TPU's F@H Team

[t_ski]

For those of you that might be interested, I listed my WCG rigs for sale in the forum. There's a 4P G34 rig and a Dual Xeon rig available.

BTW, I forgot to mention the discount that I normally give to crunchers would apply to serious folders, too.

 

[thebluebumblebee]

Is there any way we can get 0x17 wu's consistently?

Nope. We get whatever they give us.

 

[thebluebumblebee]

Consider this a commentary:
On January 15, 2014, Vijay Pande announced the end of BigAdv effective January 31, 2015. At the time, the news was met with surprise and anger, especially by those who had made the investment in systems capable of running BigAdv. (if you're really bored: https://foldingforum.org/viewtopic.php?f=16&t=25411) I saw posts from people who stated, with much vitriol, that they'd had enough, that they were done, that January 31, 2015 would be their last day (or even sooner). But, did Pande do the DC community a favor? Did he know something that he couldn't say? Did he know that in 9 months GPU's would come along that would make those systems look downright energy inefficient? (even the GTX 780 and 780 Ti are better at PPD/watt) I remember trying to read his reason for ending "the BA experiment" and it didn't make sense to me. Now, looking at sub 200 watt GPU's that get ~300,000 PPD, it does. Imagine how someone would feel if they had built a BA system last month, when the GTX 9x0's came out? Buck has said that his 4P system uses about 800 watts, and IIRC, gets about 750,000 PPD. A person should be able to set up a system with 4 GTX 980's in that 800 watt envelope that would get ~1,200,000 PPD (conservative estimate - I can't find any information for the 980 as I write this) That's a 37.5% increase in PPD/watt. I often feel like PG yanks its contributors around like a dog on a chain, but this time they may have yanked in the right direction.

 

[mstenholm]

Consider this a commentary:
On January 15, 2014, Vijay Pande announced the end of BigAdv effective January 31, 2015. At the time, the news was met with surprise and anger, especially by those who had made the investment in systems capable of running BigAdv. (if you're really bored: https://foldingforum.org/viewtopic.php?f=16&t=25411) I saw posts from people who stated, with much vitriol, that they'd had enough, that they were done, that January 31, 2015 would be their last day (or even sooner). But, did Pande do the DC community a favor? Did he know something that he couldn't say? Did he know that in 9 months GPU's would come along that would make those systems look downright energy inefficient? (even the GTX 780 and 780 Ti are better at PPD/watt) I remember trying to read his reason for ending "the BA experiment" and it didn't make sense to me. Now, looking at sub 200 watt GPU's that get ~300,000 PPD, it does. Imagine how someone would feel if they had built a BA system last month, when the GTX 9x0's came out? Buck has said that his 4P system uses about 800 watts, and IIRC, gets about 750,000 PPD. A person should be able to set up a system with 4 GTX 980's in that 800 watt envelope that would get ~1,200,000 PPD (conservative estimate - I can't find any information for the 980 as I write this) That's a 37.5% increase in PPD/watt. I often feel like PG yanks its contributors around like a dog on a chain, but this time they may have yanked in the right direction.

I don't know if he knew but yes once the newer GTX's can get a steady stream of the right core WU's folding will be back to where it began - GPU's. I'm serious tempted to replace a 7970 or two for the coming winter. Ups, just ordered a MSI 970. Not cheap in Denmark - $470 incl tax ($777 for the GTX 980). Update in next week. I'm out of country atm.

 

[mx500torid]

no work for my nvidia cards again today.
And the 290x has a damn x16 less than 3000 ppd

 

[mstenholm]

no work for my nvidia cards again today.
And the 290x has a damn x16 less than 3000 ppd

This is why I waited this long to order a new nvidia.

 

[mx500torid]

Keep us posted on how that 970 does ms.

 

[mstenholm]

Keep us posted on how that 970 does ms.

Sure will. It was dispatched today and I will return home on Saturday. Base-line and a gentle OC in a i7-920 rig if the right WUs turn up.

 

[thebluebumblebee]

Keep us posted on how that 970 does ms.

There are some results at: https://docs.google.com/spreadsheet/pub?key=0AmQ6YZGMTUpFdHhxNkptVncyR2k4dVV6WDBmQXRJYVE&gid=0

 

[mstenholm]

There are some results at: https://docs.google.com/spreadsheet/pub?key=0AmQ6YZGMTUpFdHhxNkptVncyR2k4dVV6WDBmQXRJYVE&gid=0

I can see that $ilent push it far beyond what I consider a mild OC . My results will be a at "it can last at least one year" clock.

 

[mx500torid]

yes my thoughts exactly Ill go by what ms gets.

 

[bubbleawsome]

Also, a "mild" OC on maxwell may well be higher than what many people are used to. I could actually see them pulling an AMD and pushing out that clockspeed on another card. Say a GTX 980 1.5Ghz edition kind of card to go against anything AMD puts out. These clocks are pretty conservative for a top card.

 

[Jstn7477]

Due to growing parental agitation concerning electricity costs, I have begun removing my 6 Fermi cards from service at work and should be done tonight. I only kept them running as I wanted them to die off as part of my farm evolution cycle, but they seemingly survived well. The Radeons are staying on as they are 6-8x better than a Fermi (excluding the HD 7770), so overall I am probably forgoing the power of one Tahiti chip but losing 1+kW of power consumption. I may wait a bit and buy a GTX 970/980 for my home Haswell-E system and move the R9 290 space heater to work.

 

[johnspack]

My occasional reminder post.... a lot of cancer can be prevented, or postponed, if only people would quit smoking. Encourage those you know to quit smoking, or try electronic cigarettes and quit smoking. Gum, patches, even the drugs, do not work for all. E-cigs can save lives, try one today!

