TR Forums

All of a sudden 3 of my farm 'puters have the 2million step WU's. Any way to "skip" them. Probably not and its not helping them if I do. So I'll just cry about it hehe.

Yha, those are beasts

Yep, I'm losing almost 1500 PPD on my best SMP machine, a C2D 6400 OC'ed to 3152 MHz, because it's crunching through a 5102. If they could make that WU work with FahCore_a2.exe instead of the old core, FahCore_a1.exe, I'm sure it would run a lot quicker.

Yeah, I've noticed that there's a pretty huge difference in PPD between a1 and a2 core in general. Not sure if that's because the benchmarking is unintentionally skewed, or if they are adding bonus points to a2 WUs on purpose because it is still considered to be experimental.

Man, I'm testing my new work machine over the weekend and wham! A 5102 is what it got immediately. It's only a an Opty 1214 so I don't think it has much chance. I would actually like it to be on my E2160 rig so I can see how bad this bad boy is. Life sucks, I guess.

Euw. I use the uni core client x4, and the bigger point WUs take about the same length of time to crunch as the higher step, lower point WUs. It's the difference between 250ppd / core and 800-1000ppd /core.

Flying Fox wrote:

Man, I'm testing my new work machine over the weekend and wham! A 5102 is what it got immediately. It's only a an Opty 1214 so I don't think it has much chance. I would actually like it to be on my E2160 rig so I can see how bad this bad boy is. Life sucks, I guess.

Just checked on the work box and it takes 1.5 hours just to process one step of 5102. I have a feeling that it may not make it.

Flying Fox wrote:

Flying Fox wrote:

Man, I'm testing my new work machine over the weekend and wham! A 5102 is what it got immediately. It's only a an Opty 1214 so I don't think it has much chance. I would actually like it to be on my E2160 rig so I can see how bad this bad boy is. Life sucks, I guess.

Just checked on the work box and it takes 1.5 hours just to process one step of 5102. I have a feeling that it may not make it.

I'm wondering if I should tactically avoid this WU, and how I would go about doing such a thing. I've never tried to avoid WUs in the past, as I figure it's all needed research, but this guy is ridiculous.

I just killed my 5102 and DL'd a new WU. Luckily I got a 2653.

I was at 7% on the 5102, and it was taking 2 hours per step.

This is the part where you have to pick between philanthropy and e-peen. Nice show of colours, people.

Gerbil Jedidiah wrote:

I just killed my 5102 and DL'd a new WU. Luckily I got a 2653.

I was at 7% on the 5102, and it was taking 2 hours per step.

Oh which box? My "puny" AM2 Opty at 2.2GHz (can't beat the Core 2's) is doing 1.5 hours per step. Granted it is running on a Ubuntu VM (again 64-bit guest on 32-bit host FTW) so it may be that much faster than WinSMP?

Meadows wrote:

This is the part where you have to pick between philanthropy and e-peen. Nice show of colours, people.

And this is the part where your uninformed trolling makes you look bad. At 2 hours per step it means it will take 8 and 1/3 days, meaning that box will miss the 8-day deadline (at least they have the decency to give this monster WU a longer deadline than the usual 3-4 days) anyway. On my box it is going to finish in 6 and 1/4 days so I am going to let it finish. I'm testing my new box while I am slowly installing apps and moving data. So I have time.

Flying Fox wrote:

And this is the part where your uninformed trolling makes you look bad. At 2 hours per step it means it will take 8 and 1/3 days, meaning that box will miss the 8-day deadline (at least they have the decency to give this monster WU a longer deadline) anyway. On my box it is going to finish in 6 and 1/4 days so I am going to let it finish. I'm testing my new box anyway while I am slowly installing apps and moving data. So I have time.

I will humbly admit my ignorance, simply because the topic doesn't interest me in the slightest. I'm neither for philanthropy nor for e-member-enlargement.
I have tried seti@home in the past just to get a perspective of the idea, but I discontinued my efforts pretty soon because it was a rather hillbilly method of going about things. I trust aliens in our vicinity will be more than ready to take matters in their own body extensions.

About this project however, I will say this: some people might have slower machines but want to finish it anyway, so this 8-day deadline is a rather dumb move and should be changed to be in proportion to the work you're about to complete.
Now if you're one of the people who can complete those pieces, that's good, go get them tiger. Somebody should complain to... who or whatever is in charge of the project.

Meadows wrote:

About this project however, I will say this: some people might have slower machines but want to finish it anyway, so this 8-day deadline is a rather dumb move and should be changed to be in proportion to the work you're about to complete.
Now if you're one of the people who can complete those pieces, that's good, go get them tiger. Somebody should complain to... who or whatever is in charge of the project.

