Simple - the limit is a simple numeric base, not a performance base.
For each CPU you get a number of WU
For each GPU you get A larger number of WU

It appears that a multi-core CPU is only counted as one CPU, in the same way that a multi-core GPU only counts as one GPU

---

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

I'm starting to wonder about how my single-core machine can have more than 50 in progress tasks. I know those of you with GPUs have demonstrated that once you get to 450, you get the limit message, but has there been a test where there is no GPU at all? I know you can just disable GPU crunching, but I'm trying to figure out if this is just one of those things where it's an old, pre-GPU client, or if it is because I don't have a GPU and haven't hit 450, or what?

Any quad-core people out there want to pull their GPU and see if that's the case?

My computer in the last hour or so has started making separate requests for cpu and gpu. I haven't counted, but I get the reached limit for CPU and sometimes get taasks for the GPU. At present total in progress is 192.

I've been bouncing off the rev limiter (450 w/u's) for days now and still bouncing as we speak.

Cheers.

---

Simple - the limit is a simple numeric base, not a performance base.
For each CPU you get a number of WU
For each GPU you get A larger number of WU

It appears that a multi-core CPU is only counted as one CPU, in the same way that a multi-core GPU only counts as one GPU

I get that. 1-200+ cores is still only supposed to get 50 tasks total. I've got right around double that and am not getting the message about a limit. Not at all complaining or trying to boast about my good fortune, but either there's a glitch in the system, or something makes my machine exempt/unique.

---

Linux laptop:
record uptime: 1511d 20h 19m (ended due to the power brick giving-up)

Simple - the limit is a simple numeric base, not a performance base.
For each CPU you get a number of WU
For each GPU you get A larger number of WU

It appears that a multi-core CPU is only counted as one CPU, in the same way that a multi-core GPU only counts as one GPU

I get that. 1-200+ cores is still only supposed to get 50 tasks total.

That's not my understanding.
I understand it as the limit being per Processor. My E6600 is Dual Core, so it counts as 2 Processors. My i7 is Quad Core with HyperThreading, so it counts as 8 Processors.
With a limit of 50 per processor my E6600 can get upto 100, the i7 upto 400 (would be nice if it were possible, but with the rate i return them at, and the amount of time the Scheduler is not available i'm not likely to hit those limits any time soon).

---

Grant
Darwin NT

It appears that a multi-core CPU is only counted as one CPU, in the same way that a multi-core GPU only counts as one GPU

AFAIK a multi core CPU counts as multiple processors. Multiple cores in a GPU isn't the same thing as multiple cores in a CPU, hence 1 GPU is 1 GPU. A video card with 2 GPUs on it counts as 2 GPUs.

---

Grant
Darwin NT

The limit is for one host.
I´ve checked that out yesterday no one gets more than 450 units no matter how much CPUs and GPUs they have running.

---

With each crime and every kindness we birth our future.

The limit is for one host.
I´ve checked that out yesterday no one gets more than 450 units no matter how much CPUs and GPUs they have running.

I provide this information as information. I am not being argumentative.

6 of my 19 machines have more than 450 wu right now. None of the 6 machines has a GPU, they are dual core pentium.

I believe that Boinc/Seti has issues and is somewhat out of control.

---

Dave

The limit is for one host.
I´ve checked that out yesterday no one gets more than 450 units no matter how much CPUs and GPUs they have running.

I provide this information as information. I am not being argumentative.

6 of my 19 machines have more than 450 wu right now. None of the 6 machines has a GPU, they are dual core pentium.

I believe that Boinc/Seti has issues and is somewhat out of control.

I didn´t say the code works pretty well.

I dont get 100 to keep my machine busy.
Others getting to much.
Thats life.

---

With each crime and every kindness we birth our future.

The limit is for one host.
I´ve checked that out yesterday no one gets more than 450 units no matter how much CPUs and GPUs they have running.

Bugger.
That sucks.

---

Grant
Darwin NT

That's not my understanding.
I understand it as the limit being per Processor. My E6600 is Dual Core, so it counts as 2 Processors. My i7 is Quad Core with HyperThreading, so it counts as 8 Processors.
With a limit of 50 per processor my E6600 can get upto 100, the i7 upto 400 (would be nice if it were possible, but with the rate i return them at, and the amount of time the Scheduler is not available i'm not likely to hit those limits any time soon).

That might be the way your PC is working but it doesn't seem that way on mine, at least as far as CPU work is concerned. I have a dual core with a WU unit being worked on by each core but the limit seems to be a total of 50. Whenever it drops to 49 or 48 it gets bumped back up to 50 (and only 50) within 30 minutes so I do not believe it is a case of there not being any work or the machine not being able to receive it. It is either a case of a limit being set somewhere of either 25 WUs per core or 50 per machine. If you are receiving 50 per core all I can say is lucky you and I am envious.


