I do not speak English, sorry for language errors.

I have an Asus Nvidia GTX 650 TI Boost graphics card, the card failed after only five months to process for GPUGrid and being useless. As surely still in warranty will achieve that the seller replace me with a new one just like it or another model of similar abilities and little price difference.
If I get a new NVIDIA graphics i want to continue processing GPUGRID, but would do so without forcing the graphics card as much as I did with the previous graphics card.

Can anyone tell me how I can get GPUGRID units are processed using only 60% or 65% of the total capacity of the graphics card? will achieve so perhaps reduce the risk of damage to the new NVIDIA graphics card.


Thank you.

I use EVGA Precision software http://www.evga.com/precision/ to control both EVGA and Gigabyte cards. Supposed to work with all Nvidia cards. In one case which does not have good airflow, I run one card at the 90% power setting. This will auto undervolt (and underclock as a result) the card a little bit. This is one simple way to reduce the stress on the card.

The big thing to reduce the stress on the card without slowing it down is to keep it cool. Increase the fan speed on the card (can do a custom fan profile with the EVGA software). Note other people use MSI Afterburner which I have not used, but I imagine does all the same.

Also, good airflow through the case. Or you can go liquid cooling.

I do not like letting my cards run >72. Not necessarily a magic number, but the cards will last longer and have fewer errors.

Hello

I use EVGA Precision software http://www.evga.com/precision/ to control both EVGA and Gigabyte cards. Supposed to work with all Nvidia cards. In one case which does not have good airflow, I run one card at the 90% power setting. This will auto undervolt (and underclock as a result) the card a little bit. This is one simple way to reduce the stress on the card.

The big thing to reduce the stress on the card without slowing it down is to keep it cool. Increase the fan speed on the card (can do a custom fan profile with the EVGA software). Note other people use MSI Afterburner which I have not used, but I imagine does all the same.

Also, good airflow through the case. Or you can go liquid cooling.

I do not like letting my cards run >72. Not necessarily a magic number, but the cards will last longer and have fewer errors.

Hello

I use the GPU Tweak software, which also allows you to control the fan speed, but mostly I've kept in "auto". I do not know which is the optimal operating temperature of a graphics card but I guess that values ​​below 65 or 66 degrees Celsius would be the safest.
Could you explain to me how I can do this?: "I run one card at the 90% power setting. This will auto undervolt (and underclock as a result) the card a little bit. This is one simple way to reduce the stress on the card."
Before I had a very bad air flow, I have now changed the case and now I have good air flow but also interested me to protect my new graphics card with your simple way to reduce the stress on the card.


Tanks for your answer.

In the EVGA Precision software it is as easy as sliding a bar. Older cards have the "linked" unchecked and unable to check such as the 460 (could not set temp targets). So I just slide the Power Target to 90% in above situation. I have since moved it to 105% as I fixed the airflow problem I had with my case situation.

The default Fan Curve is for a real quiet (meaning hot) card. Click the Fan Curve to change it up a bit.

On my newer 780Ti cards, I unlink the power and temp target and leave Power at 105%, Temp at 72, and Prioritize Temp. That way, it will run full boost for SANTI WU's, and will throttle back on a combination of warm days and NOELIA WU's to keep it from going past 72 (even when fans have moved to 100%) if needed.

That's some very good advice from Jeremy! I'm running my GTX660Ti at a reduced power target (108 W instead of 130 W - but the software will only show percentages). In additiioin to this I'm applying a 50 MHz GPU overclock, as I know the card can still take it. This way I'm saving power and running more energy-efficient due to the lower voltage automatically applied. In addition to this I have the memory OC'ed slightly, as this boosts performance somewhat while costing pretty much nothing.

Generally It's very probably just bad luck that your preevious card failed so early. There's always a "U" curve when plotting failure rate over runtime: some early failures not caught by the manufacturer, and increased failure rate nearing the end of life.

In your case you very probably got one of those early failing chips (I also had one once) and it would have failed relatively early without GPU-Grid too. What we have been talking about can not protect you from such a chip, but if you have a normal one it will push the "end of life" time point further away. And make you run more energy-efficient ;)

BTW: you could ask to get a GTX750 or GTX750Ti instead as replacement. These use the new Maxwell architecture and are significantly more power efficient than Keplers (like your previous card).

