When overclocking a GPU one normally can modify three values: core, shader and memory. When altering these values which one will be the most effective to fasten up crunching ?

Btw: the PC, or better, the GPU is only used for crunching, so there's no need for a 'good' picture... ;)

[quote]

You should require both, core and shader clock, whereas memory clock should be pretty irrelevant.

MrS

Wow, this was unexpected! First result is in:

Previously I was at 61.0 - 62.5 ms/step and never below 61 ms with app 6.45. The WU with mem clock increased from 1100 to 1160 MHz needed 59.5 ms/step. 5.5% mem clock increase -> 2.5 - 5.0% performance increase, depending on what the normal time would have been for that WU.

So I take everything back and claim the opposite: mem clock does matter in GPU-Grid.

MrS

When running at stock values (600/1500/1000) I had about 65 ms/step...
Now I'm running at 790/1800/1100 and had 52 ms/step for my last WU... ;)
But I'm still wondering which one has got the highest impact on decreasing the calculation times...

There are CoolBits from Nvidia and NVclock...

If you are lucky and the card and driver support it, somehow...
I tried with nvclock (even compiled the newest one from SVN) and coolbits set in xorg.conf, but both give me a white screen when ever I change a value. Here it does not matter if I lower the clocks or increase them.
I logged in via SSH and nvclock -i reported the core speed somewhere over 900MHz, way too much...

I ended up with overclocking my 8800GTS under windows, modifying and flashing the video BIOS to the new values.

I've got an ASUS 9600GT stock at 650/1625/1800 and am wondering the 'best' way to OC it. Do I keep all the ratio's the same or just the engine/shader ratio?

I've upped it to 700/1800/1900 without seeing a noticeable difference and now have it set to 725/1815/1944.

Live long and BOINC!

ps. If I had of known that an 8800 was 'faster' than a 9600 then I would have bought an 8800! Gotta love marketing! Bet then I was a real nOOb, now I'm just a noob.

Not a GTX280, but a GTX260 (if that helps)

I figured something out:

Overclocking Core&shaders ~15% did have almost a linear impact on WU times.
Overclocking the memory 10% did get me a performance gain ~1-2%.

Stock Clockings:
Core 576 - Shader 1242 - Memory 999 : Time per step: 41,7
Overclocking Core & Shaders:
Core 650 - Shader 1401 - Memory 999 : Time per step: 37,5
Overclocking Memory (and a little bit more on Core & shaders again)
Core 661 - Shader 1425 - Memory 1101 : Time per step: 37,3

It seems to me, that the last gain of 0.2ms is only because of the minimal raised Core & Shaderspeed and not because of the 10% higher memoryclock


Cu KyleFL

I've got an ASUS 9600GT stock at 650/1625/1800 and am wondering the 'best' way to OC it. Do I keep all the ratio's the same or just the engine/shader ratio?

I've upped it to 700/1800/1900 without seeing a noticeable difference and now have it set to 725/1815/1944.

My suggestion: find the maximum stable clock for engine and shader first, then for the memory and afterwards back off a bit for safety. And don't care about the ratios at all.

MrS

I have it (GTX280) overclocked to:

clock: 697, shaders: 1500, mem: stock

In my last three units I got these times per step:

29.688 ms
29.096 ms
29.391 ms

This OC is not the maximum possible. Simply I tried to put shaders to 1500 (linked clock) and I found it works fine. I have not tried to go beyond it. I have never made an error WU with these settings as you can see in my account.

My suggestion: find the maximum stable clock for engine and shader first, then for the memory and afterwards back off a bit for safety. And don't care about the ratios at all.
MrS

Agreed, that's pretty much how I approach it....

Maybe I should run that test utility that find out what the card can do ... :)

THen again ... with my luck ... maybe I shouldn't ....

I have an XFX GTX280 XXX with all the following stock settings:

clock: 670, shader: 1458, memory: 2500

In my last three WU's I got:

27.644 ms,
26.656 ms, and
26.796 ms.

I believe that most GTX280's could attain these speeds without great difficulty. Just keep them cool. Mine run between 79-82°C.

Pat

I have an XFX GTX280 XXX with all the following stock settings:

clock: 670, shader: 1458, memory: 2500

In my last three WU's I got:

27.644 ms,
26.656 ms, and
26.796 ms.

I believe that most GTX280's could attain these speeds without great difficulty. Just keep them cool. Mine run between 79-82°C.

