You're right Juan, looks like 1 GPU task is about 13 minutes for me while 2 GPU tasks finish between 18 to 20 minutes, so that is an improvement. I'll add AP config too.

I'm trying this out:
<app_config>
<app>
<name>setiathome_v7</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>0.06</cpu_usage>
</gpu_versions>
</app>
<app>
<name>astropulse_v6</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>1.00</cpu_usage>
</gpu_versions>
</app>
</app_config>

Is AP finished? I haven't seen one in a long time.

Glad it works, you could try 3 and see if works even faster, if no back to 2. I realy can´t tell because i don't have any 660TI to test.

In my 670/690 a MB WU takes 15-18 mins when crunching 2 WU at a time, and 10-12 min when crunch one, so your times are in the expected range. In theory the 670 is a little faster an the 660TI but thats heavely depends if you use OC or not. I don´t use.

No AP is not finished but the problem is after V7, AP "paids" 2x more credit than MB so most of the heavy crunchers starts to crunch AP then it´s hard to get any new AP WU this times, but if you allow AP in your preferences eventualy you will receive some.

Did you leave one core free? Try to run with and without that core free, and look the times, choose the fast of the options.

One tip: you must try to use the lunatics instaler, that will help you to configurate even better your host since you have a cuda5 GPU and a AVX capable CPU.

---

One tip: you must try to use the lunatics instaler, that will help you to configurate even better your host since you have a cuda5 GPU and a AVX capable CPU.

I installed lunatic, it changed the exe files for the tasks, but doesn't seem to have done anything except change cpu_usage for MB from .06 to .04. I was hoping to get AP, but none were downloaded. I went and changed app_info.xml count filed from 1 to .5, that's it.

Its not going to be too long before another batch of tapes are loaded and the AP splitters can get their teeth into them. Each 50Gb "tape" is split twice, once by the AP splitters, and once by the MB splitters. Because APs are bigger and have less overlap than MBs there are fewer produced from each tape, and each AP work unit (the product of a splitter) takes less time than an MB work unit to produce.

---

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

One tip: you must try to use the lunatics instaler, that will help you to configurate even better your host since you have a cuda5 GPU and a AVX capable CPU.

I installed lunatic, it changed the exe files for the tasks, but doesn't seem to have done anything except change cpu_usage for MB from .06 to .04. I was hoping to get AP, but none were downloaded. I went and changed app_info.xml count filed from 1 to .5, that's it.

When you install you select cuda5 and crunching ap on the nvidia?

You don´t need to change the app_info the app_config do the job for you.

---

One tip: you must try to use the lunatics instaler, that will help you to configurate even better your host since you have a cuda5 GPU and a AVX capable CPU.

I installed lunatic, it changed the exe files for the tasks, but doesn't seem to have done anything except change cpu_usage for MB from .06 to .04. I was hoping to get AP, but none were downloaded. I went and changed app_info.xml count filed from 1 to .5, that's it.

When you install you select cuda5 and crunching ap on the nvidia?

You don´t need to change the app_info the app_config do the job for you.

yes

One tip: you must try to use the lunatics instaler, that will help you to configurate even better your host since you have a cuda5 GPU and a AVX capable CPU.

I installed lunatic, it changed the exe files for the tasks, but doesn't seem to have done anything except change cpu_usage for MB from .06 to .04. I was hoping to get AP, but none were downloaded. I went and changed app_info.xml count filed from 1 to .5, that's it.

When you install you select cuda5 and crunching ap on the nvidia?

You don´t need to change the app_info the app_config do the job for you.

yes

Then you could look by taskmgr you are crunching cuda5 on the GPU (cuda5 is faster than the previous cuda versions if you kave a keppler GPU like your) even if the boincmgr could say youar ecrunching other cuda. So you will obtain an aditional gain in performance. You could see what i talk, by going to your compleatesd WU and look the times compare those with cuda5 with the others (with the same AR of course).

We all are waiting for AP WU, maybe they apears tomorrow after they DL the new tapes. Just wait.

---

So now I have 32 threads.

