What do they mean by "2-step deep" ?

In this project every workunit we process is a piece of a "longer" Molecular Dynamics simulation. So when a workunit is finished, uploaded and validated, then its result will be sent to another host to process the MD simulation from where the previous one has finished, until the whole given timeframe of the MD simulation is completed. For example: the whole timeframe for the MD simulation is 1µsec, it will be divided into 100 pieces of 10ns workunits (aka "steps"). However in the case of NOELIA_hfXA_long, there are only 2 such pieces.
On the server status page there are 22.000 unsent long workunits right now, in this case every one of them will be processed twice, while in other cases there are only 2.200 unsent long workunits, but those usually processed for 100 times.

... A lot of people seem to be struggling to finish them in 24hrs.

How about this for a down-to-the-wire full bonus WU:



And this after a 2hr:40min upload!

It does seem a bit daft that upload time should be added to processing time. Doesn't a done WU package include processing-end time??

---

Credit is applied when a task is reported, it's not based on runtime. While you obviously have a point, report time is the time it tasks to download, run, upload and report a task.
Tasks aren't of any research use until they are uploaded. Reporting is different; it's a separate tiny upload, mainly for credits, and goes to a different database.

Good to see you managed to get that task reported in time. I suggest people lower their cache (where possible) if you are close to missing out on a time bonus.

---

FAQ's

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

Now that we have these units uploading successfully, I noticed that the server status has 22,000 + work units listed as unsent. I have never seen that number that high, and since in the last few days, I have been getting nothing but these Noelia units, putting 2 and 2 together, I ask that question again, what's the big thing that we are crunching here?

I've been able to get nothing but the Noelia WUs either. They take too long to run to meet the 24hr deadline so have moved the NVidias to other projects. The Toni WUs are the only ones that work OK on my 4 cards as the Nathans take too much memory to run efficiently. Will check back from time to time as I would really like to be running GPUGrid...

What do they mean by "2-step deep" ?

In this project every workunit we process is a piece of a "longer" Molecular Dynamics simulation. So when a workunit is finished, uploaded and validated, then its result will be sent to another host to process the MD simulation from where the previous one has finished, until the whole given timeframe of the MD simulation is completed. For example: the whole timeframe for the MD simulation is 1µsec, it will be divided into 100 pieces of 10ns workunits (aka "steps"). However in the case of NOELIA_hfXA_long, there are only 2 such pieces.
On the server status page there are 22.000 unsent long workunits right now, in this case every one of them will be processed twice, while in other cases there are only 2.200 unsent long workunits, but those usually processed for 100 times.

Would this explain the size of these WU? (2 step deep means longer timeframe of simulation per WU, right?)

Now that we have these units uploading successfully, I noticed that the server status has 22,000 + work units listed as unsent. I have never seen that number that high, and since in the last few days, I have been getting nothing but these Noelia units, putting 2 and 2 together, I ask that question again, what's the big thing that we are crunching here?

I've been able to get nothing but the Noelia WUs either. They take too long to run to meet the 24hr deadline so have moved the NVidias to other projects. The Toni WUs are the only ones that work OK on my 4 cards as the Nathans take too much memory to run efficiently. Will check back from time to time as I would really like to be running GPUGrid...

Why don't you switch to the short task queue? AFAIK they still get a 24 hour bonus and are at least 1/2 the length or less of the long tasks.

---

XtremeSystems.org - #1 Team in GPUGrid

Why don't you switch to the short task queue?

Pretty please, tell me how to do that. Thanks!

---

Why don't you switch to the short task queue?

Pretty please, tell me how to do that. Thanks!

1) Go to your account area
2) Under "Preferences", select "GPUGRID Preferences"
3) Select "Edit GPUGRID preferences" for setting appropriate for your host (default, home, school, work)
4) Under "Run only the selected applications", check "ACEMD standard" and be sure the others are not checked.

Hope that helps!

MarkR

---

1) Go to your account area

Hi Mark! Thanks for answering.

In BOINC I clicked "Your Account". Came here:



Me no see "Preferences"!!