Stanford doesn't want just any machine folding these WUs. They are intended to be run on multi-core systems with lots of power. Slowpokes need not apply.

The deadlines exist so that Stanford will finish a project with ALL WUs folded. The longer they wait, the more likely something went wrong on the folder's side and the WU will never be returned. That's why Stanford has a deadlines (both prefered and final), so that they can re-release a WU to complete the research.

Also, I'm not sure it's much appreciated you coming in to a folding thread and trolling since you have no intention of being helpful. And with your e-peen remarks I suspect you're just a bit jealous we have the money to spend on this hobby.

Flying Fox wrote:

Gerbil Jedidiah wrote:

I just killed my 5102 and DL'd a new WU. Luckily I got a 2653.

I was at 7% on the 5102, and it was taking 2 hours per step.

Oh which box? My "puny" AM2 Opty at 2.2GHz (can't beat the Core 2's) is doing 1.5 hours per step. Granted it is running on a Ubuntu VM (again 64-bit guest on 32-bit host FTW) so it may be that much faster than WinSMP?

This is one of my vitual UNIX environments running on my 3.4GHZ Q9450. The 2653 WU is taking 30 min/step. I suspect something is wrong with that environment. Perhaps my GPU2 client is interfering with it. My second SMP client is moving along nicely at 1 step every 10 minutes. I'm off to do some tweaking

Gerbil Jedidiah wrote:

Stanford doesn't want just any machine folding these WUs. They are intended to be run on multi-core systems with lots of power. Slowpokes need not apply.

The deadlines exist so that Stanford will finish a project with ALL WUs folded. The longer they wait, the more likely something went wrong on the folder's side and the WU will never be returned. That's why Stanford has a deadlines (both prefered and final), so that they can re-release a WU to complete the research.

A lot of "large computer owners" don't own those machines to process hypothetical proteins. As always, the meat of the crowd is elsewhere.
I don't see any problems with dynamically setting deadlines such as 2 days for simpler tasks and 10 days (nothing more than a rough 12% extension) for these "5102" things or what you called them. Enabling more people to finish them in the first place may prove optimal over letting those few powerhouses handle it, but I'm nobody to guess the likelihood of that.

Otherwise, all they're creating is more issues for average people like you and potentially a lot of wasted electricity if you forget to abandon work you can't finish in time. This is no way to participate in an otherwise automated project, having to babysit the progress all the time.

Gerbil Jedidiah wrote:

Stanford doesn't want just any machine folding these WUs. They are intended to be run on multi-core systems with lots of power. Slowpokes need not apply.

One would assume their stance would be More the merrier. Why would they want to limit the number of machines able to crunch this WU?

lordtottuu wrote:

Gerbil Jedidiah wrote:

Stanford doesn't want just any machine folding these WUs. They are intended to be run on multi-core systems with lots of power. Slowpokes need not apply.

One would assume their stance would be More the merrier. Why would they want to limit the number of machines able to crunch this WU?

They have all kinds of WUs to make use of all different kinds of computers. These 5102 WUs just happen to be monster WUs that take a lot of processing power and aren't suited for all computers.

lordtottuu wrote:

Gerbil Jedidiah wrote:

Stanford doesn't want just any machine folding these WUs. They are intended to be run on multi-core systems with lots of power. Slowpokes need not apply.

One would assume their stance would be More the merrier. Why would they want to limit the number of machines able to crunch this WU?

The short-deadline WUs are on projects where they want quick turnaround, so that they can release the next round of WUs. The WUs they issue are determined in part by the results from previous ones, so waiting (say) 2 months before reissuing a "lost" WU to someone else can significantly delay their research.

So they are trying to balance getting more people to crunch these big WUs, versus getting them back as quickly as possible so they can start the next round of research.

In general, the deadlines aren't an issue. The only time I recall having trouble with missed deadlines was when I tried to run the Windows SMP client on an older dual Athlon MP (which is a bit of an oddball system these days). But there is also some variation in how well certain WUs run on particular hardware (things like cache size, memory bandwidth, AMD vs. Intel's FPU implementation can all come into play), so occasionally the deadline is set a little too tight.

Gerbil Jedidiah wrote:

Flying Fox wrote:

Flying Fox wrote:

Man, I'm testing my new work machine over the weekend and wham! A 5102 is what it got immediately. It's only a an Opty 1214 so I don't think it has much chance. I would actually like it to be on my E2160 rig so I can see how bad this bad boy is. Life sucks, I guess.