MrS

---

Scanning for our furry friends since Jan 2002

Hello

I still do not know which will be the graphics card with the seller to replace my GTX 650 TI Boost, will give me a list of graphics cards available and I will have to choose one. When I get my new graphics card i return to this forum to write some more doubts. :)

I have some more questions:

Does anyone know what is the average time that a PC can live processing data from Boinc? because my graphics card worked steadily to 90% of its capacity and the processor at 80% of its capacity, and it also puts stress on the motherboard, the memory and other components.

Does anyone know if in the future GPUGrid also worked with graphics cards AMD Radeon? If this happens it would be interesting.



Thanks for the help.

My home server has been crunching with BOINC for several years uninterrupted. Never had any problem whatsoever, even in the hot Greek summers.

Pretty soon I will be crunching for GPUGrid for a year on my GTX 650 Ti full time, 24/7. No problem with that too.

If your computer is well-built, you will not have a problem with crunching.
Well-built means in general:

High-quality components (motherboard, memory, CPU heatsink, GPU)
High-quality, high-efficiency power supply.
Very good airflow

---

I have some more questions:

Does anyone know what is the average time that a PC can live processing data from Boinc? because my graphics card worked steadily to 90% of its capacity and the processor at 80% of its capacity, and it also puts stress on the motherboard, the memory and other components.

Does anyone know if in the future GPUGrid also worked with graphics cards AMD Radeon? If this happens it would be interesting.

Thanks for the help.

As Vagelis says "High-quality, high-efficiency power supply" and "Very good airflow" are two very important keys. "High quality components" though sort of depends on how often you want to get your hands dirty working to keep the pc running. I personally fix pc's for friends and get their old parts instead of money in return, those are alot of the harddrives etc I use for my machines, they tend to crash more often then new ones, I also buy refurbished harddrives and they too tend to crash more often then brand new ones. But I do monthly backups of my Boinc only machines so if they crash I am only down for a couple of hours and then right back up again with the only thing being lost are the units.

I have been crunching 24/7 since 1999 and although those original machines are no longer crunching, I do have machines that are at least 5 years old and still going strong. Most of my pc's, I have 15 here at home, have gpu's in them and all of them crunch 24/7, most of the time even when I am on vacation. While Boinc is stressful it is not doing anything to the chip that it wasn't designed to do in the first place. Overclocking and all that kind of stuff can damage a chip, but as you are reading about those that do that stuff here at Boinc you will also see that they are not willy nilly overclocking their chips, they are doing it with care and thought. They are not trying to just get max performance, they are trying to get better performance intelligently.

Does anyone know if in the future GPUGrid also worked with graphics cards AMD Radeon? If this happens it would be interesting.

No. They tried to in the past, but results were bad. Things have probably improved, but the GPU-Grid app is comparably complex, so porting it to OpenCL (as would be neccessary for AMD GPUs) and afterwards maintaining 2 separate code paths (OpenCL and CUDA) would require significantly more man-power. Which they'd rather use for science, as long as nVidia GPUs are enough for them. And dropping CUDA entirely in favor of OpenCL is surely not an option either, as this would make the nVidias run slower and less efficient.

MrS

---

Scanning for our furry friends since Jan 2002

Hello

The answers you have given me I have been very useful. At the moment I have only one pc to BOINC processing (currently do not have any, because the only one I have still not even your graphics card). I've never done oveclock, but once I thought, I prefer to use the components as manufactured.
When my graphics card stopped working and send to the seller by it's warranty, people who worked with it had asked me if overclocked, which I never did or will, I think like you.

I have another question: does anyone know if this GPUGrid planned to complete its work in the future and become a retired project? because there are not many projects that use graphics cards ...


Thank you all for responding.

Nah, as long as we can simulate we will :) There are no plans for closing down GPUGRID

There are actually several projects that use GPUs. Off the top of my head, I am connected to the following projects that issue GPU tasks:
- GPUGrid (almost always has GPU tasks available)
- Poem@Home (rarely has)
- World Community Grid (rarely has)
- Einstein@Home (always has)
- Albert@Home (always has)
- Milkyway@Home (always has)
- SETI@Home (always has)
- SETI Beta (always has)
... and I'm sure there are more that I'm not connected to.

There is no harm in connecting to several projects (I'm connected to 30), and configuring Resource Shares and GPU Exclusions so that you keep all your resources busy, while prioritizing exactly which projects you want them to run on. The flexibility of BOINC is awesome.

