I have always crunched MB's because I felt my Nvidia card was "too small" for AP.

By bad timing i by chance replaced my 550TI with a 560TI(448 core) on the same day MB v7 was introduced.

SO ...

Now that I have a "bigger" GPU engine I want to try my hand on AP's.

My simple minded question is this:

Is it "most efficient" to run
two MB's per gpu card or
one MP and one AP per GPU card or
two AP's per GPU card?

(assuming a 448 core 560 TI)

"most efficient" == most science done / real-time second (credits-be-hanged).

Ed F

Let me assure you, that your 550 Ti's are not puny for AP work. I have two machines that run AP on 550 Ti's.

I only run one wu per GPU but I read where most can run two per card or 3 But the only way you can tell is to try it.

As for your 560 Ti cards, try 2 wu per card if its stable try 3.

---

[/quote]

Old James

My simple minded question is this:

Is it "most efficient" to run
two MB's per gpu card or
one MP and one AP per GPU card or
two AP's per GPU card?

(assuming a 448 core 560 TI)

"most efficient" == most science done / real-time second (credits-be-hanged).

Ed F

The simple answer: You don't get to choose. The manager picks up task and set them to run. It could be two MB, it could be two AP, it could be 1 MB and 1 AP. You don't get to choose.

The other thing. The limiting factor on how many tasks to run on gpu is the memory on the video card. If you have 512mb, you can run 1 or 2. 1GB will let you run 3, 4 is possible, but would need to be examined with gpu-z to make sure it will fit. If you had 2GB, you could have run 7 tasks, maybe 8.

I run GTX 460, I do just one since I also play games and stuff.

---

I only run one wu per GPU but I read where most can run two per card or 3 But the only way you can tell is to try it.

Boy am I confused ... I'll have to give it some thought ...

My GPU has 448 cores you have 192
My GPU runs at 1594 Mhz yours at 1800 Mhz
My typical AP WU takes 3760 sec's yours 6585 sec's
I'm using driver 314.22 you are using driver 320.49.
both have the same version of opt code.
I'm running 2 AP WU's at a time and you 1?? ... are you running MB WU's at the same time as AP WU's on the GPU??

please let me know ... thanks

Ed F

The simple answer: You don't get to choose. The manager picks up task and set them to run. It could be two MB, it could be two AP, it could be 1 MB and 1 AP. You don't get to choose.

yes ... and ... no

I have it set up to run only on the GPU and to only run AP WU's .

I can run AP/AP or MB/MB or MB/AP or mix-MB-AP/mix-MB-AP because I am running 2 copies of BOINC on the same computer. I only need to change what "location" the particular BOINC instance (i.e. computer) references in order to lock it to AP only, MB only, or 'take whatever you've got' etc.

So I still need to (would like advice to) know what is the best mix for getting science done.

perhaps it's just ---leave it alone ---

Ed F

I only run one wu per GPU but I read where most can run two per card or 3 But the only way you can tell is to try it.

Boy am I confused ... I'll have to give it some thought ...

My GPU has 448 cores you have 192
My GPU runs at 1594 Mhz yours at 1800 Mhz
My typical AP WU takes 3760 sec's yours 6585 sec's
I'm using driver 314.22 you are using driver 320.49.
both have the same version of opt code.
I'm running 2 AP WU's at a time and you 1?? ... are you running MB WU's at the same time as AP WU's on the GPU??

please let me know ... thanks

Ed F

The limiting factor on how many tasks to run on gpu is the memory on the video card.



Here is example of 8 tasks. 1 AP and 7 MB. It uses 1.8GB of memory on the memory card.

---

I don't know what size your GTX550TI is but you shouldn't have any problem running it alongside the GTX560TI as they are both Fermi class GPUs. I wouldn't want to run 3 WUs at a time, although it is possible. I just got finished running a test running 3 v7 MB WUs on 2 GTX660SC @ 2Gb and I felt that the time spent to process them was not worth it.

I see that machine 7005699 is also working OpenCL, have you thought of installing the 550 on that machine and turning it into a pure AP cruncher? If not, you might want to give it a thought. Take a look at my i7/950 machine as it is primarily an AP machine and v7 is only allowed on it if there are no AP WUs available for a couple of days as it happened this past week.

---

I don't buy computers, I build them!!

I only run one wu per GPU but I read where most can run two per card or 3 But the only way you can tell is to try it.

Boy am I confused ... I'll have to give it some thought ...

My GPU has 448 cores you have 192
My GPU runs at 1594 Mhz yours at 1800 Mhz
My typical AP WU takes 3760 sec's yours 6585 sec's
I'm using driver 314.22 you are using driver 320.49.
both have the same version of opt code.
I'm running 2 AP WU's at a time and you 1?? ... are you running MB WU's at the same time as AP WU's on the GPU??

please let me know ... thanks

Ed F

I choose to only run one AP or MB at a time on my rigs with the 550Ti's. They are stable and temps run cool. I dont overclock for the same reasons.

---

[/quote]

Old James

I have always crunched MB's because I felt my Nvidia card was "too small" for AP.

By bad timing i by chance replaced my 550TI with a 560TI(448 core) on the same day MB v7 was introduced.

SO ...

Now that I have a "bigger" GPU engine I want to try my hand on AP's.

My simple minded question is this:

Is it "most efficient" to run
two MB's per gpu card or
one MP and one AP per GPU card or
two AP's per GPU card?

(assuming a 448 core 560 TI)

"most efficient" == most science done / real-time second (credits-be-hanged).