Pat

Sorry guys, dumb question..
Where you getting these numbers for ms/step? Wondering what my cards doing..Thanks!

Also, the overclocking..this is pretty much my thoughts.
number of shaders X core clock = amount of work done

So I always OC to highest stable core and the rest no matter really.
Does anyone know...My 260 won't OC. Is there a proggy besides the Nvidia stuffs? I tried Ntune and the new one but no have the perfomance tab on the 260 machine!?! It is an msi card btw.

My 9500 GT overclocks like a champ and is maybe the best rac/$ of my current setups. ..checking...Gpu-z gives it's current clocks at 700/570/1750 stock is 550/500/1375. It is made by EV3A. The 9800GTX runs 800/1115/1984.

Mahalos


____________

Sorry guys, dumb question..
Where you getting these numbers for ms/step? Wondering what my cards doing..Thanks!

My 280 is running 27 - 32 ms per step over a bunch of tasks ... the 9800 GT is 91 - 92 ms ...

My 280 is running 27 - 32 ms per step over a bunch of tasks ... the 9800 GT is 91 - 92 ms ...

The older posts in this thread from October were during a period where only one kind of workunit was being examined. Now we have three different kinds (with credits of 32xx, 24xx, and 18xx) that have widely varying ms per step on the same card. It would be interesting to see how cards are doing with these different types of work and how much spread there really is. I suspect that the spread is fairly tight on high-end cards with the differences becoming more substantial as one moves to the low-end. Not sure if different aspects of different OC would have different performance effects on different types of work? and yes, I could not fit another "different" into that last sentence ;)

The older posts in this thread from October were during a period where only one kind of workunit was being examined. Now we have three different kinds (with credits of 32xx, 24xx, and 18xx) that have widely varying ms per step on the same card. It would be interesting to see how cards are doing with these different types of work and how much spread there really is.

• 203862 > 3232 > 30.451 ms
  
• 202155 > 3232 > 30.363 ms
  
• 199331 > 3232 > 30.188 ms
  
• 198010 > 3232 > 43.650 ms (MDIO ERROR)
  
• 201276 > 2933 > 57.040 ms (MDIO ERROR)
  
• 200798 > 2435 > 36.197 ms
  
• 200793 > 2435 > 36.363 ms
  
• 196008 > 2435 > 42.375 ms (MDIO ERROR)
  
• 202088 > 1888 > 38.864 ms
  
• 197912 > 1888 > 38.721 ms
  
• 196876 > 1888 > 39.508 ms
  

GTX 260/216 (666/1500/1150) (TaskID > credits > ms/step)

• 202904 > 3232 > 27.505 ms (MDIO ERROR)
  
• 202898 > 3232 > 27.503 ms (MDIO ERROR)
  
• 182989 > 3232 > 27.513 ms (MDIO ERROR)
  
• 185601 > 2435 > 33.772 ms (MDIO ERROR)
  
• 185199 > 2435 > 33.813 ms (MDIO ERROR)
  
• 184240 > 2435 > 34.298 ms (MDIO ERROR)
  
• 186263 > 1888 > 35.726 ms (MDIO ERROR)
  
• 185932 > 1888 > 35.593 ms (MDIO ERROR)
  
• 185160 > 1888 > 37.406 ms (MDIO ERROR)
  

Are those MDIO errors signs of serious problems ? Do not hope with OC-settings.

Regarding and comparing the values from the oced GTX 280 I can assume there is a measurable effect in the difference of the shaders 192-216-240, which can be reduced by high overclocking, rudimental at least.

Okay...On the low end I have numbers from a 9500GT with 512mb(700 core, 1750 shader, 2000 memory) and a mid-range 9600 GSO with 384mb(600, 1700, 1800). Using 6.5.0 for all unless noted otherwise.

9500GT

3232> 210ms
2435> 272ms
1888> 293ms


9600GSO

3232> 85ms to 89ms
2435> 111ms to 116ms
1888> 102ms to 115ms

Odd thing with the 9600GSO 1888 credit units is that the range is deceiving since there were really no middle units--couple at 102ms with the rest around 115ms.

It does appear that the spread does magnify as one moves from faster to slower cards.


*Note that 1)The 9500GT numbers are from single workunits for each credit level 2)The 9500GT shader clock increased from 1750 to 1800 for 3232 credit unit 3)The 9600GSO 1888 credit units all with BOINC 6.3.21

Are those MDIO errors signs of serious problems ? Do not hope with OC-settings.

