Hi, I know you guys are running on the fresh new cards, but those of us with a bit older iron are now out of the game!

The long runs that you have introduced lately run 7 days on GT640, but have about 5 days deadline. Later submit is ignored as the task has already been sent to somebody else meanwhile. I know these are not hyper super trooper heavy gamer cards, but they can still do a lot of useful work.

Can't you extend the deadline, or break the looooong tasks to a somewhat shorter ones? I mean even faster cards must have trouble, if they do not work 24/7, which is kind of crazy to expect at a volunteer computing network, isn't it?

Can't you extend the deadline, or break the looooong tasks to a somewhat shorter ones? I mean even faster cards must have trouble, if they do not work 24/7, which is kind of crazy to expect at a volunteer computing network, isn't it?

Yes, you have a point.

I would say, though, that a lot of people here do indeed run their computers 24/7. If you do not run your computer 24/7, I'm with you on that one. I think it would be hard to crunch unless you have say for example a gtx 1080.

Just a helpful suggestion, but You can try messaging one of the scientists if they do not answer this. I do sometimes! Let's keep our fingers crossed and hope that there will be many short tasks soon for people who have older cards. :)

---

Cruncher/Learner in progress.

Hi, I know you guys are running on the fresh new cards, but those of us with a bit older iron are now out of the game!

We're not running slow cards out of the game, you just have to go into account settings, GPUGrid preferences, and uncheck long runs. This is far easier than recompiling all the tasks and fundamentally changing the entire system. And you still contribute in the short runs because all the fast cards are tied up doing long runs.

Alright, but just about a week ago, there was no problem with the long tasks. I do not think you would find another project where you get 250k+ points for a single task...

But thanks for the hint. Ok, if I have to, I'll switch.

Who knows how many people who do not watch their stats regularly do not even know that they burn their electrons for nothing. ;)

I think it would be hard to crunch unless you have say for example a gtx 1080.

Just a side note on this, if I may. As I have got one 1070(OC) and one 1080 ... and from my experience there is almost no difference in crunching Performance. As for the Pascals, the GFLOPs specs are only halfway reflected by the PPD, whereas there is a huge price difference between the 1070 and 1080. So the 1070 has the better price/performance ratio by far.

Load utilization looks even better for the smaller cards e.g. the 1060. No need to go for the 1080 for the long runs. Unless you want to play Mass Effect Andromeda next year ;-)

So in case you seek for a favorable upgrade in the near future, my recommendation would be the new GTX 1060, 1050ti or the little sister 1050 for 100$. Even the smallest 1050 has about the same power draw as your GT640 but almost 3x (THREE) times more computing power. Time flies.

Alternatively, there are many second hand GTX 660/660ti/670/750ti/760 now on market for maybe 60 USD... still capable cards.

---

I would love to see HCF1 protein folding and interaction simulations to help my little boy... someday.

I do not think you would find another project where you get 250k+ points for a single task.

The credit among different projects is not directly comparable, with SETI, you're doing very well if you can manage 10k RAC. They do not represent GFLOPS, the project owners can assign any amount of credit they want. The way cryptocurrencies like Gridcoin even this out is averaging everone's credit for that one project and placing you where you are so they can reward you with Gridcoin.

Thanks Joerg for the useful hints! I think I will choose AMD next time as I am disappointed by NVIDIA's lack of OpenCL 2.x standard support.

I know, Pappa, the credits are a mess, but the same is true for the amount of computation: you do not find such long tasks in other projects. Now I got some long short tasks, which are like the previous long tasks, so it's fine with the settings updated.

I try to stay away from all *coin virtual currencies, it's all garbage. I think the state has it's important responsibilities and should maintain its official currency, other currencies should be made illegal. And especially those that require computer crunching are very bad - for the environment for instance.

Back to the topic, it would be great, if the master were able to recognize that the particular slave machine could not meet the deadline and restrain from sending such tasks regardless the individual's project settings.

happy holidays.