17567 ? RNl 42:51 ../../projects/setiathome.berkeley.edu/AK_V8_linux32_ssse3
21529 ? RNl 30:09 ../../projects/setiathome.berkeley.edu/AK_V8_linux32_ssse3

With twin Xeons( 8 physical/ 16 HT cores) and a GTX 590(2 GPU's)
I can confirm: I reach a max of 450 tasks, when the scheduler connection permits.

---

Janice

My Intel® Core™2 Duo CPU T8100 has two processors, and the current limit for this rig is 50 tasks.

---

Can +1 on being over limits for maybe 24 hours. Now back under, due mostly to connect failures and intermittent limits response on successful connects. Don't think it is platform. Will provide additional details/comments if there is interest.

---

Another Fred
Support SETI@home when you search the Web with GoodSearch or shop online with GoodShop.

Close to running on empty.

---

The limit is for one host.
I´ve checked that out yesterday no one gets more than 450 units no matter how much CPUs and GPUs they have running.

I provide this information as information. I am not being argumentative.

6 of my 19 machines have more than 450 wu right now. None of the 6 machines has a GPU, they are dual core pentium.

I believe that Boinc/Seti has issues and is somewhat out of control.

You and Cosmic_Ocean are both running older versions of BOINC, and it looks like that may be the reason the limits are not being applied. Looking in the top_hosts list starting several thousand back so there would be a reasonable number of hosts running only CPU crunching, those with BOINC 6.6.20 and later are being limited, but those with 6.4.5 and earlier are not.

The most obvious client change between those versions is that 6.4 and earlier don't report runtime, but there were many other changes too.

Note that the limits are at least in part being applied as protection against overfetching as the capping of server scaling of task estimates is returned toward normal operation. IOW if you have a large cache setting and a small fractional DCF now, if the servers send much shorter estimates your host will ask for more work than you really want, and continue doing so until one of those new tasks is crunched and drives DCF up to something near 1. Anyone considering going to an old version of BOINC to build cache will risk that overfetch unless they also set a reasonably small cache, and of course the older versions are not practical for those doing GPU crunching.

                                                                  Joe

The limit is for one host.
I´ve checked that out yesterday no one gets more than 450 units no matter how much CPUs and GPUs they have running.

I provide this information as information. I am not being argumentative.

6 of my 19 machines have more than 450 wu right now. None of the 6 machines has a GPU, they are dual core pentium.

I believe that Boinc/Seti has issues and is somewhat out of control.

You and Cosmic_Ocean are both running older versions of BOINC, and it looks like that may be the reason the limits are not being applied. Looking in the top_hosts list starting several thousand back so there would be a reasonable number of hosts running only CPU crunching, those with BOINC 6.6.20 and later are being limited, but those with 6.4.5 and earlier are not.

The most obvious client change between those versions is that 6.4 and earlier don't report runtime, but there were many other changes too.

Note that the limits are at least in part being applied as protection against overfetching as the capping of server scaling of task estimates is returned toward normal operation. IOW if you have a large cache setting and a small fractional DCF now, if the servers send much shorter estimates your host will ask for more work than you really want, and continue doing so until one of those new tasks is crunched and drives DCF up to something near 1. Anyone considering going to an old version of BOINC to build cache will risk that overfetch unless they also set a reasonably small cache, and of course the older versions are not practical for those doing GPU crunching.

                                                                  Joe

Well I don't know about right now, but a few weeks ago when my same single-core machine in question went from AP-only to MB-only, the estimates were ridiculously tiny. So much so, that shorties take right at 2:20 (h:mm) but the estimates were showing between 3 and 8 minutes. This led my 10-day cache to turn into.. I think it was 586 WUs. I was completing these WUs as the runaway cache was being filled and the estimates were increasing, but not drastically. 8 minutes turned into 12, then just under 20, etc.

It was a gradual increase that resulted in a reasonable decrease in the number of seconds in the work request. Eventually, all the estimates were pretty close to correct and EDF mode began. I pulled up BoincTasks and it said it was ~267 days worth of cache. I selected all but the first 20 in the list and aborted them.

Point is.. if anyone is going to do that.. go with a tiny cache.. like 0.5 days max, until you figure out what the estimates are really going to be, then increase it from there.

---

Linux laptop:
record uptime: 1511d 20h 19m (ended due to the power brick giving-up)

Given the difficulties getting work over the last month or so, i can't see too big a cache being a problem for a while yet.

---

Grant
Darwin NT

With the current scheduler difficulties, I think it would be virtually impossible to get an overload of units in under a 24 hour period.

Edit: plus, I just got a bunch of these...



core_client_version>6.10.58</core_client_version>
<![CDATA[
<message>
WU download error: couldn't get input files:
<file_xfer_error>
<file_name>20ap11ad.30869.11928.12.10.59</file_name>
<error_code>-200</error_code>
</file_xfer_error>

</message>
]]>

---

Janice

Ok, things are changing. After several days of not running, ap_validate3 is now online.

Wow. I just looked at the server status page and somehow didn't notice that. It's amazing. Maybe a bunch of my pendings will finally be validated and moved into the ridiculous purge backlog.

---

Linux laptop:
record uptime: 1511d 20h 19m (ended due to the power brick giving-up)