---

Tomba, you are viewing your account information that everyone can see.

To view your preferences however, click on your name on the top right of the GPUGRID web site. It should be next to server status, and is above the Volunteers and the Science links. This will take you to the correct account area where you can change your preferences.

Look at the top of the picture you posted, and click on tomba.

Tomba, you are viewing your account information that everyone can see.

To view your preferences however, click on your name on the top right of the GPUGRID web site. It should be next to server status, and is above the Volunteers and the Science links. This will take you to the correct account area where you can change your preferences.

Look at the top of the picture you posted, and click on tomba.

Got it! Thank you!!

---

Why don't you switch to the short task queue?

Pretty please, tell me how to do that. Thanks!

It's in your project preferences. Be advised that the credit earned is much lower though.

Just had another Noelia "Error while computing", after eight hours of crunching. That's the sixth out of a total of 15.

Can I please have more Nathan benHPs...?

---

Are you overclocking your card? Is your computer crashing? Things like that can cause these errors.

Just had another Noelia "Error while computing", after eight hours of crunching. That's the sixth out of a total of 15.

Can I please have more Nathan benHPs...?

Obviously you can't have more of them.
Your error is an "Energies have become nan", so see the appropriate thread.

Is there any way to stop the old 31 wus from hitting my machine other than checking and aborting manually? These are insanely slow on my old 560ti. The server reports 1000s of wus ready to send and when I abort a 31 wu it almost always send a 42 task so I'm thinking the server isn't out of 42 wus. I saw this rarely in the past with the long queue but it seems much worse with these tasks.

You can try my workaround on Linux:
1. turn off file size checking in cc_config.xml
2. overwrite the cuda31 binary with the cuda42 binary

We can remove already 3.1 apps on the long queue, but we will not upgrade the application until new year.

gdf

GDF says yes, server says no!

Your error is an "Energies have become nan", so see the appropriate thread.

Thank you for responding. I went there. All a bit confusing BUT I think I got the message that I need to under-clock my GPU. In just over two months I've had 16 "errors" out of 94 "good" WUs (yes - I log all my WUs...)

I have an ASUS GTX 460 1 gig. Stock settings are Core: 675, Shader: 1350, Mem: 1800.

For the 18 months I've had the 460, I've been GPUGRID-ing at: Core: 850, Shader: 1700, Mem 2000. It's only recently the errors have reared their ugly heads...

I have 'EVGA Precision', that lets me do changes, though I have nothing that lets me change the voltage...

Can I please ask you to recommend which of the three parameters I should change to give me the best chance of avoiding errors.

Many thanks, Tom

---

I have an ASUS GTX 460 1 gig. Stock settings are Core: 675, Shader: 1350, Mem: 1800.

ASUS have their own GPU overclocking tool, but I don't recommend it.

For the 18 months I've had the 460, I've been GPUGRID-ing at: Core: 850, Shader: 1700, Mem 2000. It's only recently the errors have reared their ugly heads...

As the GPU is getting older, it can tolerate less overclocking.
Also, the CUDA4.2 tasks tolerate less overclocking than the old CUDA3.1 tasks.

I have 'EVGA Precision', that lets me do changes, though I have nothing that lets me change the voltage...

I use and recommend MSI Afterburner for that purpose.

Can I please ask you to recommend which of the three parameters I should change to give me the best chance of avoiding errors.

1. Cleaning the cooling fins of the GPU and the CPU with high pressure air duster (using a vacuum-cleaner at the same time is recommended)
- (also the PSU, as much as you can do it through the rear grille, or the fan)
2. If there is any overclock applied to the PCIe bus, cancel it, and set the PCIe bus frequency to its default of 100MHz
3. GPU memory clock is irrelevant, raising it is just asking for trouble. (set it to factory default)
4. You can raise the GPU voltage by 0.025V increments as long as your GPU temp is below 80°C (and gets stable).
5. Lower your GPU core clock by 10MHz decrements, until it gets stable. (note that not every frequency can be set, so the actual GPU frequency can be ~3MHz lower than the value beside the slider)