You are welcome, however please keep in mind that GPUGRID does not support AMD cards (yet)! So you would have to switch to another project e.g. Folding@Home. The RX480 8G would be an attractive card, although not as efficient as the new Nvidia Pascal.

---

I would love to see HCF1 protein folding and interaction simulations to help my little boy... someday.

This thread has deviated too far from its topic, so I've answered here.

Back on topic, I would like to see a finer adjustment setting for running either long or short runs per host or even per card in a host from the preferences page.

If you have a wide range of cards available it is difficult to adjust the workload accordingly. I am aware of the cc_config file but it is awkward to setup and adjust later if you want to change it.

For example I have a GTX 670 in one host and two GTX 980's in another two hosts. I would like to always have the short runs go to the 670 and the long runs to the 980's.

If Gpugrid runs out of short runs for the GTX 670 it could switch to another project. If Gpugrid runs out of long runs for the 980's they could switch to short runs but then go back to long when they are available again.

I also use Boinc Stats and can adjust the projects running per host but not the CPU or GPU applications. Having the settings available to adjust there would be a good option.

Another thought would be to define which cards can run which type of jobs. Ideally GPUgrid could determine which GPU card is available to send work to and distribute the jobs accordingly.

Even if this is too much overhead a well documented guideline would help most users decide what their HW capability is and make better decisions on what makes good sense to run or when they just need to upgrade to something better.

JJCH

Tasks are not always the same size, and sometimes they end up in the wrong queue by mistake.

Use two profiles:

Home
670 - select short tasks and other work if short tasks are not available (If no work for selected applications is available, accept work from other applications?) - if you want to crunch long tasks when short tasks are not available, otherwise don't select this option.

Work
980's - select long tasks and other work (short tasks) if Long tasks are not available.

Also, have GPUGrid Resource share set high (10000) and have one or more GPU backup projects with low resource shares; (10), (1), (0).

Note that you can also exclude apps for different projects against a specific GPU in cc_config.xml or disable a GPU:

<exclude_gpu>
<url>http://albert.phys.uwm.edu/</url>
<device_num>0</device_num>
<app>einsteinbinary_BRP4</app>
</exclude_gpu>

<exclude_gpu>
<url>http://www.gpugrid.net/</url>
<device_num>1</device_num>
</exclude_gpu>

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Right, back to the topic.

I would say, though, that a lot of people here do indeed run their computers 24/7. If you do not run your computer 24/7, I'm with you on that one.

So looking at the current ratio between the short and long tasks available (33/657) and to be clear: GPUGrid is targeted primarily at people who buy a recent nifty expensive hardware and dedicate 100% of its utility just to participate in this single project and perhaps some other BOINC projects with a priority that is lower by several degrees of magnitude. Right or wrong?

Wrong, in my personal opinion.

Don't look too much at the 'cavalry' surging ahead with their shiny 980ti or 1070ies. They look good at the RAC high scores, but in fact, their contribution is nothing without the countless 'common' supporters, all the people with proven gtx 460ies, 560ies or 660ies, who form the backbone or 'infantry' of every BOINC project. Same applies here.

My deep respect belongs to every woman and man who contributes to this worthy cause according to her/his possibilities, no matter what the RAC and the position in the score table.

---

I would love to see HCF1 protein folding and interaction simulations to help my little boy... someday.

Requesting a constant supply of short tasks to greater facilitate the masses; when you're enhancing the project management of your 'free' DC-supper computer. We can't expect people to fork out £300 for mid-range GPU's, never mind £650 high end GPU's, only to encounter work shortages. Note my position on the price of the Pascals (it's too hight ATM) and position regarding system builds for here/crunching in general (wait until the competition is better).

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

@skgiven At one point it did try using the Home/Work profiles but these three machines are all at home. I also have a few at work in my office area and a bunch in a computer lab. I ended up setting the profiles to where they were located which was somewhat related to keeping track of them before I started using Boinc Stats.

