GPU OCing Help Thread

[INFRNL]

I'm at 1512 right now and it's lookin like I can pull 27k, but that might be a little optimistic. Only 2 days here, so we'll have to see. Probably won't clock it up any more till after spring break (two weeks).

I'm experimenting on a few hunches here... If I can get my numbers substantially higher than normal then I did something right.

So I am not much further than you. I need to remember which card is my GTX 260. But I think your right its not much further off my 275's. I have 4 GTX 275's still and 2 GTX 260; so if I gain 1k ppd more on the 275's thats 4k more a day. When I had 10 275's it was more noticeable. every little bit adds up. main reason why I went with the 275's.

Either way i think now the 260's are probably the best bang for the buck right now.

[Mabyboi]

So I am not much further than you. I need to remember which card is my GTX 260. But I think your right its not much further off my 275's. I have 4 GTX 275's still and 2 GTX 260; so if I gain 1k ppd more on the 275's thats 4k more a day. When I had 10 275's it was more noticeable. every little bit adds up. main reason why I went with the 275's.

Either way i think now the 260's are probably the best bang for the buck right now.

wait... 10x 275's?!

[mreuter80]

Either way i think now the 260's are probably the best bang for the buck right now.

unless you get a used 275 for the same price as a new 260...

OK, OK ... one can also get used 260s for less ...

[PoppaGeek]

wait... 10x 275's?!

Yep, he is like that.

[Mabyboi]

Okay, i need a bit of help now..

Before i was running at 701/1577/1233, and GPUgrid was stable and completing WU's in about 22-24,000 seconds..

after overclocking to 727/1623/1308 im furmark stable, and my first 2 WU's completed and validated.. but took 50,000 seconds to do so... double the amount of time...

why would this be?!

one WU just error'd at 7 hours

[Otis11]

Already said this, but we'll see what others think.

I'm guessing mem OC is too high and failing. Shaders or core fail you get an error... if mem can't tell what value it's supposed to be, it'll make a call back to main memory (ideally, depending on how the drivers are written). That's the only thing I can think of to get it to validate but still be crazy slow...


Hope it works out!

[Mabyboi]

Already said this, but we'll see what others think.

I'm guessing mem OC is too high and failing. Shaders or core fail you get an error... if mem can't tell what value it's supposed to be, it'll make a call back to main memory (ideally, depending on how the drivers are written). That's the only thing I can think of to get it to validate but still be crazy slow...


Hope it works out!

If the clock was too high, wouldnt i have had errors in furmark? like im video gaming stable.. its the GPUgrid that isnt stable... as you said earlier, is my memory clock too high?

dropped memory clock to 1243 lets see if it changes.

[Otis11]

If the clock was too high, wouldnt i have had errors in furmark? like im video gaming stable.. its the GPUgrid that isnt stable... as you said earlier, is my memory clock too high?

dropped memory clock to 1243 lets see if it changes.

wow 1,111 posts at 11:11 (my time anyway)...

I think you should try clocking that card to 902/1940/1526... Just saying, might be lucky...

[Snow Crash]

The thing about being stable, especially with 295, is that you can go just far enough that the driver will reset itself without crashing (erroring) the WU. The big problem when that happens is that it resets itself into 2D mode ... very slow. Pop open Precision and see what it says what your shaders are really doing. I change the Precision graphs to show me gpu temps for both cards and shaders for both cards because on a 295 the downcloack can happen on one card but not the other .... you really need to check both. Don't use GPUZ if you suspect you may have downclocked because the way it interogates your GPU forces it to step back up to high performance 3D. While that might seem like it gets you back to where you need to be it is only temporary and your GPU will downclock again shortly. The only way I have found out of the downclock due to a driver reset is to first suspend BOINC, close BOINC, reboot the machine, change my OC (remember, that's how this all started to begin with) and then let BOINC get back to work.

