Sorry, slight exaggeration on my part. Snowmain wants to get more work for his ATI card and someone suggested disabling the nVidia MB app. The exact quote was ...
[quote]Your hardware isn´t to shabby to be honest.
I would simply disable nvidia GPU in your preferences and only fetch ATI for a while.

Which part of disabling Nvidia work fetch didn't you understand? Snowmain has 5 days of Nvidia work, he can survive at least 3 days without asking for any Nvidia work, at which point he can re-enable Nvidia GPU work fetch.

Claggy

... pulling his nVidia card (good one)

Who said that???
I'm sure no-one (except the confused owner - people say something different which he understand as 'throw the nVidia in the trash')

Sorry, slight exaggeration on my part. Snowmain wants to get more work for his ATI card and someone suggested disabling the nVidia MB app. The exact quote was ...

Your hardware isn´t to shabby to be honest.
I would simply disable nvidia GPU in your preferences and only fetch ATI for a while.

... run a background utility to keep retrying the servers for more work

NOT "for more work", but to (more often) retry the downloads you already have.

Which doesn't address the actual problem that few if any ATI work units were being fetched and those that were had extremely large estimated completion times compared to the actual time it took for the ATI app to finish them.

If Snowmain simply sticks <flops>80e9</flops> into the MB ATI portion of his app_info ...

Simply for you.
Do you really think he will do that?
And even if <flops>80e9</flops> will be used - it may make BOINC ask for more work but will not fix the long backoffs (which in turn stops BOINC to ask for more work)
(he complained that "The 10 ati unit have been trying to download all night")

Well I get those long backoffs as well but they wouldn't be as annoying if we had a decent size cache of ATI units that, in Snowmain's case, would last for more than a few hours and for me a day.

And there is an entire thread dedicated for novices about how to determine the optimal number of work units each nVidia GPU can handle, which requires, once the number is determined, editing the app_info file. If somebody can explain those changes somebody can explain a simpler change like adding the <flops> line.

The 80e9 value is a SWAG on my part

Probably is good enough (my AMD Radeon HD 6570 shows ~50e9 for MB ATI tasks)

Of course editing of the app_info file requires a full shutdown of the BOINC manager

If we have to be literally correct - BOINC manager do not know about the existence of app_info.xml

Fine, since which part of BOINC uses which xml file isn't publicly documented, at least somewhere obvious, I took the precaution of not only shutting down my apps but BOINC manager as well so I didn't mess anything up. Sorry for erring on the side of caution.

I just saw Snowmain's problem as similar to mine, we weren't getting enough work downloaded for our GPU MB cruncher. From my limited understanding, if you ask for work, the server in theory sends enough work based on the estimates on how fast each work unit will be crunched by that app (an estimate) and the amount of "time" requested. Ask for 24 hours worth of units and if the current estimated completion time is 2 hours per work unit, you will end up getting 12 units (per core). The problem is when you are crunching them in 2 hours but the server constantly estimates you will take 12 hours. And if this estimate isn't changing as you complete more and more validated work then there is a problem.

To me the solutions that should have been suggested were ways to inform the server that the app is faster than what the server thinks it is. From what I understand <flops> does exactly this. Yes, it's deprecated, not the preferred method anymore and seemingly ignored if the server actually has it's own current flops value for your app but it's apparent that the server isn't tracking ATI MB apps, as least on newer installs.

People have been trying to treat the symptoms and not the underlying cause. Jammed up downloads aren't as much of a problem if you have 6 days worth of cache because the jam ups tend to go away in a few hours, unless it's a server issue of course. But if you are continuously running dry, jam ups are annoying "look I have ATI work units for you, but you have them, TTHHPPTT!!"

I know because I'm on dial-up and can't tie up my phone line for hours (not that ATT would let me) to go through the usual retry cycle. I've resorted to disabling and enabling my network activity setting to restart jammed downloads, which seems to work better than hitting the retry button every ten to fifteen minutes.