 

[mstenholm]

yes my thoughts exactly Ill go by what ms gets.

Came home and kissed the wife and installed my new MSI GTX 970 in a i7 970 rig. All is for now run at stock clocks.

The first WU failed (core 15, MEMtest) so did the second (a core 17) and the third and THEN I installed the drivers . The first WU is a core 15, 7622 and that is of course low yield points wise but since they are in the mix you get that figure as well. 68k PPD, 51% fan speed in a 21 C room and I can't hear it above my NH-14D run at 9V so pretty quiet. The power readings will come at a later state once I find a good time to power down the rig and install my meter.

 

[Nordic]

Some of you might remember my experiment recently.

Ok. Awhile back, @thebluebumblebee you were wondering how low of a cpu one can use for folding.

I crunch my 750ti/s with an Intel celeron g1620 2.7ghz dual core
1 and 2 gpu's it is maxing out the gpu's and feels responsive. I know it maxes it out, because I have folded with all the hardware the same except instead an overclocked 2500k.
With 1 gpu, I can crunch on both cores with WCG and not harm folding ppd.
So I just tonight dropped to 1 core at 2.7ghz. Felt responsive. Appears not to harm folding performance.
Still with 1 core, I dropped the clocks to 1.6ghz. System is not very responsive even without folding. I will leave in in this configuration for a few days to see the effect. It will be hard to tell at first because at the moment I have 0x15 low point work.

So folding with a 750ti can be a bit of roulette between 0x17 and 0x15.

So in response to overclocked over not overclocked.

Spoiler: hard numbers

I fold at 1300mhz core and 1525 memory which is the highest folding stable clock I can achieve at stock volts. With cpu at a normal dual core 2.7ghz and folding with 1 gpu overclocked as such I get a total system power usage of 88.8w at the wall. With out the gpu overclock I use 82.5w at the wall. 88.8w-82.5w= 6.3w

Gpuz data shows I use ~89.8% of tdp when overclocked which equates to 60w*.898= 53.88w overclocked. Not overclocked I see a ~80.6% tdp which equates to 60w*.806= 48.36w not overclocked.
53.88-48.36= 5.52w.

So to summarize: my overclock of 1300/1525 gives me ~6w more power used for what I believe is 5k-15k more ppd with a 0x17 core work.

My exact gpu model is this.

With 1 core at 1.6ghz it did not seem to have an affect on the ppd of 0x15 work. Still though at 1.6ghz the desktop was laggy even at idle. So james888's recommendation for lowest power cpu one can use would be an undervolted 2ghz or better sandy bridge core per gpu.
The cheapest intel desktop cpu you can buy right now is an Intel Celeron G1840 2.8ghz haswell for $46.99. You could pair this with the cheapest dual pci x16 motherboard which at the time of writing is an msi h97 guard-pro for $99.99. Paired with two gtx 970's and you would have the cheapest most performance per watt folding system.

 

[thebluebumblebee]

Some of you might remember my experiment recently.


With 1 core at 1.6ghz it did not seem to have an affect on the ppd of 0x15 work. Still though at 1.6ghz the desktop was laggy even at idle. So james888's recommendation for lowest power cpu one can use would be an undervolted 2ghz or better sandy bridge core per gpu.
The cheapest intel desktop cpu you can buy right now is an Intel Celeron G1840 2.8ghz haswell for $46.99. You could pair this with the cheapest dual pci x16 motherboard which at the time of writing is an msi h97 guard-pro for $99.99. Paired with two gtx 970's and you would have the cheapest most performance per watt folding system.

Past experience has been that faster cards put a bigger load on the CPU. Also, there's virtually no CPU load with Core_15 WU's. (they're working on Core_19!) IIRC, that conversation was more about running 1-2 750 Ti type GPU's. I think I'd want a fast i3 to keep up with a couple of GTX 9x0's. Like this: http://portland.craigslist.org/wsc/sop/4648210685.html

 

[mstenholm]

GTX 970 folding with client-type advanced- after two core 15 units and low scores I got ten 9406 core 17 in a row. They all failed before they even began:

00:05:55:WU00:FS01:0x17:ERROR:exception: Force RMSE error of 418.296 with threshold of 5
00:05:55:WU00:FS01:0x17:Saving result file logfile_01.txt
00:05:55:WU00:FS01:0x17:Saving result file log.txt
00:05:55:WU00:FS01:0x17:Folding@home Core Shutdown: BAD_WORK_UNIT
00:05:55:WARNING:WU00:FS01:FahCore returned: BAD_WORK_UNIT (114 = 0x72)

 

[thebluebumblebee]

What version of the core are you using?

 

[Nordic]

Past experience has been that faster cards put a bigger load on the CPU. Also, there's virtually no CPU load with Core_15 WU's. (they're working on Core_19!) IIRC, that conversation was more about running 1-2 750 Ti type GPU's. I think I'd want a fast i3 to keep up with a couple of GTX 9x0's. Like this: http://portland.craigslist.org/wsc/sop/4648210685.html

I would go for that overclocked pentium, or does it benefit from the i3's hyperthreading? I would think faster core clock would matter more.

 

[mstenholm]

What version of the core are you using?

52 but I can see that somebody else https://foldingforum.org/viewtopic.php?f=66&t=26836
has a similar problem with 55.

 

[thebluebumblebee]

Aren't 9406's old?

 

[mstenholm]

Aren't 9406's old?

9401 is old.
I tried to un-installed FAH and restall but I ended up with version 52 again and a new error. Any hints?

 

[thebluebumblebee]

@james888 is the only other one running Maxwell that I know of.

 

[mstenholm]

@james888 is the only other one running Maxwell that I know of.

It seem to be a fairly common problem with 9x0 and 9406 https://foldingforum.org/viewtopic.php?f=18&t=26807#p269098