Nope. They just tell you there was no file to resume computation from, because you didn't stop & restart BOINC during the WU.

9600GSO

3232> 85ms to 89ms
2435> 111ms to 116ms
1888> 102ms to 115ms

I was having the same problem and I found an app put out by EVGA simply called

EVGA_Precision_1.40.exe that I had read worked on almost any brand of card.

I have a BFG card and it seems to be working for me, by that I mean, I just started using it today and don't understand the effect the various changes mean as far as decreasing my run times. GPU over-clocking is new territory for me, CPU'S yes GPU'S no. It seems really simple to use, it simply has four sliders, which are Core Clock, Shader Clock, Memory Clock, and Fan Speed which I was glad to see as my fan was stuck on Auto and never moved, it was always 40% no matter what the conditions were, now I can increase it as needed. The danger posed by overheating when overclocking CPU's I am very familiar with and I assume the same holds for GPU's and I can't afford to risk hurting my GTX 260.

Hope this helps you

I was having the same problem and I found an app put out by EVGA simply called

EVGA_Precision_1.40.exe that I had read worked on almost any brand of card.

I have a BFG card and it seems to be working for me, by that I mean, I just started using it today and don't understand the effect the various changes mean as far as decreasing my run times. GPU over-clocking is new territory for me, CPU'S yes GPU'S no. It seems really simple to use, it simply has four sliders, which are Core Clock, Shader Clock, Memory Clock, and Fan Speed which I was glad to see as my fan was stuck on Auto and never moved, it was always 40% no matter what the conditions were, now I can increase it as needed. The danger posed by overheating when overclocking CPU's I am very familiar with and I assume the same holds for GPU's and I can't afford to risk hurting my GTX 260.

Hope this helps you

I didn't know about the program only controlling one core of a GTX 295 because I have yet to work out the nerve to buy one. I am looking to build a new system but am going nuts trying to figure out what hardware to purchase. It's main purpose will be crunching Boinc App's and a lot of video conversions, both of which can benefit from a great GPU, but I am going to be putting together a whole system, case, power supply, video card, motherboard, ram, processor, the list seems endless. I am hoping that during the time I spend trying to get the nerve up to lay out what for me will be a small fortune, the prices on video cards at least will stabilize. I saw a BFG GTX 280OC for $365 CAN today and the 295 is around $660 but I think the 295's and now the 285's will push the 280's price down even further very soon. The problem is I don't game at all, it is something I have never been interested in and while this will not only date me it's also a little embarrassing to say the last game I played was something like frogger or Pac-Man back in University. I was even considering getting dual 295's as I can probably scrap up the cash but can I justify that kind of outlay just to increase my RAC and hopefully do some worthwhile science in the process.

The problem is I don't game at all, it is something I have never been interested in and while this will not only date me it's also a little embarrassing to say the last game I played was something like frogger or Pac-Man back in University.

I was even considering getting dual 295's as I can probably scrap up the cash but can I justify that kind of outlay just to increase my RAC and hopefully do some worthwhile science in the process.

If this is going to be the rig for a few years ... get the best MB with 3 PCIe slots. Then the best and fastest CPU and memory you can afford ... if needs be, skimp on the memory as that can be replaced more cheaply than the CPU ... the high quality MB for the same reason.

Get the best GPU you can afford with what is left. YOu have expansion slots and can slowly add GPUs as money comes in later on ... so, for a start lets say you get a 9800GT for $100 ... does about 5K per day ... a year from now you get a 295 ... and now you are well above 15K a day add another 295 ... then replace the 9800 GT with a third ... by then may be you will be in the market for a new PC ... but now you have a small farm of GPUs to start ...

Rinse ... repeat ...

h

hello everybody,i would want your advices on my overclocking,it is good?

geforce gtx 260 216sp 55nm (original clock (576/1242/999)

now overclocked at: (gpu:756/shader:1512/memory:1188)
temp idle:55°c
temp full load with furemark in extreme mode:81°c


Task ID 282813
Name hG13938-SH2_US_5-3-10-SH2_US_51910000_1
Workunit 210038
Created 6 Feb 2009 0:48:21 UTC
Sent 6 Feb 2009 0:49:32 UTC
Received 6 Feb 2009 7:39:45 UTC
Server state Over
Outcome Success
Client state Done
Exit status 0 (0x0)
Computer ID 22576
Report deadline 10 Feb 2009 0:49:32 UTC
CPU time 2416.602
stderr out