Sorry if I come off as angry, just frustrated that in this and my thread the general response was do nothing it'll get better or that it's a run of bad work units, give it more time.

PS: Thanks for the info on SIV. I'm not using the auto retry feature but since my CPU doesn't have a unified memory cache for all cores, I've use the CPU affinity setting to lock each CPU app to it's own core and dedicated memory cache. Might make them a little faster.

---

"Life is just nature's way of keeping meat fresh." - The Doctor

Well the nVidia technocrati wins again as an ATI cruncher is convinced it's not worth trying anymore.

Well played.

Nobody tried to convince him, he convinced himself!
(despite all advices - most of which he didn't follow)

... pulling his nVidia card (good one)

Who said that???
I'm sure no-one (except the confused owner - people say something different which he understand as 'throw the nVidia in the trash')

... run a background utility to keep retrying the servers for more work

NOT "for more work", but to (more often) retry the downloads you already have.

If Snowmain simply sticks <flops>80e9</flops> into the MB ATI portion of his app_info ...

Simply for you.
Do you really think he will do that?
And even if <flops>80e9</flops> will be used - it may make BOINC ask for more work but will not fix the long backoffs (which in turn stops BOINC to ask for more work)
(he complained that "The 10 ati unit have been trying to download all night")

The 80e9 value is a SWAG on my part

Probably is good enough (my AMD Radeon HD 6570 shows ~50e9 for MB ATI tasks)

Of course editing of the app_info file requires a full shutdown of the BOINC manager

If we have to be literally correct - BOINC manager do not know about the existence of app_info.xml

---

 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 

If Snowmain simply sticks <flops>80e9</flops> into the MB ATI portion of his app_info (I put it right before the <coproc> line in both ATI sections) he should get a bucket load of MB ATI work units. The 80e9 value is a SWAG on my part, if someone is already running an HD 5770 and actually has an average processing rate for it please share.

69.76e09 is what i got from my HD5770 when it was running MB.

Claggy

Of course editing of the app_info file requires a full shutdown of the BOINC manager yada yada, YMMV, yada yada.

Boinc Manager is just the GUI, i often run with Boinc Manager shut down, saves a few cycles, after editing the app_info, a restart of the Boinc client is required.

One of the simpler solutions is just to run Boinc 6.10.x, it doesn't have the excessively backed of downloads of 6.12.x and 7.0.x, and will ask for work once the project backoff has ended, unlike 6.12.x and 7.0.x,
But it's not easy to revert to Boinc 6.10.x once 7.0.x has been installed.

Claggy

Well the nVidia technocrati wins again as an ATI cruncher is convinced it's not worth trying anymore.

Well played.

There seems to be a simple solution that doesn't required a beta of BOINC, switching to the AP ATI app, pulling his nVidia card (good one), run a background utility to keep retrying the servers for more work, I surprised nobody suggested reinstalling Windows with this shotgun approach.

If Snowmain simply sticks <flops>80e9</flops> into the MB ATI portion of his app_info (I put it right before the <coproc> line in both ATI sections) he should get a bucket load of MB ATI work units. The 80e9 value is a SWAG on my part, if someone is already running an HD 5770 and actually has an average processing rate for it please share.

Of course editing of the app_info file requires a full shutdown of the BOINC manager yada yada, YMMV, yada yada.

What harm is it to try?

---

"Life is just nature's way of keeping meat fresh." - The Doctor

I also AP WU enabled...I one every other day or so. Like the regular work units it doesn't appear boinc understands that the video card can eat 1,800,000 Gflops in 2 hours...so it waits 2 days.

My system currently has 18(ATI) seti enhanced units(which is more than I have ever seen @ one time) downloaded now and crunching threw them. But it went all night with no ATI WU's. It will complete these units in about 8 hours...then it will again sit idle.

Then tomorrow I will pull the ati card and post it on ebay and replace it with another 8800gts. Overclock to 702mhz(from 500) with the memory over clocked to 1107Mhz(from 800).