Ed F

Both those cards are capable of the 1st 2, the 560Ti 448 may be capable of the last (I only had the standard 560Ti), but you will need to watch your temps with the 560Ti.

And remember to have a free CPU core for each AP you do on the cards.

Cheers.

So I still need to (would like advice to) know what is the best mix for getting science done.

Ed F

Theoretically MB + AP would be the best mix. Optimal AP number varies with blanking. But MB load relatively steady. So, MB would "backup" your GPU in case of high blanking AP set.
Also, time to time there is lack of AP tasks. Again, MB on GPU will help with that.
Regarding to worth or not to run NV AP at all - different point of view exists. I think it's worth on FERMI at least. But did not done relative comparison of timings so it's just guess.
(how I would decide worth or not: to get mean CPU MB and AP times. Get mean GPU MB and AP times, compare CPU/GPU ratio for both apps. App with bigger ratio better suit for GPU cause provide more speedup).

---

SETI apps news
We're not gonna fight them. We're gonna transcend them.

I only run one wu per GPU but I read where most can run two per card or 3 But the only way you can tell is to try it.

Boy am I confused ... I'll have to give it some thought ...

My GPU has 448 cores you have 192
My GPU runs at 1594 Mhz yours at 1800 Mhz
My typical AP WU takes 3760 sec's yours 6585 sec's
I'm using driver 314.22 you are using driver 320.49.
both have the same version of opt code.
I'm running 2 AP WU's at a time and you 1?? ... are you running MB WU's at the same time as AP WU's on the GPU??

please let me know ... thanks

Ed F

The limiting factor on how many tasks to run on gpu is the memory on the video card.



Here is example of 8 tasks. 1 AP and 7 MB. It uses 1.8GB of memory on the memory card.

Actually you are way off base and giving bad advice. Notice your GPU tasks are at 1% and have an estimated time of over 2 hours. That is just plain bad.

Here's a little experiment you can do. See how many you can get done in say 3 hours running 8 at a time. Then for giggles try 3 at a time.

Remember that GPU-Z is your friend. Most people attempt to run 2 at a time then if the GPU isn't maxxed out to 100% you can run a 3rd WU. most if not all GPU's cannot handle more tasks than 3 without slowing down the process as to make it take longer than a CPU. THis seems to be the case with your GPU.

Please read what I've written carefully otherwise you are wasting your time and your GPU power.

summary run 2 and if it doesn't max out your GPU to 100% then do 3.

---

In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope

I only run one wu per GPU but I read where most can run two per card or 3 But the only way you can tell is to try it.

Boy am I confused ... I'll have to give it some thought ...

My GPU has 448 cores you have 192
My GPU runs at 1594 Mhz yours at 1800 Mhz
My typical AP WU takes 3760 sec's yours 6585 sec's
I'm using driver 314.22 you are using driver 320.49.
both have the same version of opt code.
I'm running 2 AP WU's at a time and you 1?? ... are you running MB WU's at the same time as AP WU's on the GPU??

please let me know ... thanks

Ed F

The limiting factor on how many tasks to run on gpu is the memory on the video card.



Here is example of 8 tasks. 1 AP and 7 MB. It uses 1.8GB of memory on the memory card.

Actually you are way off base and giving bad advice. Notice your GPU tasks are at 1% and have an estimated time of over 2 hours. That is just plain bad.

Here's a little experiment you can do. See how many you can get done in say 3 hours running 8 at a time. Then for giggles try 3 at a time.

Remember that GPU-Z is your friend. Most people attempt to run 2 at a time then if the GPU isn't maxxed out to 100% you can run a 3rd WU. most if not all GPU's cannot handle more tasks than 3 without slowing down the process as to make it take longer than a CPU. THis seems to be the case with your GPU.

Please read what I've written carefully otherwise you are wasting your time and your GPU power.

summary run 2 and if it doesn't max out your GPU to 100% then do 3.

The limiting factor on how many tasks to run on gpu is the memory on the video card.
I run GTX 460, I do just one since I also play games and stuff.

---

Slightly correction.
Some cards can even run 5 instances.
Depends on VRAM and CU`s.

---

With each crime and every kindness we birth our future.

Again, wrong. I have a 3 Gb AMD 7970 and it can only run 3 tasks at a time. You could set it to run 50 if you want and the GPU will always read 100% but you are sharing bandwidth with the other tasks. My GPU-z reads 95-100% GPU load. So the reality is your GPU can run 8 10 or even 20 at a time but you lose efficiency when you run to many. You even slow them down to a point that you are doing less than if you'd just ran 1 at a time.

The whole purpose of running multiple tasks is to get the most benefit from your GPU without taking longer to run than running 1 at a time.

So to recap. start by turning on GPU-z. Start BOINC and see how running 1 at a time looks. If it runs smoothly move up to 2 at a time. Did it hit more 90% or more GPU load? if yes then you are done. If not, continuing with 3 and dod the same process over. keep going if you don't hit 100%.

As for 8 tasks at once, all you need to do is look at the estimated times. 3+ hours to complete a task. I derive this by estimating your % done vs time. That seems slow even for a 460. assuming those tasks take 3 hours to complete, you've done 8 tasks in 3 hours. when you are running tasks at 20 minutes or less for standard 0.42 AR WU's when you could have done 12 just by running 1 at a time. Imagine running 24 in about the same time by just running 2X.

---

In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope

I dont want to argue.
It works fine on a 7970.
GPU usage is not everthing.
It depends mostly on blankings.

---

With each crime and every kindness we birth our future.