There are actually several projects that use GPUs. Off the top of my head, I am connected to the following projects that issue GPU tasks:
- GPUGrid (almost always has GPU tasks available)
- Poem@Home (rarely has)
- World Community Grid (rarely has)
- Einstein@Home (always has)
- Albert@Home (always has)
- Milkyway@Home (always has)
- SETI@Home (always has)
- SETI Beta (always has)
... and I'm sure there are more that I'm not connected to.

There is no harm in connecting to several projects (I'm connected to 30), and configuring Resource Shares and GPU Exclusions so that you keep all your resources busy, while prioritizing exactly which projects you want them to run on. The flexibility of BOINC is awesome.

DistRTgen is one that you missed, so are Moo, Asteroids, PrimeGrid and Collatz. Although not ALL projects can use AMD cards. For credits DistRTgen absolutely hands down pays the most, but it favors the high end cards the most too. Each project is different and each uses your gpu slightly differently, some get close to 90+% usage on just one unit, some you can run multiple units at once on with no problems.

On my newer 780Ti cards, I unlink the power and temp target and leave Power at 105%, Temp at 72, and Prioritize Temp. That way, it will run full boost for SANTI WU's, and will throttle back on a combination of warm days and NOELIA WU's to keep it from going past 72 (even when fans have moved to 100%) if needed.

Jeremy,

Thanks for this tip! I've only ever used Precision X for monitoring the cards so the settings were at default, but I tried your tweaks and instantly saw some improvement from one card which was always powering down to base clock speeds. It is now properly applying boost as it should.

---

DistRTgen is one that you missed, so are Moo, ... PrimeGrid and Collatz.

And *some* will question the scientific value of these projects ;)

MrS

---

Scanning for our furry friends since Jan 2002

DistRTgen is one that you missed, so are Moo, ... PrimeGrid and Collatz.

And *some* will question the scientific value of these projects ;)

MrS

Those projects might be backed by power companies ;-) Some of them don't provide any value, just waste power...

DistRTgen is one that you missed, so are Moo, ... PrimeGrid and Collatz.

And *some* will question the scientific value of these projects ;)

MrS

Those projects might be backed by power companies ;-) Some of them don't provide any value, just waste power...

I wasn't trying to do an evaluation, just provide the names of some other gpu projects.

I agree with Mikey. Let's let the users decide what they want to attach to. I had just provided a list of ones I was quite familiar with, because I run them. Mikey added others that I was less familiar with, because I don't run them.

Nobody asked for an evaluation. I may agree with some of the sentiments that were stated, but I don't think they should have been stated.

Long story short: There are plenty of BOINC GPU projects out there to keep a GPU busy. :)

Hello

They have already given me a list of graphics cards that I get to replace the Nvidia GTX 650 TI Boost I had and unfortunately in this list are not included the graphics card I expected to find , they offer me are these:

* ASUS GeForce GTX660 TI DirectCU II OC 2GB DDR5 PCI -E

* EVGA GeForce GTX660 2GB DDR5 ACX Dual PCI -E
 
* GIGABYTE GeForce GTX660 TI 2GB PCI -E

I leave some questions:

The motherboard I have is an Asus but i can not remember the model, do you think the EVGA card can work well for Boinc on an Asus motherboard ? ( GPUGrid and other projects ) ? I do not know this graphics card and i can not find it on the site EVGA Latinoamerica : http://latam.evga.com/products/prodlist.asp?family=GeForce % 20Series % 20Family % 20600 , or the EVGA site of USA : http://www.evga.com/Products/ProductList.aspx?type=0&family=GeForce+600+Series+Family&chipset=GTX+660

Another question: any of these three graphics cards can run on a pc with an AMD FX- 8350 processor and with a power supplie of 850 Watts 80 + Gold? I worry consumption can be plotted Asus first listed , I'm not sure that the power supplie is able to supply it with the processor working at 100 % capacity .


Thank you all.

do you think the EVGA card can work well for Boinc on an Asus motherboard ?

The brands of the motherboard and graphics card don't make a difference as long as they're compatible. I use EVGA cards and an ASUS motherboard in my rig.

Another question: any of these three graphics cards can run on a pc with an AMD FX- 8350 processor and with a power supplie of 850 Watts 80 + Gold?

850W should be enough depending on what else you're powering. I previously ran two GTX680s and i7-3770k on a 750W power supply with no problems.

Here's a link to a pretty good calculator for determining PSU needs:
http://extreme.outervision.com/psucalculatorlite.jsp

---