---

A couple of things - your "valid" list for the GPU only shows about 4 or 5 valid, non-blanked tasks, a long way from your claimed 50 (which is more like the figure for your CPU). Second, "officially" we aren't meant to use <FLOPS>, but it works a dream, far better than the guestimates that BOINC generates, but they are a manual process and BOINC/S@H is meant to be a "fit and forget" process...

It's been crunching for a week, the validated list only has work units validated within the last 24 hours or so.

At this moment I have 49 validated tasks, 8 of them ATI and 5 of those that didn't overflow. Yet I still have 0 tasks completed according to the application details page. And this is just one 24 hour period, I started crunching GPU work units about 7 days ago. Ignoring the overflows, you can't say that the rest are all outliers.

Something's stinky in Denmark and it's not the cheese. <flops> is a workaround that works for me and I can go back to connecting to the servers only once or twice a day rather than dialing in every few hours or leeching off a neighbor's open wifi connection continuously (I'm trying to be good about not using it, their cap is only 50GB a month and they have kids with game consoles).

Snowmain could do the same thing since his primary complaint is his ATI card isn't getting enough work and is running dry. It's just a matter of coming up with a good SWAG at a flops value that gives estimates that are close to what he is actually seeing as completion times.

I'm not suggesting that it's a long term solution, just one that's useable until the server decides to declare my validated work units as "complete" and start reporting average turn around time.

---

"Life is just nature's way of keeping meat fresh." - The Doctor

You and Snowmain could always abandon running the ATI MB app for now, and run the ATI Astropulse app instead:
1) The outier detection works on the the ATI AP app, so it will scale to the GPU.
2) The r555 ATI AP app is more efficient/faster than the ATI MB app, and the r1316/Stock version is faster still.

Claggy

A couple of things - your "valid" list for the GPU only shows about 4 or 5 valid, non-blanked tasks, a long way from your claimed 50 (which is more like the figure for your CPU). Second, "officially" we aren't meant to use <FLOPS>, but it works a dream, far better than the guestimates that BOINC generates, but they are a manual process and BOINC/S@H is meant to be a "fit and forget" process...

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

Yes I realize this but I had over 50 work units validated, matching results from nVidia GPU and CPU apps and NONE were considered valid enough to count?

Someone suggested and I puzzled out how and use the <flops> tag in my app_info.xml file to encourage more than a handful of gpu workunits to be cached. That's good enough for now.

If someday my apps actually start reporting "tasks completed", keep track of average turnaround time and calculate an average processing rate, I'll pull that line. Until then, at least I don't have to connect every few hours just to prevent the GPU queue from running dry.

Snow and me aren't the only ones seeing this. I found this thread while Googling for some insight.

---

"Life is just nature's way of keeping meat fresh." - The Doctor

Keith, a quick look at your stats suggests that your GPU hasn't managed to have enough results validated to get the various fudge factors corrected. These fudge factors enable the run time guestimate to be less inaccurate. This is a "feature" of the way BOINC works, each processor needs to be calibrated and until that calibration is done the amount of work in its cache will be rather limited as the cache size calculation uses the initial estimated times not the real times.

So if you have a cache set for five days, and a nice shiny new GPU installed BOINC may guess at a conservative 12hours runtime, so will limit the cache to ten jobs. The GPU may go though these jobs in 1 hour each, and return them in no short order, however you still have to wait for the wingman to return their copy of each of your jobs. And the cache is topped up with another 10 jobs as BOINC still doesn't know how fast your GPU is...This repeats until you have had enough work validated (its either ten or twenty VALIDATED, non-blanked tasks) for BOINC to work out that your GPU is taking 1 hour not 12, now suddenly you start to get work to fill that 5 day cache - instead of getting 10 jobs in the cache you find its got 120.

I know its a pain, but let BOINC do its thing - I allow a couple of weeks at least for a major hardware change like installing/enabling a GPU to properly take effect and get bedded in. (Its going to be "fun" watching my next cruncher get going in a few weeks...)

