I like that Stefan explains what the WUs are and how they interact. I wish the other researchers would do that, it takes 5 minutes of their time.

Here is my result for a BNBS Project

Took about 2 days on a gtx 960 with windows xp.

http://www.gpugrid.net/result.php?resultid=15854803

It would be much shorter if you would have a PCIe3.0 capable MB/CPU pair (at least a 3rd gen i5, or a 4th gen i3, Pentium or Celeron CPU)

I dont think so... I have an old PCIe2.0 Mainboard and still it doesn't slow down my gtx 1080 noticeably, maybe 2-3%. PCIe2x16 yields the same speed as PCIe3x8 which is OK for most SLI settings. See multiple reviews in the web which have examined the speed difference in detail.

So IMHO with 2.0x16 there is enough throughput for a middle class GPU like the gtx 960. But I think you do even use PCIe1.0, don't you?

---

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

It would be much shorter if you would have a PCIe3.0 capable MB/CPU pair (at least a 3rd gen i5, or a 4th gen i3, Pentium or Celeron CPU)

I dont think so... I have an old PCIe2.0 Mainboard and still it doesn't slow down my gtx 1080 noticeably, maybe 2-3%. PCIe2x16 yields the same speed as PCIe3x8 which is OK for most SLI settings. See multiple reviews in the web which have examined the speed difference in detail.

You use operating systems which have WDDM (Windows Vista and later), and this is a much bigger bottleneck than PCIe bandwidth.
The reviews usually about gaming or bitcoin mining which does not need that much interaction between the GPU and the CPU as GPUGrid does, so the WDDM does not hinder their performance as much as GPUGrid's.

On my computer (1060 3GB), BNBS works take 17 hours. In general for 410.000 credits, for some 490.000.

First, I am happy to have a recent Pascal card...
In second: before the new server, I had 100.000 credits more by day. We don't do that just for credits but I am sad nevertheless. And the error rate is 30%, it is huge. (always good for me I have to say)
A BNB work (with a huge 50% error rate) finished with error couple days ago. Between 14 and 17 hours for ... nothing; sorry: still, I will always have my electricity bill :-(

On my computer (1060 3GB), BNBS works take 17 hours. In general for 410.000 credits, for some 490.000.

You should set a short work queue (less than 5 hours = 0.2 days) to achieve 490.000 credits per workunit (this amount includes the bonus for returning the result within 24 hours).

First, I am happy to have a recent Pascal card...
In second: before the new server, I had 100.000 credits more by day. We don't do that just for credits but I am sad nevertheless.

The BNBS workunits are exatly as long as the BNB was, so your PPD will return to the previous level. It's probably lower than before because the project aborted one of your workunits mid-run.

And the error rate is 30%, it is huge. (always good for me I have to say)

The error rate starts at 100% every time, then it is continuously falling, as more and more successful tasks are returned. (now it's down to 29%.)

A BNB work (with a huge 50% error rate) finished with error couple days ago.

The BNB batch had an error (they are not chained correctly, it's explained here), and had to be aborted by the project which explains the huge error rate, and your lost workunit.

Between 14 and 17 hours for ... nothing; sorry: still, I will always have my electricity bill :-(

This is that workunit, it was aborted at 28,686.86 seconds (almost 8 hours).

You should set a short work queue (less than 5 hours = 0.2 days) to achieve 490.000 credits per workunit (this amount includes the bonus for returning the result within 24 hours).

Ok... Good to know, thanks for the advise ! I was still with a "long" queue because it was a time when the download of jobs was... You know... slooowwww
So now: biomolecular simulations for GPUGRID; credits and eternal glory for me !! (too much ??? Nooooope....)

Between 14 and 17 hours for ... nothing; sorry: still, I will always have my electricity bill :-(

This is that workunit, it was aborted at 28,686.86 seconds (almost 8 hours).

Thanks for the exactitude ! Still frustrating (and even if it is my first work in error => French guy) :D

You use operating systems which have WDDM (Windows Vista and later), and this is a much bigger bottleneck than PCIe bandwidth.
The reviews usually about gaming or bitcoin mining which does not need that much interaction between the GPU and the CPU as GPUGrid does, so the WDDM does not hinder their performance as much as GPUGrid's.

