I have 2 750Ti's and in general I'm happy with them, they've added 25% to the RAC of the computer compared to the 550Ti's I was running previously

The problem I do have is that if the card temperature exceeds 55 deg C the cards either down clock, crash the driver or both. As the room temperature at this time of year is often around 30deg C this is proving a PITA.

The cooling fans are running at 100% and I have extra fans to blow additional air to the cards.

I'm not after help, just giving a warning to anyone from a hot climate who may be thinking of buying these cards.

T.A.

The problem I do have is that if the card temperature exceeds 55 deg C the cards either down clock, crash the driver or both. As the room temperature at this time of year is often around 30deg C this is proving a PITA.

Thanks for posting that. Considering how cool they run when ambient temperatures are low, I have wondered how far they would go without down-clocking.

Just to clarify for me, do you have Precision X set to allow for higher temperatures?

For anyone who doesn't know, Precision X will let you set a power target or a temperature target and prioritize either way.

The problem I do have is that if the card temperature exceeds 55 deg C the cards either down clock, crash the driver or both. As the room temperature at this time of year is often around 30deg C this is proving a PITA.

The cooling fans are running at 100% and I have extra fans to blow additional air to the cards.

What is your card? I'm running with Asus GTX750TI-OC-2GD5 and it doesn't have any problems, GPU-Z shows did show about 55C temps and fans did run about 70% when room temperature was about 30 at summer.

The cards are Asus GTX750TI-PH-2GD5

I've set the fans to 100% on both cards using Precision X.

I only crunch MB at 1 unit per card on this machine (It's my Daily Driver so I can't push it too hard), GPUZ shows the average TDP at ~60%.

T.A.

The cards are Asus GTX750TI-PH-2GD5

I've set the fans to 100% on both cards using Precision X.

I only crunch MB at 1 unit per card on this machine (It's my Daily Driver so I can't push it too hard), GPUZ shows the average TDP at ~60%.

T.A.

One of the first questions that I would ask at this point is - How's your cable management? Does your case allow for you to put the majority of cables behind the board? If not are you running the cables out of the airflow to the top exhaust? I'm running 2 x GTX750Ti FTW w/ 2tasks each and very seldom do they get above 55c even when the ambient temps are above 30c. Two of the reasons why is cable management and the case, which is a full tower allowing for max air flow. Even when I ran a Q6600 w/ 2 x GTX660SC, in the same case, I didn't have to worry about GPU temps. Also, when was the last time you cleaned out the dust bunnies on the GPUs?

---

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

The problem I do have is that if the card temperature exceeds 55 deg C the cards either down clock, crash the driver or both.

Sounds like you have an issue with your cards or your power supply.
I'm running 4 GTX 750Tis, 2 in each system, and here the temperatures are around 34°c for most of the day (30° being the coolest it's been in that room for several months now).
The hottest card is 60° at 75% of maximum fan speed when the ambient temp is at 34.5°c. No problems at all.

---

Grant
Darwin NT

Still running stock AP v7 (with two v6 in queue), RAC is now about 13.5K.

RAC now about 16K, didn't get any task for while so RAC is lower as it should be.

RAC still about 16K, it was over 17K but then "You know what" hit the S@H fans... so waiting things to clear up.

---

"Please keep Your signature under four lines so Internet traffic doesn't go up too much"

- In 1992 when I had my first e-mail address -

Which one is fastest running MBv7 stock application?

My host runs only 1 task and cuda42 seems to be faster.

I don't know how many tasks these hosts run at time but Average processing rate is much lower than mine:

This host cuda42 is faster

This host cuda50 is faster

cuda42 or cuda50?

Just to know if I install optimized application, which one to select.

---

"Please keep Your signature under four lines so Internet traffic doesn't go up too much"

- In 1992 when I had my first e-mail address -

Stick with cuda 50 for now

Stick with cuda 50 for now

Yeah, I know and I will install someday cuda50 optimized app.

I did run several stock application tasks (over required 11), cuda50 was fastest. Then I did test optimized cuda50 application, faster than stock.

But now stock-cuda42 seems to be faster? What did happen that servers do think that cuda42 is faster than cuda50? Shorties? Or something else?

---

"Please keep Your signature under four lines so Internet traffic doesn't go up too much"

- In 1992 when I had my first e-mail address -

Here's the thing. Neither Cuda 42 nor cuda 50 were written for the Maxwell chip.

I've seen the same thing as you. Sometimes the 42 seems faster, other times the 50 seems faster.

Since the Maxwells just came out (relatively speaking, lol) they haven't adapted the cuda for them yet.

So, you are correct. When they finally do adapt the code for Maxwell there will be improvement but until that happens, you are going to see differences from 1 day to the next. Which one is closer? 50 since that generation of GPU is closer to Maxwell.

But in the end it's your choice.


Happy Crunching..

Zalster

edited to correct Never to Neither

Here's the thing. Never Cuda 42 nor cuda 50 were written for the Maxwell chip.

I've seen the same thing as you. Sometimes the 42 seems faster, other times the 50 seems faster.

Since the Maxwells just came out (relatively speaking, lol) they haven't adapted the cuda for them yet.

So, you are correct. When they finally do adapt the code for Maxwell there will be improvement but until that happens, you are going to see differences from 1 day to the next. Which one is closer? 50 since that generation of GPU is closer to Maxwell.

Thanks Zalster, You did give very good explanation to my questions.

Happy Crunching..

Yep, as long as We can get workunits... -> http://setiathome.berkeley.edu/forum_thread.php?id=76106


(setting up barricades against S@H withdrawal) :)

Stick with cuda 50 for now

Yeah, I know and I will install someday cuda50 optimized app.