This morning I did AP GPU in 67 minutes, but I noticed my settings had cpu_usage at 0.04 by mistake. Now that AP GPU is set to cpu_usage 1, I'll see if the next one is done faster.

Tonight is the first time I'm seeing AP CPU, I'm estimating 14 hours for each, wow. I guess my cpu clock speed is too slow, with turbo my frequency goes from 2.3 to 2.6 ghz. I will be getting better cooling, so that should help turbo in the future.

So now I have 32 threads.

This morning I did AP GPU in 67 minutes, but I noticed my settings had cpu_usage at 0.04 by mistake. Now that AP GPU is set to cpu_usage 1, I'll see if the next one is done faster.

Tonight is the first time I'm seeing AP CPU, I'm estimating 14 hours for each, wow. I guess my cpu clock speed is too slow, with turbo my frequency goes from 2.3 to 2.6 ghz. I will be getting better cooling, so that should help turbo in the future.

32 threads...Fine i wish i have too (i only have 4 cores...), you don´t need to worry about the estimated times, the times will be adapt, mines for example are 6hr in this slow I5 host @ 2.90GHz, but when it´s start you will notice they will drop very fast.

Running 2 WU at a time a 65 min AP is within the expected time. In a 670/690 the times are in the range of 55-65 min. This times are highly dependant onn the % or Blanks in the file. But remember i allow to run only 1AP+1MB at the same time (i have few cores avaiable to feed the GPU´s)

Better cooling is allways good, the crunchers runs hotter than normal, and heat is allways our enemy. So keep an eye on the temperatures is allways important.

If you have some trouble with tthe heat you could use a program like Fred´s Throttle, it helps to keep your CPU within the safe temperature range, take a look on it at: http://www.efmer.eu/boinc/

And don´t forget to use a program like EVGA Presision/MSI Afterfurner (or any other) to increse the GPU fan speed to keep the GPU temp´s within the safe range too.

---

@Juan, I'm using HWMONITOR, both Nvidia and Intel will throttle down due to heat, i'm not worried.

AP CPU came in around 13.5 hours, I have 4 more crunching, my own estimated says 13.5 hours too.

Thanks everyone for all your help, I'm getting AP gpu/cpu, cuda tasks are more efficient too.

@Juan, I'm using HWMONITOR, both Nvidia and Intel will throttle down due to heat, i'm not worried.

AP CPU came in around 13.5 hours, I have 4 more crunching, my own estimated says 13.5 hours too.

Thanks everyone for all your help, I'm getting AP gpu/cpu, cuda tasks are more efficient too.

Throttling is something you want to avoid otherwise you'll start generating "exceed time limit" errors. ;-)

Cheers.

@Juan, I'm using HWMONITOR, both Nvidia and Intel will throttle down due to heat, i'm not worried.

AP CPU came in around 13.5 hours, I have 4 more crunching, my own estimated says 13.5 hours too.

Thanks everyone for all your help, I'm getting AP gpu/cpu, cuda tasks are more efficient too.

Throttling is something you want to avoid otherwise you'll start generating "exceed time limit" errors. ;-)

Cheers.

Aw hell, ok, so I have 58C on CPU1 and 66 on CPU2, intel's states 73C is their max temp for the e5-2660. I expect these temps to go down once I add new CPU coolers (contained liquid coolers).

As for the gtx 660 ti, I saw max 69C and fans were spinning at 35%, strange. Nvidia states 97C is the max for this card.

I don't know when throttling starts, but CPU2 was still going into turbo, around 2.3-2.4ghz while at 66C, so that's good.

I don´t know what MB you use, mine have a set in the bios to keep the CPU fan at haigh speed, maydbe you have something similar.

35% of fan speed when crunching on the GPU is unusual, but your temp are fine 69C are you sure? i use EVGA precision and modify the fan curve to mantain the GPU cool as possible.

i could be radical but i don´t trust in normal setings they could work for gaming, but not for 24/7 usage, so i allways use turbo mode on the temp controls to keep the tempslower as possible.

---

As for the gtx 660 ti, I saw max 69C and fans were spinning at 35%, strange. Nvidia states 97C is the max for this card.