Well, even my PCIe2x16 conncected gtx1080 runs at pretty 95% per GPU-Z... which is a very pleasing figure. The BNBS accommodate the fast Pascal I presume and reduces the WDDM influence. Anyway, I am happy with the throughput.

---

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

Well, even my PCIe2x16 conncected gtx1080 runs at pretty 95% per GPU-Z... which is a very pleasing figure. The BNBS accommodate the fast Pascal I presume and reduces the WDDM influence. Anyway, I am happy with the throughput.

Exactly. There's unusually little interaction between the CPU and the GPU by these workunits.

As far as I could see, all BNBS WUs yield 492.000 credits if crunched within 24 hours.

this evening, for GLU73ALA_S13F6_C2-SDOERR_BNBS2-3-4-RND1150_1 which was crunched in about the same time as all the others before (i.e. approx. 18 hours on my GTX970) I got 328.000 points.
How come?

As far as I could see, all BNBS WUs yield 492.000 credits if crunched within 24 hours.

this evening, for GLU73ALA_S13F6_C2-SDOERR_BNBS2-3-4-RND1150_1 which was crunched in about the same time as all the others before (i.e. approx. 18 hours on my GTX970) I got 328.000 points.
How come?

Probably the same way I did with GLU73ALA_S2F7_C0-SDOERR_BNBS2-2-4-RND7864_1.

The first attempt was sent to a user with a slow card (by current standards), who took over 5 days and lots of restarts to complete the task.

When that task passed the initial deadline without response, a second copy was generated and sent to my host. I completed it within 24 hours, but in the meantime, the original task was returned and validated - earning its owner the basic credit score, without any fast completion bonus. That sets the 'tariff' for the workunit, and the second copy to be reported gets the same as the first. That's the way the bonus system has been set up, and it's the same for everyone - nothing personal about it.

Richard, thanks for the quick response. I would never have thought of something like that.
I had a similar case a few days ago already, where I could not explain the low credits either.
Still, a strange system though :-(

As far as I could see, all BNBS WUs yield 492.000 credits if crunched within 24 hours.

this evening, for GLU73ALA_S13F6_C2-SDOERR_BNBS2-3-4-RND1150_1 which was crunched in about the same time as all the others before (i.e. approx. 18 hours on my GTX970) I got 328.000 points.
How come?

Probably the same way I did with GLU73ALA_S2F7_C0-SDOERR_BNBS2-2-4-RND7864_1.

The first attempt was sent to a user with a slow card (by current standards), who took over 5 days and lots of restarts to complete the task.

When that task passed the initial deadline without response, a second copy was generated and sent to my host. I completed it within 24 hours, but in the meantime, the original task was returned and validated - earning its owner the basic credit score, without any fast completion bonus. That sets the 'tariff' for the workunit, and the second copy to be reported gets the same as the first. That's the way the bonus system has been set up, and it's the same for everyone - nothing personal about it.

This has been happening for years. See link below:

http://www.gpugrid.net/forum_thread.php?id=2795#22917

This has been happening for years. See link below:
http://www.gpugrid.net/forum_thread.php?id=2795#22917

hm, it's obviously never happened to me until few days ago. So I was even lucky :-)

Greetings;
First, I want to say that I really like supporting GPUGRID as a project. The team publishes results. Also, it seems so many other distributed (grid) computing projects are in theoretical math or in astronomy. I LIKE supporting a project which we can hope will improve human health down the road.
I have a pretty modest GPU (GT 640**) compared to many GPUGRID hosts. However, I have been averaging right at 11 WU a month. Also, I imagine that I am far from unique... that there are many BOINC volunteers 'out there' who have a more modest system (not costing a grand or more, and with no external power supply.)
I appreciate that timely turn-around is important, but there is no other BOINC project of which I am aware with such a short time-out limit on WUs (I am active in 15 projects, not counting several which have by now ended). Ten to 20 days is pretty typically the time limit. SETI has a time limit of over 5 weeks.
So, my First question: Is GPUGRID snubbing potential BOINC volunteers who have, say, a GeForce, but not a Titan or whatever? (This part of the post is perhaps better fitted to another forum.)