Just checked on the work box and it takes 1.5 hours just to process one step of 5102. I have a feeling that it may not make it.

I'm wondering if I should tactically avoid this WU, and how I would go about doing such a thing. I've never tried to avoid WUs in the past, as I figure it's all needed research, but this guy is ridiculous.

Don't try to game the system for points Jed. Once you start doing that, you are heading for the dark side of the force! Don't become a points whore!! Roll with the big crunchers. The points come around.

Besides, check out our numbers. We are heading in the right direction.

jeffry55 wrote:

Don't try to game the system for points Jed. Once you start doing that, you are heading for the dark side of the force! Don't become a points whore!! Roll with the big crunchers. The points come around.

Besides, check out our numbers. We are heading in the right direction.

It wasn't that I was a point whore, it was more that I wasn't going to make the deadlines. It turns out though that there was something wrong with my virtual environment. I deleted the old, created a new, and my points are WAY up

If I get another 5102 I expect I will be able to crunch it just fine

I agree that being picky about WUs is definitely going to the dark side

Welp, after a couple weeks of 1000-1100 PPD WU's on every single one of my farm computers save one or two every now and then, I just noticed today they are all back crunching 2500+ PPD WU's. *knock on wood* hope it does those for a little while now. I also noticed my GPU2 client, after spending most of its life crunching 4000 and 8000 step WU's just downloaded a 20,000 stepper. Never seen one of them before. heh

The 5102's may be killers, but the 5103's & 5777's are delicious!!

My GPU's are being dominated by the crappy 5903/5904 1888-point WU's, and they really waste time on my machines.

Ragnar Dan wrote:

My GPU's are being dominated by the crappy 5903/5904 1888-point WU's, and they really waste time on my machines.

Yeah, that's what I meant to say! In that the 5903,5904 & 57.. series of WUs are giving me big points on my farm. My G92s luv 'em.

I did notice that my ppd numbers on FahMon went up after I updated the FahMon WU database. Maybe it just coincided with a change in WU mix.

Ragnar Dan wrote:

My GPU's are being dominated by the crappy 5903/5904 1888-point WU's, and they really waste time on my machines.

ARRGH! I got those very same ones right now

Gerbil Jedidiah wrote:

Ragnar Dan wrote:

My GPU's are being dominated by the crappy 5903/5904 1888-point WU's, and they really waste time on my machines.

ARRGH! I got those very same ones right now

They go quick on the G92's. You should be smokin' 'em with your powerhouse dual GPU 240 x 2 shaders beast!!

I get 6850 ppd (FahMon) on my GTX 260 with the 5903/5904 WUs.

I don't know why they're not fast on 285 or 295 cards, if they're fast on your 260. Meanwhile I'm losing about 5000 PPD with them on my 2 cards compared to what they produced before the change to crappy WU's, and about 3800 PPD compared to more recent WU's.

Ragnar Dan wrote:

I don't know why they're not fast on 285 or 295 cards, if they're fast on your 260. Meanwhile I'm losing about 5000 PPD with them on my 2 cards compared to what they produced before the change to crappy WU's, and about 3800 PPD compared to more recent WU's.

Maybe the new NVidia cards are just too darn advanced to crunch them effectively. Need a little yester-tech to do the job right!

jeffry55 wrote:

Ragnar Dan wrote:

I don't know why they're not fast on 285 or 295 cards, if they're fast on your 260. Meanwhile I'm losing about 5000 PPD with them on my 2 cards compared to what they produced before the change to crappy WU's, and about 3800 PPD compared to more recent WU's.

Maybe the new NVidia cards are just too darn advanced to crunch them effectively. Need a little yester-tech to do the job right!

A 5904 vs a 5774

Gerbil Jedidiah wrote:

jeffry55 wrote:

Ragnar Dan wrote:

I don't know why they're not fast on 285 or 295 cards, if they're fast on your 260. Meanwhile I'm losing about 5000 PPD with them on my 2 cards compared to what they produced before the change to crappy WU's, and about 3800 PPD compared to more recent WU's.

Maybe the new NVidia cards are just too darn advanced to crunch them effectively. Need a little yester-tech to do the job right!

A 5904 vs a 5774

Wow Jed! That is interesting to see. My GTX 260 was getting 6858 ppd while crunching a 5904 yesterday!


Maybe it is just a two GPU on one card thing. My 9800GX2's don't put out the same points per GPU as my 8800GTS's.

I'll have to try the screenshot thing again and send it for comparison.