<core_client_version>6.4.5</core_client_version>
<![CDATA[
<stderr_txt>
# Using CUDA device 0
# Device 0: "GeForce GTX 260"
# Clock rate: 1512000 kilohertz
# Number of multiprocessors: 27
# Number of cores: 216
MDIO ERROR: cannot open file "restart.coor"
# Time per step: 31.006 ms
# Approximate elapsed time for entire WU: 15503.221 s
called boinc_finish

</stderr_txt>
]]>

Validate state Valid
Claimed credit 2478.98611111111
Granted credit 2478.98611111111
application version 6.59
***********************************
Task ID 281496
Name Fm24458-GRA1-4-5-acemd_0
Workunit 210000
Created 5 Feb 2009 14:10:57 UTC
Sent 5 Feb 2009 19:55:55 UTC
Received 6 Feb 2009 7:23:43 UTC
Server state Over
Outcome Success
Client state Done
Exit status 0 (0x0)
Computer ID 22576
Report deadline 9 Feb 2009 19:55:55 UTC
CPU time 2636.021
stderr out

<core_client_version>6.4.5</core_client_version>
<![CDATA[
<stderr_txt>
# Using CUDA device 0
# Device 0: "GeForce GTX 260"
# Clock rate: 1512000 kilohertz
# Number of multiprocessors: 27
# Number of cores: 216
MDIO ERROR: cannot open file "restart.coor"
# Using CUDA device 0
# Device 0: "GeForce GTX 260"
# Clock rate: 1512000 kilohertz
# Number of multiprocessors: 27
# Number of cores: 216
# Using CUDA device 0
# Device 0: "GeForce GTX 260"
# Clock rate: 1512000 kilohertz
# Number of multiprocessors: 27
# Number of cores: 216
# Time per step: 24.897 ms
# Approximate elapsed time for entire WU: 21162.294 s
called boinc_finish

</stderr_txt>
]]>

Validate state Valid
Claimed credit 3215.36111111111
Granted credit 3215.36111111111
application version 6.59

it is good job for a gtx260?
____________

How did you test stability?

So the answer is a big fat "yes"!

hello all,

Ive tried o/c-ing my 9800gtx card. got it up about 15%, and its stable everywhere else but gpugrid is not liking it very much. temps at full load are around 60deg.
I can only complete maybe 1 out of 5 wus issued by gpugrid. Any idea why this might be the case? as i said the card is perfectly stable in other apps.

Cheers

Had the same experience like you: my card errored out each & every WU on GG and when I switched to SETI it ran without any probs...
If you want to continue to crunch GPUGrid-WUs you have to lower your OCing until you get no more computation errors...

Had the same experience like you: my card errored out each & every WU on GG and when I switched to SETI it ran without any probs...
If you want to continue to crunch GPUGrid-WUs you have to lower your OCing until you get no more computation errors...

Ive ran benchmarks on 3d mark, stressed on sisoft sandra, several games, and seti wus.

they are all fine.

also like i said 1 out of every 5 gg wus will calculate fine, the rest error out.

strange

Ive ran benchmarks on 3d mark, stressed on sisoft sandra, several games, and seti wus.

they are all fine.

also like i said 1 out of every 5 gg wus will calculate fine, the rest error out.

strange

Had the same experience like you: my card errored out each & every WU on GG and when I switched to SETI it ran without any probs...
If you want to continue to crunch GPUGrid-WUs you have to lower your OCing until you get no more computation errors...

I believe something has changed in the Application's myself. I have a 8800GT OC that was running the WU's just fine then a few days ago started erring out every WU after 3 Seconds. Like other people I can run the SETI GPU WU's just fine though.

I've even under clocked the 8800GT OC & it still errors out the WU's, so lowering your clock speed isn't necessarily a cure for the errors, at least it wasn't in my case.

The 8800GT OC runs perfectly fine though for everything else @ it's Stock or the Overclocked Speed it came with except for running the GPUGrid WU's so like I said I think it's something in the Application's that's changed ... ???

Had the same experience like you: my card errored out each & every WU on GG and when I switched to SETI it ran without any probs...
If you want to continue to crunch GPUGrid-WUs you have to lower your OCing until you get no more computation errors...

I believe something has changed in the Application's myself. I have a 8800GT OC that was running the WU's just fine then a few days ago started erring out every WU after 3 Seconds. Like other people I can run the SETI GPU WU's just fine though.

I've even under clocked the 8800GT OC & it still errors out the WU's, so lowering your clock speed isn't necessarily a cure for the errors, at least it wasn't in my case.