With this plan I would have to change all of the systems with low end cards in the GTX 670 range and whatever Low End Quadros to the Home setting and Mid Range cards GTX 980 and better Quadros to Work. There is also a School profile I could use for High End GTX 10x0 and equivalent Quadro cards if I had any and wanted to break it down a little further.

The GPUgrid Resource share is already set to 99,999 which gives GPUgrid the maximum priority. Normally I have all the systems with Nvidia cards set to run GPUgrid only using Boinc Stats and just turn on the other projects manually when GPUgrid runs out of work. The way I'm using it this setting doesn't make much difference unless there are multiple GPU projects queued up.

At one time I did try setting up fancy cc_config files for all of the different systems and GPU's but it ended up being a bit of a maintenance nightmare. Invariably there would be an app change or some other type of conflict and it would require updating the cc_config files everywhere. If there was a more automated way of doing this it could be very useful.

It may just be more simple and make more sense to update the GTX 670 to a 1060/70 so it would at least have a chance to complete a long run within 24hrs if I left it running. This would also mean looking at upgrading the low end Quadro cards too. Funny even the GTX cards can outrun a Quadro so I may even look at converting those over.

From what I am seeing on the Performance Stats the Long runs are taking around 16hrs on a GTX 980 and the bulk of the rest of the cards I have are running over 24hrs. Most of them are completing jobs in anywhere from about 36 to 48hrs. Way too long to get a top credit bonus. I had previously set all the cards to run two work units but I'm thinking I need to back that down to one and see if they can complete in less than 24hrs.

Don't look too much at the 'cavalry' surging ahead with their shiny 980ti or 1070ies.

well, I look, I look, but what I see is that the only available tasks at GPUGrid are such that the NVIDIA in my notebook that is only 1 year old require about 5.03 days to complete, but there is deadline of 5 days set on them. and of course, the notebook cannot compute 24/7.

Could that be that the people who are very hungry after their gridcoins have hijacked this project? :-O

Your laptop is probably a 940m or lower. Low end desktop GPUs are something like 660ti, 760s etc, more than 10 times faster than your laptop gpu.

From the computer list it seems to be a 920M with either the old GF117 Fermi chip or the Kepler GF208. If the Laptop is only one year of age I would presume the latter. This one yields 440 GFLOPS SP which is about the same speed as a Desktop GT240, GT545 or GT640 but significantly less Performance than a GT730. No matter how you look at it ... this mobile GPU is not particularly suited for crunching. For comparison, a 100 USD entry level desktop GTX 1050 yields 1700 GFLOPS SP.

---

I would love to see HCF1 protein folding and interaction simulations to help my little boy... someday.

Your laptop is probably a 940m or lower. Low end desktop GPUs are something like 660ti, 760s etc, more than 10 times faster than your laptop gpu.

It's even worse: a 920M, which is based on GK208 (Kepler).

Pavel wrote:

...the NVIDIA in my notebook that is only 1 year old require about 5.03 days to complete...

The 920M is released on March 13, 2015, but as it is a rebranded Kepler chip (GT 730M), which is released Mar 6, 2014 in reality it is almost 3 years old. But it's not the age of the card: even a recent low-end ("entry level") GPU / mobile GPU would be inappropriate for long runs.

this mobile GPU is not particularly suited for crunching. For comparison, a 100 USD entry level desktop GTX 1050 yields 1700 GFLOPS SP.

You are missing the point. I am not buying hardware for crunching somebody else's data, I have enough of mine, but I'd like to use my HW for useful purpose when I don't use it for my other things.

I am happy with this GPU. It can do much better than CPU.
But there is no useful project at BOINC for this.

I don't care so much about E.T. and neither about crypto cows, I do not see much else around at BOINC after Poem has discontinued. Any hints?

But I see 12979 long tasks, and 0 short tasks at GPUGrid. That means I have been pissed off by the crazy gridcoiners here and ought to leave. It was not like that a month ago. The long tasks were not extreme long.

That is the fact.