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

Well, I'm also not getting any of my valid ATI credits counting as completed which may be the reason the estimates aren't being adjusted and results in few ATI work units being downloaded, same as you.

And yes, in actuality my ATI units do fall in a broader range of 5 1/2 to 20 1/2 hour estimates which are actually completed in 20 minutes to 4 1/2 hours, depending if my GPU is running at full speed or is being throttled at under 1/2 clock speed, but that's an entirely different problem. It's the fact that both of us aren't getting enough work units for a single day rather than the amount we are asking for that leads me to believe we have the same problem.

The integrated ATI graphics in my A8-3820 are a tad slower than a discrete HD 5570 card (400SPs at 600MHz with DDR3 memory Vs 800SPs at 820MHz with GDDR5 memory in the case of your HD 5770) so that's where the real world performance difference comes from.

---

"Life is just nature's way of keeping meat fresh." - The Doctor

"Also my estimates for ATI units are way, way high. As in 19-20 hours per unit when they normally get done in under two (when the GPU isn't being throttled, then it's under four hours).

As it is I'm getting less than 1/2 a days worth of ATI units instead of 6 days my cache is set to and being without an "always on" internet connection I have to manually connect several times a day if I don't want the GPU to run out of work."

Exactly my same problem.(but not for much longer) Estimates 4-20 hours actually finish in 30minutes.


Not really sure if this could be related.

I downloaded MSI Afterburner to overclock the 8800gts. For that it has worked fine, however MSI can't see the ATI card. I have found posts of other HD5770 owners using MSI Afterburner to overclock this very card. Next I downloaded the AMD OverDrive application, it gives me am error message stating it can't find an ATI/AMD video card right when I start it.

http://www.overclock.net/t/672763/review-xfx-hd5770-1gb-xxx-edition
My card is a ref ATI manufacture card so perhaps the XFX version in this article is the difference.

I don't really think that has anything to do with the issue I am having but ya never know.

---

I'm in a similar situation with the Snowmain. I recently started to use the integrated GPU in my AMD A8 desktop and my application details also report 0 tasks complete even though I have 13 tasks completed and validated right now and none of them have errored or overflow out and most are paired with normal CPU wingmen and their credits seem to be in a normal range.

Also my estimates for ATI units are way, way high. As in 19-20 hours per unit when they normally get done in under two (when the GPU isn't being throttled, then it's under four hours).

As it is I'm getting less than 1/2 a days worth of ATI units instead of 6 days my cache is set to and being without an "always on" internet connection I have to manually connect several times a day if I don't want the GPU to run out of work.

I don't know if there is something missing in one of my XML files or what but I would like to be able to get more ATI work units down without having to connect more frequently than I want.

---

"Life is just nature's way of keeping meat fresh." - The Doctor

SLI is of no benefit to S@H, but may be of use in some other applications.

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

My 8800gts is an evga superclocker card...stock 8800gts run @ 500mhz. My card runs stock at 576Mhz....currently its overclocked to 703Mhz :)

.....................GPU....Shaders...Memory
Stock.............500....1200......800
superclocker...576....1350......850
overclocked....703....1674......1035

I really think memory bandwidth is one of the more important attributes of these cards. Overclocking my card has increased the already overclocked bandwidth from 67GB/s to 82.1GB/s.

I am infact gonna pull the ati. I have now given the system more than enough time to figure out what ever it was supposed to. The Ati is again out of work.

A freind has another 8800gts that I can probably get for for $20. Should I sli them or just let them run seperatly?

Or save the $20 get $50 or so from the sale of the ati and buy something else still?

---

The 10 ati unit have been trying to download all night ......

