Yessir, when I stop for a reboot all my nodes using athlon xp's are automatically detecting sse optimizations and utilizing them all on their own without any -forcesse flags. Hmmm, I wonder what this other flag does? hehe
KingFish
Well this much I can say for sure ...that flag will get you the very most unstable WU's if they are indeed unstable ...that's about it.
Even -advmethods still gets unstable WU's ...more unstable than no flags at all anyway.
You can really only count on no flags to get the most stable of WU's at this point.
Unstable WU's are the last thing I need at this point. Now that the heat wave has begun my farm nodes are having probs. This hasn't been a very good week for the KingFish Farm. I've had two nodes spitting back special_exit and early_unit_end this week. No mo' overclocking for most of my nodes for a while now. Production has been down for the past week or so but hopefully I'll have a rebound now that I've fixed some of those probs. The only flags I'm using is -service when applicable and -verbosity 9 and am plugging along quite well with them. I wouldn't want that extra headache at this point in time.
Unstable WU's are the last thing I need at this point. Now that the heat wave has begun my farm nodes are having probs. This hasn't been a very good week for the KingFish Farm. I've had two nodes spitting back special_exit and early_unit_end this week. No mo' overclocking for most of my nodes for a while now. Production has been down for the past week or so but hopefully I'll have a rebound now that I've fixed some of those probs. The only flags I'm using is -service when applicable and -verbosity 9 and am plugging along quite well with them. I wouldn't want that extra headache at this point in time.
KingFish
I agree ...the heat is effecting me as well at home as you might have figured. The farm is ok as long as the A/C doesn't go down ...they don't overclock anyway!!! The best production you can afford is the most stable I find.
Agreed csimon. I'm in the process of clocking the nodes back down to stock speeds and returning the voltages to stock as well. This one room may as well be called the boiler room. I had to break down and buy a window unit for this room to bring it to comfortable temps. With this many nodes I doubt I'll mess with OC'ing again as stability is better for managing the farm rather than having to resolve instability issues all the time while those aussies bogard all the production ranks ahead of me.
just a tad over stock voltage ...tomorrow morning before I leave for school I'll drop the fsb to 400 and it should be able to weather the heat. Gets soooooooo damn hot in here during the summer. I go to see the contractor this week about building a new house!!
The -advmethods switch is supposed to be used to request work units that are in the final stages of development, just prior to general release. It is not intended to express a preference for Gromacs or DGromacs work units.
There was a period of time (during the extended gradual introduction of the Gromacs cores) when -advmethods provided greater numbers of Gromacs WUs, and for SSE-enabled machines, a large increase in points production. Since Gromacs cores have been available for general release, use of the -advmethods switch no longer necessarily results in a competitive advantage. In fact, in some circumstances, -advmethods can reduce long-term points production.
Before choosing to invoke -advmethods, consider the following points:
The use of -advmethods is accompanied by a small risk of receiving an unstable work unit. The developers at Stanford ask that problems with work units be reported at Folding-community.org.
The use of -advmethods overrides the servers' algorithms for matching optimized software to advanced hardware. This can result, for example, in Athlon XPs and PIIs getting DGromacs units, when it's better for everyone if they go to SSE2-enabled machines.
Since the recent upgrade of the points determination system, there is no competitive advantage to Gromacs WUs over Tinker WUs. Although there is now such an advantage when a SSE2-enabled machine receives a DGromacs WU, the -advmethods switch will not, over time, result in an increase in the number of DGromacs WUs that are downloaded.
The -forceSSE switch was introduced to manually force SSE optimizations when running on AMD hardware. There was, at one time, an issue with AMD processors that resulted in instability when SSE optimiztions were used, so 3DNow! optimizations were used by default. This issue has since been resolved, and the most recent cores use SSE optimiztions by default on all SSE-enabled hardware.
Descriptions for all of the options can be found in the Folding@home Console Client: User Guide.
Very well said ...this is the point I was trying to make from the beginning of this thread but this info has been trickling down so gradually since this thread began.
Agreed csimon. I'm in the process of clocking the nodes back down to stock speeds and returning the voltages to stock as well. This one room may as well be called the boiler room. I had to break down and buy a window unit for this room to bring it to comfortable temps. With this many nodes I doubt I'll mess with OC'ing again as stability is better for managing the farm rather than having to resolve instability issues all the time while those aussies bogard all the production ranks ahead of me.
KingFish
Interesting thread. Sally might understand what is going on, me, i just keep lifting heavy weights, lol.
It appears a few of us have lower than normal production figures. Sally had a few probs with heat and at least one big WU not turning in or a few shutting down early. Too late to do a copy of the logs.
KingFish, i'll have a talk to Sally later and ask her to slow down a bit if you like, hehehe.
Going to shut a few puters down this week anyway, one at the time, to do maintenance. The new papst fans arrived and a few other things aswell. That is so i don't get bored on my week off, starting today. Sally shut them down for disk maintenance the other day so i guess in a sense we are all in the same boat.