The 8800GT OC runs perfectly fine though for everything else @ it's Stock or the Overclocked Speed it came with except for running the GPUGrid WU's so like I said I think it's something in the Application's that's changed ... ???

I must say that I am seeing the same basic effect with my Asus 8800GT. I run 3x 8800GS and 1x 8800GT. The GS's are running fine as usual. The GT collected dust for several months, but is now back up and running--and kicking out 1 compute error after another. This particular card has always been the biggest pita out of the 4 (even on F@H). I just lowered my OC a bit once again, but I expect it to start throwing errors again...maybe not though. I thought it was the card and/or OC....however, after reading this: I'm wondering if there is an 8800GT issue. It's a G92 card.....

Some programs are more sensitive to errors than others. Aside from occasional errors I have been running for two months with very few errors. The system may pass all tests, yet not run a specific application. If you bring down the OC and the application begins to run without errors, then regardless of other tests the OC was at fault.

Another cause can be the amount of memory on the card.

A third cause can be other events on the system that the system as a whole does not react well to and causes tasks to error out...

There is also a slight amount of evidence that some of the tasks issued here may not run well or at all on the lower end cards.

At SaH one of the error modes is that once tasks fail, all will fail until the system is rebooted. There are dozens of reasons that this can happen with the simplest being that the API does not properly initialize the GPU after certain errors so that the error on one task will contaminate all other tasks started.

I run 5 8800gt's (all o/c'ed) from three differnt manufactures and have the extremeely rare work unit have a problem...I am still running 6.5.0 on all my machines so maybe the problem is with the 6.6.xx application

If you bring down the OC and the application begins to run without errors, then regardless of other tests the OC was at fault.

Another cause can be the amount of memory on the card

A third cause can be other events on the system that the system as a whole does not react well to and causes tasks to error out

At SaH one of the error modes is that once tasks fail, all will fail until the system is rebooted. There are dozens of reasons that this can happen with the simplest being that the API does not properly initialize the GPU after certain errors so that the error on one task will contaminate all other tasks started

I must say that I am seeing the same basic effect with my Asus 8800GT. I run 3x 8800GS and 1x 8800GT. The GS's are running fine as usual. The GT collected dust for several months, but is now back up and running--and kicking out 1 compute error after another. This particular card has always been the biggest pita out of the 4 (even on F@H). I just lowered my OC a bit once again, but I expect it to start throwing errors again...maybe not though. I thought it was the card and/or OC....however, after reading this: I'm wondering if there is an 8800GT issue. It's a G92 card.....

Just trying to brainstorm ...

I hate it when we cannot find reasons for things ... like why one of my computers all of a sudden seems to want to run the hard drive all the time to the point where nothing gets done ...

Indexing is off, computer is only used for BOINC ... no AV installed ...

ATItool is a small free app that's very useful for testing video overclocks. It contains a "scan for artifacts" function that can catch things the naked eye cannot. My usual arsenal of apps for testing the stability of my system is:

Prime95 - 64 bit, multi core version. Run on all cores using Small FFTs, Round Off Checking enabled
FurMark - Stability Test, Xtreme Burning Mode, non-fullscreen, set resolution to match desktop resolution and minimize to get it out of the way
ATItool - set to "scan for artifacts"
SpeedFan - temps calibrated for accuracy

Prime + Fur will put more heat and load into your system than almost anything else. The Round Off Checking in Prime95 will check for CPU computation errors, the Scan for Artifacts tool in ATItool will check for GPU computation errors. Let it run for a few hours. If no errors present, your overclock isn't causing issues. If you also have a system memory/FSB overclock going, run Prime95 on "Blend" to test for memory errors. Typically, I'll find the max stable overclock, then reduce clocks one step down from that level to ensure stability.

Happy overclocking!

Hmm lol i am not really a hero in this area called oc.
But i finally decided to oc my EVGA 9600 GT
And found it running nice at 725/1800/1100 real(2200).
Now lets wait and see if it really gains me some speed ;)
Untill now most units did :
2478 / 139-143 ms
3718 / 117-131 ms
3848 / 187-206 ms

UPDATE, just in case anybody was using the stability test I encouraged earlier in this thread. CPU testing with Prime95 small FFTs is NOT sufficient unless you let it run for around 40 hours.

Instead, I'm now using a program called "Intel Burn Test". It'll tell you if you're CPU is operating within spec in 30 minutes.