In the past couple of weeks, the lion's share of WUs have been BNBS -- and they have been running very, very long.
My second and MAIN question is: is the 5 day timeout enforced on tasks, even if they are making progress?
In the past week or so, I have abort a dozen or more BNBS WUs because, while running, they were heading to 300+ hours of compute time.
At this posting I have a WU running with anticipated completion time of jut over 11 days (270 hours). I really would like to run this one, since I have not been able to run any GPUGRID WUs in nearly two weeks. (In the past POEM was a good backup project, but they have recently terminated.)

My uncertainty about the timeout being enforced comes from the fact that the Performance page will list several batches with maximum run time (in the box-plots) of well over 120 hours.
Thanks.
LLP, PhD PE

** Some 6 years ago, I had a system with a 530. A couple of years later, I upgraded to a 640... I didn't go for more so as to keep a reasonable budget, and not to have to buy an external power source.
By the way, I am NOT a gamer at all. I bought the GPU solely to support distributed computing science projects.

---

I think ∴ I THINK I am
My thinking neither is the source of my being
NOR proves it to you
God Is Love, Jesus proves it! ∴ we are

Hi,

Well, for that kind of hardware GPUGRID offers short runs. But I admit that this queue sometimes is barely stuffed. We hope that Stefan and his colleagues can get more work for this queue by opening GPUGRID for lab tutorials, educational purpose or something like that, but this is just guesswork for now and we have to wait for their own considerations and decision.

But in case you consider an GPU upgrade for accessing long runs without buying a new power supply, there is good News. The new Pascal GPU generation is so power efficient that you can triple your crunching speed at the same power usage. Most of the mainstream gtx 1050ti can manage without extra 6pin connector and get their power from the mainboard only. And it is relatively favorable, Zotac, Inno3D, Gigabyte, KFA and Palit offer their Mini/single fan cards from 150€. Alternatively there is the slower gtx1050 for 120€ and up. Still much faster than your gtx640.

In my humble opinion the gtx 1050ti (and probably the speed equivalent but less efficient gtx960 or 660ti; careful: unlike the 1050ti they need extra power) are the most favorable Cards for a long run (here: just one per GPU), it does not make sense to use slower cards. Having said this, even the 1050ti will miss the deadline for receiving the early completion bonus. So don't expect too much.

---

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

Having said this, even the 1050ti will miss the deadline for receiving the early completion bonus. So don't expect too much.

You make this sound so much worse than it is. The 24 hour completion bonus for BNBS2 is 492k credits, the non-bonus is 410k. No matter what it is, it is hundreds of times faster than a gt640 and will dramatically contribute compared to it.

Having said this, even the 1050ti will miss the deadline for receiving the early completion bonus. So don't expect too much.

You make this sound so much worse than it is. The 24 hour completion bonus for BNBS2 is 492k credits, the non-bonus is 410k. No matter what it is, it is hundreds of times faster than a gt640 and will dramatically contribute compared to it.

Yes, it will. But I didn't want to create the impression on the other hand that the 1050 goes off like a missile. Seemed worth mentioning to me. And no... it is not hundred times faster than the 640 (in fact it should be about 3x as I wrote) ....but I get your point ;-)

---

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

Well it takes him 11 days to complete a BNBS2, now it should take slightly over 24 hours, so over 10 times as fast. I like to exaggerate as well :)

Well it takes him 11 days to complete a BNBS2, now it should take slightly over 24 hours, so over 10 times as fast. I like to exaggerate as well :)

Well, to avoiding extending the exaggerations ... my 1050Ti turns around the big BNBS2 WUs in something of the order of 29 to 30 hours (plus or minus). The current one is on course for the lower value; I've had ones that take longer.

So, yes, I "only" get 410k for such WUs and I miss the 20% bonus for sub-24 hour completion (which I get from the 18 hours my 1060 takes for equivalent WUs) but it's way better than my 750Ti (which has been relegated to the slower box and no longer does GPUGrid except for the occasional short run WUs) as that was taking closer to 40 hours.

It's not a matter of "snubbing" users. It's a matter of doing practical research in practical timeframes. We understand that not everyone can have a 1080 but the most interesting research is done on slow biological processes which need really long simulations.
When we test methods instead we sometimes run short simulations on the short queue like we do these days with Adria and will do more the coming month.