Personal computing discussed
Moderators: renee, farmpuma, just brew it!
Ragnar Dan wrote:My questions and suggestion assuming the answers to them remain unanswered.
Edit: BTW, I read the other day that apparently that 260-216 can use up to 282 watts all by itself. Sheesh.
Gerbil Jedidiah wrote:Ragnar Dan wrote:My questions and suggestion assuming the answers to them remain unanswered.
I have no idea what you meant by this. Yes I'm slow. Maybe you mean rhetorical?
Gerbil Jedidiah wrote:Edit: BTW, I read the other day that apparently that 260-216 can use up to 282 watts all by itself. Sheesh.
Not while folding though I imagine. I have 720 shaders total on a 750w PSU and system runs fine. I'm actually a bit amazed at the load this PSU is handling... GTX295, GTX280, and i7 920 @ 3.6GHZ all 100% utilized... Rig just purrs along, if only the SMP software would catch up...
Ragnar Dan wrote:I asked if you had the SMP clients set up such that they were all launched by shortcuts specifying processor affinity.
Flying Fox wrote:GJ, your experience pretty much sums up why the TR community is not as involved as before and the result that we are dropping fast in the rankings. It's the sad state of affairs of these new fangled clients that even experienced folders finding the problems a little too troublesome for our taste. Sure we are more technically minded and are not afraid of changing settings here and there, but there is a limit on the amount of trouble we are willing to tolerate just to keep the clients up. Until Stanford addresses some of these issues I don't think our participation can improve a whole lot, although the huge points advantage of the GPU client can offset the loss in points from farms whenever we get people with single machines with single GPU to join up.
Why the **** do we have to babysit so much? I just got a couple of 2665's EUE'ed on me too. Seems like this box can only do 2653's but it never bothered to try the A2 WUs. Stupid assignment server.
/rant
Gerbil Jedidiah wrote:~Closing one Win SMP client crashes the other~
farmpuma wrote:Gerbil Jedidiah wrote:~Closing one Win SMP client crashes the other~
IIRC, the SMP client spreads out over as many processors as it can find, so when running two SMP clients you are sharing already shared (HT) cores. I imagine Windoz gets confused and just shuts them all down causing the second client to crash.
How many PPD do you lose when running only one Win SMP client instead of two? A single SMP WU crunching on eight processors should crunch fairly fast and it might increase your GPU PPD.
Flying Fox wrote:farmpuma wrote:Gerbil Jedidiah wrote:~Closing one Win SMP client crashes the other~
IIRC, the SMP client spreads out over as many processors as it can find, so when running two SMP clients you are sharing already shared (HT) cores. I imagine Windoz gets confused and just shuts them all down causing the second client to crash.
How many PPD do you lose when running only one Win SMP client instead of two? A single SMP WU crunching on eight processors should crunch fairly fast and it might increase your GPU PPD.
The default (-smp 2) setting spawns 4 instances of FahCore_A?.exe. In theory you should do (-smp [cores]) to maximize CPU usage. The WUs should finish faster but they may not be as much points as running multiple SMP clients, similar to the situation of HT'ed P4's with 2 single core clients.
Gerbil Jedidiah wrote:Flying Fox wrote:The default (-smp 2) setting spawns 4 instances of FahCore_A?.exe. In theory you should do (-smp [cores]) to maximize CPU usage. The WUs should finish faster but they may not be as much points as running multiple SMP clients, similar to the situation of HT'ed P4's with 2 single core clients.
Here is my client config line. How should I change it?
[settings]
extra_parms=-smp
Ragnar Dan wrote:Since it doesn't meet the specs Newegg lists for the card, they should be embarrassed about it and refund my return shipping cost. But they won't. I'm fairly pissed they did this.
Ragnar Dan wrote:Well I haven't talked to them yet. I'm just assuming, given the pissant warranty paragraph they show about open box products, that they'll do as little as possible, but hopefully you're right. I'll call them tomorrow, or maybe email them in the next hour or so...
I found my reboot sped up the 511-point WU, though, so there's one bright spot. It went from 2:56 to ~1:45 per % (4204 PPD). I think it must have been the installation of VMware or whatever the other thing I can't recall that was slowing it down for whatever reason.
Ragnar Dan wrote:FF: I agree that's a problem, but why does it seemingly affect TR's team more than others?
We've been left in the dust by 4 teams in the last year or so. Do you think that it's a simple matter of others recruiting a lot more than we are and even though they're having a similar rate of teammates quitting it doesn't affect them as much?
Flying Fox wrote:Ragnar Dan wrote:FF: I agree that's a problem, but why does it seemingly affect TR's team more than others?
We've been left in the dust by 4 teams in the last year or so. Do you think that it's a simple matter of others recruiting a lot more than we are and even though they're having a similar rate of teammates quitting it doesn't affect them as much?
There are lots of factors, I have a few points that hopefully shed some light in the picture:
- The TR community itself is smaller than the others. This is a fundamental issue that is not going to change any time soon. Compare to other sites like AnandTech and [H] we simply don't have as big communities as theirs. To solve this we need to dramatically step up our recruitment and retaining efforts. Problem is some of us old-timers are somewhat elitistknowledgeable and massive recruitment may mean dumbing down the audience. Notice the relatively "meh" reaction to TR joining the new fancy stuff like Facebook and you can get a glimpse of what kind of people usually hang around here. /no-offense-just-an-observation
- Now, we may be small, but the fact that we have been hanging on for so long is a testament to something. I did a quick Excel exercise and it showed. I took the top 25 teams from EOC and calculated the "total points per active user". TR is at #5 at 2.8 million points per active user. If you look at the 5 teams above us they are near the bottom actually. So we just have a relatively small group but they are already pretty productive.
- The other teams that are surpassing us most likely have a lot of members folding on one or 2 machines. We seem to have lots of people with a farm of some sort, thus jacking up the per user production numbers. This can be a focus of our next recruitment drive in which we ask people to fold on one GPU on one computer. Hopefully we can get more people this time.
- The TR community also prides itself in living in the "bang for buck" realm. However, this hurts us in our folding efforts because most of us are more cost conscious and in general are just more mature. So the hotter weather, a/c costs, etc. all lead to people not folding as much as they can. Just look at how UGN and other big-time folders downsizing their farms and we can see how much that hurts us.
- Some of our past leaders have farms of CPU nodes, but the SMP client is just in such a bad state that they have quit altogether (drfish and leor are the prime examples). For the corporate folders GPU folding is usually not accessible to them.
- When we lost drfish we lost our de facto captain so to speak. UGN, FBW, FH, and PinTO have been carrying the torch but with such a disparity of points among them, it does not capture the imagination. We need more mid-range gauntlets!
- Our other "drumming up support" efforts fizzled most of the time. We need to improve on that.
Flying Fox wrote:The default (-smp 2) setting spawns 4 instances of FahCore_A?.exe. ...
F@H SMP FAQ wrote:The SMP client was originally intended for multi-core CPUs, which generally do not support HT. For machines with 2 physical CPUs, we do recommend enabling HT for the SMP client as this presents the operating system with what looks like 4 logical processors (and our SMP client is intended for 4 processors). If you have 4 physical CPUs, we recommend against using HT, as this presents the operating system with 8 logical processors, which will make the SMP client run inefficiently (especially since the logical processors coming from HT run much slower than the normal ones).
Ragnar Dan wrote:Well I haven't talked to [Newegg] yet. I'm just assuming, given the pissant warranty paragraph they show about open box products, that they'll do as little as possible, but hopefully you're right. I'll call them tomorrow, or maybe email them in the next hour or so...