The funny thing abut this post is I have been stepping up my memory freq today and was getting impatient and then I remembered Mabyboi had high clocks there and was claiming stable so I was just stopping by real quick to see what your actual numbers were. Think I might keep stepping up slowly

[PoppaGeek]

wow 1,111 posts at 11:11 (my time anyway)...

I think you should try clocking that card to 902/1940/1526... Just saying, might be lucky...

He's the...the....the ONE!

[Mabyboi]

The thing about being stable, especially with 295, is that you can go just far enough that the driver will reset itself without crashing (erroring) the WU. The big problem when that happens is that it resets itself into 2D mode ... very slow. Pop open Precision and see what it says what your shaders are really doing. I change the Precision graphs to show me gpu temps for both cards and shaders for both cards because on a 295 the downcloack can happen on one card but not the other .... you really need to check both. Don't use GPUZ if you suspect you may have downclocked because the way it interogates your GPU forces it to step back up to high performance 3D. While that might seem like it gets you back to where you need to be it is only temporary and your GPU will downclock again shortly. The only way I have found out of the downclock due to a driver reset is to first suspend BOINC, close BOINC, reboot the machine, change my OC (remember, that's how this all started to begin with) and then let BOINC get back to work.

The funny thing abut this post is I have been stepping up my memory freq today and was getting impatient and then I remembered Mabyboi had high clocks there and was claiming stable so I was just stopping by real quick to see what your actual numbers were. Think I might keep stepping up slowly

You sir are correct.

GPU1 Core Clock: 400mhz
GPU2 Core Clock: 400mhz
GPU1 Shader Clock: 799mhz
GPU2 Shader Clock: 799mhz
GPU1 Memory Clock: 297mhz
GPU2 Memory Clock: 297mhz

How do i fix it..?

Suspend BOINC -> Restart Computer....?

[Snow Crash]

You sir are correct.

How do i fix it..?

Suspend BOINC -> Restart Computer....?

yuppers ... and then turn down the clock you think is hurting you before restarting BOINC.

I tend to work with one frequency at a time and because I am not gaming I turn core down one notch, up memory 10% and focus on shaders. Once I know where my shaders max out I start working on memory. One at a time, otherwise you just end up guessing.

Have you upped the volts?

[CaptMorgan]

You sir are correct.

How do i fix it..?

Suspend BOINC ->Lower Shaders-> Restart Computer....

Edit- Sorry Snow Crash is on the case

[Mabyboi]

yuppers ... and then turn down the clock you think is hurting you before restarting BOINC.

I tend to work with one frequency at a time and because I am not gaming I turn core down one notch, up memory 10% and focus on shaders. Once I know where my shaders max out I start working on memory. One at a time, otherwise you just end up guessing.

Have you upped the volts?

I think its memory... personally... but ill bring em back down to my stable frequencies before.. and then up my shaders a little, how much do you bump them up at at a time?

Set it back to 701/1600/1233
but when i look at my evga manager, it says: 702/1584/1242...

but putting it to 701/1611/1233 makes it say: 702/1620/1242...

thanks guys!

[Snow Crash]

Take a look at the first post ... shaders move in "straps".
How sure are you that it is memory? I can't do more than 1512 shaders without bumping volts. Once I do bump to 1188 mv I max out at 1656.
Are you using Voltage Tuner (EVGA) to push it?
If I had to OC another 295 I would start by turning core down to 540, turn shaders to 1512 and leave memory at stock 999.
Then after I was sure I am stable (I let it run for at least one complete WU, preferably a GINANNI_BIND or TONICAP_BIND as they stress the shaders to 93%), then I let it run for the rest of the day until I can be around to keep an eye on things. Next I bump to the next "strap" and let it run some more ... rinse repeat. Slow but steady ... I hate when I lose a whole day of crunching because I was at work ands it hard crashed.

[HuffPCair]

Is that F@H or GPUGrid? their temps are different so I have to ask.

Thanks for the stats!