I'm keeping the -advmethods switch on my P4's for sure. Even though they've reworked the points and are now using a 2.8 P4 for benching, the P4 rigs will still turn in a better point return by processing Gro and DGro work than Tinker, due to the relatively crappy native FP performance of their architecture. Since Tinker doesn't utilize any form of SSE or SSE2, the P4 has to depend entrely on it's native FP and is inefficient. It's much better to maximize it's processing efficiency by trying to do nothing but WU's that use SIMD instructions.
As far as for the -forceasm and -forcesse flags, they will start the client without checking for instability, which sounds bad at first glance. However, if for some reason your computer powers down without shutting down the client (power outtage, accidently unplug the wrong plug, etc) and you aren't using one of those switches, the next time you start the client back up it will disable any SIMD optimization use until the client has gone through one complete shutdown and restart with no problems. Even though your rig is perfectly stable, not running either -forceasm or -forcesse can slow your processing way down, so I keep that flag also in my startup. Since they've fixed the client for the SSE problem they were having with AMD procs, it's probably better to go back to strictly using just the -forceasm flag now, since it's possible that Stanford might drop the -forcesse flag support from future revisions of the client.
One other note; I still think I average a little better processing strictly Gro and DGro, by running -advmethods instead of getting what comes up by running no flag at all. I do know that I am running work that isn't quite tested enough to be in general release a lot times while running -advmethods, so anyone else that is using that flag needs to be aware of the fact that you can occasionally get bad work with it.
ok ...I have officially removed -advmethods from all of my machines ...you can monitor my progress if you like on the EOC stats or just ask me and I'll let you know how it's coming along.
Comments
Even -advmethods still gets unstable WU's ...more unstable than no flags at all anyway.
You can really only count on no flags to get the most stable of WU's at this point.
KingFish
KingFish
The -advmethods switch is supposed to be used to request work units that are in the final stages of development, just prior to general release. It is not intended to express a preference for Gromacs or DGromacs work units.
There was a period of time (during the extended gradual introduction of the Gromacs cores) when -advmethods provided greater numbers of Gromacs WUs, and for SSE-enabled machines, a large increase in points production. Since Gromacs cores have been available for general release, use of the -advmethods switch no longer necessarily results in a competitive advantage. In fact, in some circumstances, -advmethods can reduce long-term points production.
Before choosing to invoke -advmethods, consider the following points:
The use of -advmethods is accompanied by a small risk of receiving an unstable work unit. The developers at Stanford ask that problems with work units be reported at Folding-community.org.
The use of -advmethods overrides the servers' algorithms for matching optimized software to advanced hardware. This can result, for example, in Athlon XPs and PIIs getting DGromacs units, when it's better for everyone if they go to SSE2-enabled machines.
Since the recent upgrade of the points determination system, there is no competitive advantage to Gromacs WUs over Tinker WUs. Although there is now such an advantage when a SSE2-enabled machine receives a DGromacs WU, the -advmethods switch will not, over time, result in an increase in the number of DGromacs WUs that are downloaded.
The -forceSSE switch was introduced to manually force SSE optimizations when running on AMD hardware. There was, at one time, an issue with AMD processors that resulted in instability when SSE optimiztions were used, so 3DNow! optimizations were used by default. This issue has since been resolved, and the most recent cores use SSE optimiztions by default on all SSE-enabled hardware.
Descriptions for all of the options can be found in the Folding@home Console Client: User Guide.
Very well said ...this is the point I was trying to make from the beginning of this thread but this info has been trickling down so gradually since this thread began.
Enjoy
csimon
It appears a few of us have lower than normal production figures. Sally had a few probs with heat and at least one big WU not turning in or a few shutting down early. Too late to do a copy of the logs.
KingFish, i'll have a talk to Sally later and ask her to slow down a bit if you like, hehehe.
Going to shut a few puters down this week anyway, one at the time, to do maintenance. The new papst fans arrived and a few other things aswell. That is so i don't get bored on my week off, starting today. Sally shut them down for disk maintenance the other day so i guess in a sense we are all in the same boat.
As far as for the -forceasm and -forcesse flags, they will start the client without checking for instability, which sounds bad at first glance. However, if for some reason your computer powers down without shutting down the client (power outtage, accidently unplug the wrong plug, etc) and you aren't using one of those switches, the next time you start the client back up it will disable any SIMD optimization use until the client has gone through one complete shutdown and restart with no problems. Even though your rig is perfectly stable, not running either -forceasm or -forcesse can slow your processing way down, so I keep that flag also in my startup. Since they've fixed the client for the SSE problem they were having with AMD procs, it's probably better to go back to strictly using just the -forceasm flag now, since it's possible that Stanford might drop the -forcesse flag support from future revisions of the client.
One other note; I still think I average a little better processing strictly Gro and DGro, by running -advmethods instead of getting what comes up by running no flag at all. I do know that I am running work that isn't quite tested enough to be in general release a lot times while running -advmethods, so anyone else that is using that flag needs to be aware of the fact that you can occasionally get bad work with it.
:smokin:
But I may have a change of heart real soon!!!