I did run several stock application tasks (over required 11), cuda50 was fastest. Then I did test optimized cuda50 application, faster than stock.

But now stock-cuda42 seems to be faster? What did happen that servers do think that cuda42 is faster than cuda50? Shorties? Or something else?

Shorties, Running lots of them lowers the APR.

With MBv7 an autocorrelation search was added, this added a fixed amount of processing to all MBv7 Wu's, the autocorrelation search has since been proved to be relatively inefficient,
since Shorties processing is dominated by the autocorrelation search, they do tend to lower the APR below a slower app.
(Shorties went from 2mins on my old GTX460 to almost 7 minutes after the move from MBv6 to MBv7)

Claggy

(Shorties went from 2mins on my old GTX460 to almost 7 minutes after the move from MBv6 to MBv7)

I can't remember the actual numbers, but I found that going from a GTX 560Ti to the GTX 750Ti, the shorties ran much longer. Doing 2 at a time they went from around 10min to 14-16min. However the longer running WUs were a different story- run times only increased a minute or 2. I actually tried running 3 at a time & got even better throughput; however the slow down with the shorties was even more severe and made it not worth doing 3, hence still only doing 2 WUs at a time.
Overall, CUDA50 was the best application to use.

---

Grant
Darwin NT

Stick with cuda 50 for now

Yeah, I know and I will install someday cuda50 optimized app.

I did run several stock application tasks (over required 11), cuda50 was fastest. Then I did test optimized cuda50 application, faster than stock.

But now stock-cuda42 seems to be faster? What did happen that servers do think that cuda42 is faster than cuda50? Shorties? Or something else?

I may have founded answer to my own question.

After all outages when MB units were available, I did run stock and did receive only cuda50 tasks.

Then I did switch to run two MB tasks simultaneously. After day or two (don't remember) I did receive only cuda42's.

So I think that BOINC servers don't know are You running one/two/etc tasks at time and switches to cuda42. In time I think that will level back to cuda50, but now when AP's are running I don't know for sure.

---

"Please keep Your signature under four lines so Internet traffic doesn't go up too much"

- In 1992 when I had my first e-mail address -

Asus GTX750TI-OC-2GD5, core 1188Mhz, memory 2048MB DDR5 @ 1350Mhz, factory defaults.

Computer 6864212

Running now Lunatics v0.43a, APv7 GPU only.

I have tried to figure out what I should put in my ap_cmdline_win_x86_SSE2_OpenCL_NV.txt? Those -unroll etc parameters.

And should I run one or two tasks?

I have done some experiments, but those experiments did gave me problems, my host just freezed. Only hard reboot got it going again.

Maybe some bad parameters or it is just my crappy computer...

---

"Please keep Your signature under four lines so Internet traffic doesn't go up too much"

- In 1992 when I had my first e-mail address -

WezH,

Sorry delay, Holidays and such.

What does your command line look like?

I don't have access to my computer with the 750s right now but I believe my command line looks like this.

-use_sleep -unroll 12 -oclFFT_plan 256 16 256 -ffa_block 8192 -ffa_block_fetch 4096 -tune 1 64 4 1 -tune 2 64 4 1 -hp

I'll check it when I see those computers again on Monday.

I would only run 2 APs at a time on the 750s

If Mike makes any suggestions on different settings then I would follow what he says.

Zalster

Nope, nope and nope. There must be something wrong with my computer, every time I insert command line options my host freezes. Sometimes it takes 15 minutes, sometimes couple hours. This doesn't happen when running Lunatics without command line options.

So far I have tried following command line options:

-use_sleep -unroll 10 -ffa_block 4096 -ffa_block_fetch 2048 -hp
-use_sleep -unroll 15 -ffa_block 16384 -ffa_block_fetch 8192 -hp
-use_sleep -unroll 10 -oclfft_plan 256 16 256 -ffa_block 8192 -ffa_block_fetch 4096 -tune 1 64 4 1 -tune 2 64 4 1
-use_sleep -unroll 10 -ffa_block 16384 -ffa_block_fetch 8192 -hp -use_sleep
-use_sleep -unroll 12 -oclFFT_plan 256 16 256 -ffa_block 8192 -ffa_block_fetch 4096 -tune 1 64 4 1 -tune 2 64 4 1 -hp
-use_sleep -unroll 15 -oclFFT_plan 256 16 256 -ffa_block 16384 -ffa_block_fetch 8192 -tune 1 64 4 1 -tune 2 64 4 1 -hp
-use_sleep -unroll 15 -oclFFT_plan 256 16 256 -ffa_block 16384 -ffa_block_fetch 8192 -tune 1 64 8 1 -tune 2 64 8 1

And every one of them causes my host freeze. And it doesn't matter do I use my host or am I AFK.

---

"Please keep Your signature under four lines so Internet traffic doesn't go up too much"

- In 1992 when I had my first e-mail address -

how many work unites per GPU and what is the ration of CPU to GPU? Are you crunching any on the CPU as well?

how many work unites per GPU and what is the ration of CPU to GPU? Are you crunching any on the CPU as well?

Most of these tests I did run only one wu per GPU, couple times I did two/GPU because command line options were slower than stock Lunatics.

Always did run 1 CPU + 1 GPU (or 1 CPU + 0.5 GPU on those couple times)

Only thing running on my CPU is WUProp and that is non-CPU-intensive task (in 6 hour running period CPU usage is under 10 seconds)

---

"Please keep Your signature under four lines so Internet traffic doesn't go up too much"

- In 1992 when I had my first e-mail address -