1) Then read again Message 1272810:
http://setiathome.berkeley.edu/forum_thread.php?id=69039&postid=1272810#1272810

... and start using SIV (see the screenshot)
(SIV 'Auto Retry' stuck downloads in periods of 3-15 minutes, BOINC own mechanism may delay retries for hours)

SIV - System Information Viewer:
http://rh-software.com/


2) As already noted - go here:
http://setiathome.berkeley.edu/prefs.php?subset=project

and select:
'Use NVIDIA GPU' no

(this 'Use' means in fact: 'Get tasks for NVIDIA GPU' no
It will not stop your NVIDIA GPU from computing tasks you already have.

This preference do not exist locally in BOINC - this answers your question:
"Online preferences? Didn't we disable those with the config file?"

"What do I need to add to the file to shutoff the nvidia card?"
Nobody said you need "to shutoff the nvidia card", only to temporarily disable the work fetch for it.
)

You may leave it this way for a few days (as you have already tasks for NVIDIA in the computer sufficient for a few days computing)

***

Why you need this 'things'?
A) Because BOINC still thinks: this NVIDIA is the fastest computing unit here so I will get tasks first for it and only after NVIDIA have 5 days WU cache I will ask for this 'slow' ATI
B) Because BOINC will not ask for any new tasks if you have stuck downloads (in retry/backoff)

So you may need forever "1) use SIV" (to often retry stuck downloads, it is the same as if you stay there and click [Retry])
And you may need to do "2) 'Use NVIDIA GPU' no" several times in the next 1-2 weeks until BOINC learns that your ATI is fast

---

 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 

"I would simply disable nvidia GPU in your preferences and only fetch ATI for a while."

How?
There is no labeled controls in Boinc to control spacific gpus its all or nothing.
Online preferences? Didn't we disable those with the config file? What do I need to add to the file to shutoff the nvidia card?

There are two sets of Preferences, one of them is called Computing Preferences, and is can be changed eithier on a projects Website, or Locally in Boinc Manager, this is known as as a Global preferences, ie you only need to adjust it one project for all projects to be effected,

the other preference is the project preferences, this is where you can select what devices to fetch work for, and what type of work to get, Setiathome project preferences just effect setiathome, while CPDN project preferences will just effect CPDN, etc,
project preferences are only available on the projects Website.

Claggy

Oh ok....on the hypothetical.

Still no one sees an issue with 11:1 nvidia/ati results. When the ati card is literally 5 times faster. Everything is fine? That isn't a red flag? Again 2 cards turned on,on the same day in the same system? 15 days ago long before this current outage.
I am beating the stuffing out of this horse it was dead on saturday, this just isn't making any sense to me.

I understand.

I also understand that there is nothing you can do about it or need to do about it. It will be what it is when it becomes what it is becoming.

Check my dual-brand machines. The story is much the same as yours. I've got days of total work and 24 hours of ATI work (if that).

If you want a better ratio, **turn off the NVIDIA work-fetch**.

How?

From your Account web-page to go "SETI@Home Preferences" and click-it.

You'll see "Primary Default Preferences." Leave it alone.

Scroll down the page and you'll see a way to set-up separate preferences for "home," "work," and "school." Choose one of those that you currently are not using and "edit preferences." I'm going to hypothetically say you are using "School."

When you are making selections under "School", un-check USE NVIDIA GPU, then click "Update Preferences."

Go back to your "Account" page, go to View next to "Computers on This Account" and click it.

Go to your mixed-GPU computer and click "Details."

At the bottom of the page you'll find a drop-down box. It probably says "home" now (I haven't looked). Select "School" from the Drop-Down box.

BINGO! You shouldn't get any more NVIDIA work units.

Does that make sense?

Just remember to go back there (the drop-down box under "Details" for that computer) and move it back to "home" once you have all the ATI work-units you want.

I use this to prevent specific computers from getting AP work units. I can move back and forth at will.


HOWEVER ---- If I were you, right now, while we recover from the outtage, I'd leave it alone and let it get all the work it can get no matter what kind of work that is.

Now I'm going to be the dead-horse slapper and tell you that preventing your computer from getting NVIDIA work won't "create" ATI work for you to download. You do not have a very large cache of work. Leave it alone. Be patient.