I do Folding@home

[Mabyboi]

Take a look at the first post ... shaders move in "straps".
How sure are you that it is memory? I can't do more than 1512 shaders without bumping volts. Once I do bump to 1188 mv I max out at 1656.
Are you using Voltage Tuner (EVGA) to push it?
If I had to OC another 295 I would start by turning core down to 540, turn shaders to 1512 and leave memory at stock 999.
Then after I was sure I am stable (I let it run for at least one complete WU, preferably a GINANNI_BIND or TONICAP_BIND as they stress the shaders to 93%), then I let it run for the rest of the day until I can be around to keep an eye on things. Next I bump to the next "strap" and let it run some more ... rinse repeat. Slow but steady ... I hate when I lose a whole day of crunching because I was at work ands it hard crashed.

Well my last stable oc was at 701/1577/1233 and all my WU's were perfectly fine, after 18 of them... so i felt like bumping it up more, and as Otis said, if it was the core or shaders the WU would crash immediately, but if it was the memory it could mess it up and push it farther.. so regardless i've dropped down to 702/1620/1242 and well see if this is stable for a while...

but now it seems that the WU's are taking 8-10 hours, rather than the 6 they used to... but even looking at EVGA precision, it says 702/1620/1242...

the only thing that looks odd is the GPU1/2 memory usage... they're both clocked at 1242 but their usage is at 621...?

[Otis11]

^^ Now on that mem thing I could be completely off base... that was a complete guess just FYI. There is reasoning behind it, but as I don't know how these drivers are programmed, that reasoning could be crap...

And I go by SC process of single bumps (after an initial jump)... but I do shaders, mem, and core, all one bump at the same time. If anything fails you can take it all back one bump and proceed forward on each one individually if you wish...

[Mabyboi]

^^ Now on that mem thing I could be completely off base... that was a complete guess just FYI. There is reasoning behind it, but as I don't know how these drivers are programmed, that reasoning could be crap...

And I go by SC process of single bumps (after an initial jump)... but I do shaders, mem, and core, all one bump at the same time. If anything fails you can take it all back one bump and proceed forward on each one individually if you wish...

im holding you to it otis11!

[Otis11]

If you're willing to stick with it, we'll get your 295 to the highest possible output... but it takes work.

[Mabyboi]

If you're willing to stick with it, we'll get your 295 to the highest possible output... but it takes work.

thats what i want. i will overtake you otis. i promise you i will !

[Snow Crash]

snip
the only thing that looks odd is the GPU1/2 memory usage... they're both clocked at 1242 but their usage is at 621...?

When you look at Memory Usage in Precision it is showing you how much memory is being used not the speed at which it is being used. For freq check the Memory Clock.

If you are not familiar with Precision you can double click the graphs themselves and they will pop out into a box all by themselves which you can change the size of :-)

I just took a look and your times are moving in the right direction again

[Mabyboi]

When you look at Memory Usage in Precision it is showing you how much memory is being used not the speed at which it is being used. For freq check the Memory Clock.

If you are not familiar with Precision you can double click the graphs themselves and they will pop out into a box all by themselves which you can change the size of :-)

I just took a look and your times are moving in the right direction again

theyre coming back down to about 5.5-6 hours.. had 2 errors recently though... well see

[Snow Crash]

The only change I made yesterday was memory and I had some errors also).
I noticed something else ... the WUs both you and I were erroring on were IBUCH ... we both complete the TONIs fine.
Perhaps IBUCH is more demanding than TONI, maybe it is just memory?
My 285 is processing IBUCH just fine so I think we are running a little too high on our 295s.
I turned memory down a hair (1152) and we'll see how today goes ... so far no errors since early this morning.

<edit>I posted over at GPUGrid to see if the failuire rate fro IBUCH Pyxxxx type Wus has been higher recently ... trying to see if it might be them and not our OC.</edit>

[CaptMorgan]

Thank you Snow Crash- and everyone else for risking there stability for the greater good (of a stable OC)