons I've maintained my GPU and single core WU's, in addition to mostly A3's. Are we speculating that there is an equal amount of science that is accomplished in one -bigadv unit as say, four A3's, or maybe 50 GPU WU's?

Maybe that is a question that Xilicon can present for clarification to the board? If there is already some clarification and someone could provide some linkage to it, then it would be greatly appreciated.

Hasn't VJ posted that Big WUs turned in faster are a higher priority, hence the implementation of the bonus system.

My understanding is that: bigadv WUs are somewhat interdependent. Meaning the work we do today, affects the bigadv WU assigned next week. And that GPU WUs are not so much, rather made in large batches and analyzed less frequently. This would imply that bigadv WUs are more efficient in developing science, more targeted if you will, and that GPU WUs while lots of science is generated, that they are more of a shotgun effect.

Both are important to science, but they go about developing it two very differnt ways. Based on what has been said by VJ. If you have the capability of running bigadv fast, VJ would perfer it. if you can't regular A3 and GPU are the next most important, so if you can't run bigadv you should run that. Anything else should be left for older slower hardware. Think about the hardware we have available to us, and think about the hardware that the average user owns. We have (as demonstrated by our daily score and the number of SR2s available to us), we can run lots of bigadv. We have lots of really fast CPUs that might not be able to run bigadv, perfect for standard a3 WUs. We have lots of really fast GPUs, we can run lots of those too. Everything we do here is important, and our team has lots of ability to focus on the principal of VJs development. We should be leaving the v5 standard clients for other teams who can't do what we do.

Articles like this are the basis of what I am saying. I know that some of us do not respect what 7im has to say, but i think on this particular article is closer to the direction of the project.

*grabs the wheel and tries to steer this back on topic*

With the machines you have, a single GPU per board (looks like they have a single pci-e slot) would be a good choice. I'd look for deals on 460s, since they are probably the best bang for the buck right now. Running the smp CPU client, you are not going to hurt your production much at all with a single GPU.

The next step would be to see what you think about your production with this relatively easy and incremental upgrade path of GPUs. If you are happy with where you end up, that is fine. If not, you will need to start looking at swapping out the dual socket boards with newer archetecture. I'd wait and see what Sandy Bridge brings to the table this month. There may be some really good combinations in term of proce point and efficiency. Even if they are a folding flop (which I doubt), they will bring pricing down on the older I7s, which are still very respectable folders.

And, welcome back!

It will all depend on how fast you can turn the standard A3 work-units around if its possible to run -bigadv on your dual-quads.

If you can get around 15k PpD with them, ~10k with the slow work-units, then a -bigadv work-unit will only take ~2.5 days and net ~30k.
But at a guess you'll need to be clocked to near 3.4 Ghz to make that type of time.

My 920 i7 are clocked to ~3.8 Ghz and are pulling in the same PpD as my Skulltrails boxen did at ~3.4 Ghz.

Luck .............
[H]

Hehe. It's nice to see the passion people have for these projects.

All these machines I have are 3+ years old and I'll probably run them into the ground before I buy new. I'm a cheap bastard, I do the same with my automobiles and I never buy anything new

The socket F is a K9N Speedster and was an experiment. I picked the board up for 80 bucks and replaced the bios with an 8 meg version flashed for quad core. The is a thread on 2cpu that covers this. It just didn't work out for me. It ran for 2 days and then shut down. When I go to fire it up, it runs for 30 seconds or so, begins to post, and then turns off. I think the quad core procs are overheating the mosfets. It just can't handle it. So I found a server board that handles quad and hex out of the box for 70 bucks. I'll try the dual core F's in it. If it runs, I'll slap the K9N in a case and give it to my sister's kids and let them run it (saves me on my power bill)

I've been poking around, and it looks like I can probably drop a single Benjamin per board to upgrade to 460's. I can swing that, although I'll probably do one every couple weeks to spread the pain.

HFM is working out nicely. I haven't set up the share on the Linux box yet, but all the others are connected.

I must have a few other old clients out there somewhere. Standford says I had 20 clients reporting in the last couple weeks. God know where they are and what they are running on

musky said:

*grabs the wheel and tries to steer this back on topic*

With the machines you have, a single GPU per board (looks like they have a single pci-e slot) would be a good choice. I'd look for deals on 460s, since they are probably the best bang for the buck right now. Running the smp CPU client, you are not going to hurt your production much at all with a single GPU.

The next step would be to see what you think about your production with this relatively easy and incremental upgrade path of GPUs. If you are happy with where you end up, that is fine. If not, you will need to start looking at swapping out the dual socket boards with newer archetecture. I'd wait and see what Sandy Bridge brings to the table this month. There may be some really good combinations in term of proce point and efficiency. Even if they are a folding flop (which I doubt), they will bring pricing down on the older I7s, which are still very respectable folders.

And, welcome back!

Click to expand...

dwdawg said:

I must have a few other old clients out there somewhere. Standford says I had 20 clients reporting in the last couple weeks. God know where they are and what they are running on

Click to expand...

It is a nice thing to have some borgs out there.

jfb9301 said:

Hasn't VJ posted that Big WUs turned in faster are a higher priority, hence the implementation of the bonus system.

Click to expand...

To greater or lesser extent, this has always been true for every F@H projects but that's not what is at the heart of the controversy. We're arguing the comparative research value across different projects, and the overinflated crediting of -bigadv WUs for the tangible research being created.

Articles like this are the basis of what I am saying. I know that some of us do not respect what 7im has to say, but i think on this particular article is closer to the direction of the project.

Click to expand...

I do respect what 7im has to say even if we butted heads once or twice, but his article addresses the speed vs volume debate and this was not precisely the issue that incited the ongoing controversy. That is another discussion.

dwdawg said:

I've been poking around, and it looks like I can probably drop a single Benjamin per board to upgrade to 460's. I can swing that, although I'll probably do one every couple weeks to spread the pain.

Click to expand...

I think the GTX460 is the real winner right now in the GPU realm. Get one of the cheaper 768MB versions and let them rock. i have three of them in my farm and I am quite pleased.

jebo_4jc said:

I think the GTX460 is the real winner right now in the GPU realm. Get one of the cheaper 768MB versions and let them rock. i have three of them in my farm and I am quite pleased.

Click to expand...

Hehe, headed to Fry's tomorrow. Looks like 89 buck AR.
Might have to go four days in a row....

dwdawg said:

Hehe, headed to Fry's tomorrow. Looks like 89 buck AR.
Might have to go four days in a row....

Click to expand...

lol walk in buy one, walk out, walk in buy one.. those 1 item per customer rules are bogus you just cant walk up the counter with 3 or 4 of them but you can walk in and out of the store and buy as many as you want. i just dont trust the rebates from fry's.

It depends on who is paying the rebate more than the store where it was bought.
Posted via [H] Mobile Device