I just installed a GTX 660 Ti this lunchtime, replacing an 8800 GTS 512 in a Linux (SLC6) machine (Quad-core Q9550). I didn't look at the 8800's temps but with one instance of MB running on the 660 nvidia-smi showed 59 C and the fan at 17%. Now I'm trying two instances and showing 68 C and 37%.
Out of interest, at one instance it took 165 secs CPU in 2055 secs real-time to do a work-unit worth 88 creds (314 MB memory used); at two instances it's taking 140 secs CPU in 2205 secs for the same credit (574 MB). The 8800, which can only run one instance, was using 180 secs CPU in 5,000 secs for 110 creds -- obviously there is some variation in the WUs there.

---

As for the gtx 660 ti, I saw max 69C and fans were spinning at 35%, strange. Nvidia states 97C is the max for this card.

I just installed a GTX 660 Ti this lunchtime, replacing an 8800 GTS 512 in a Linux (SLC6) machine (Quad-core Q9550). I didn't look at the 8800's temps but with one instance of MB running on the 660 nvidia-smi showed 59 C and the fan at 17%. Now I'm trying two instances and showing 68 C and 37%.
Out of interest, at one instance it took 165 secs CPU in 2055 secs real-time to do a work-unit worth 88 creds (314 MB memory used); at two instances it's taking 140 secs CPU in 2205 secs for the same credit (574 MB). The 8800, which can only run one instance, was using 180 secs CPU in 5,000 secs for 110 creds -- obviously there is some variation in the WUs there.

With three instances, it's hovering around 69 C and 40% fan -- I should mention that it's 26 C in my office... The credits/minute/instance are still the same, or even a bit higher, but there's more variability in the WUs just now. Guess I have to try four instances tomorrow.

---

Don´t go so far, first look you GPU usage, if it is allready near to 98% you allready reach the best number. More task makes you just waste time in the task switching logic and at the end produces less work.

It´s a general wrong ideia, more are better, even if you could run a lot of WU at a time, the actual performance is a curve like a senoidal wave, when you pass the optimal number (max) you just loose time.

Very few High End GPU can handle 4 WU at a time specialy after V7. Even a 770 in a Highly OC host has troubles to run 4 at a time.

---

Don´t go so far, first look you GPU usage, if it is allready near to 98% you allready reach the best number. More task makes you just waste time in the task switching logic and at the end produces less work.

It´s a general wrong ideia, more are better, even if you could run a lot of WU at a time, the actual performance is a curve like a senoidal wave, when you pass the optimal number (max) you just loose time.

Very few High End GPU can handle 4 WU at a time specialy after V7. Even a 770 in a Highly OC host has troubles to run 4 at a time.

Unfortunately, nvidia-smi doesn't report GPU usage("GPU-Util N/A"). However, at four instances it's still 70C/40% and the credits/minute/instance are also much the same as before -- certainly no dropoff. (Remember, this thing has 1344 cores; I haven't checked the code yet to see if any of the major operations have hardwired execution parameters.) Memory usage is now up to 1093 MB. I'm going to let it sit at four instances until the RAC stabilises before I try anything else.

---

Out of interest, at one instance it took 165 secs CPU in 2055 secs real-time to do a work-unit worth 88 creds (314 MB memory used); at two instances it's taking 140 secs CPU in 2205 secs for the same credit (574 MB). The 8800, which can only run one instance, was using 180 secs CPU in 5,000 secs for 110 creds -- obviously there is some variation in the WUs there.

I have noticed some variation in the MB units; if I run one at a time, some will use ~70% GPU and be very stable with their loads, others use ~90% on their own with spikes to 96%



Ended up reverting to 1 at a time on my 560Ti; when I was getting 2 of the 'spiky' units going at the same time it caused the board to drop to a low power state, fixed only by a restart. Thought I had that solved with some adjustments, but I've got 2 units failed from this morning that did the same thing.

Edit : Graphs are from the free utility 'Nvidia Inspector'.

Thanks. Juan, do you use Nvidia Inspector or some other program to monitor gpu usage?

I use EVGA Precision that allow you monitoring the GPU usage, control the fans, power usage, etc. Try it is free and easy to use, DL it from the EVGA site.

---