5870 Users! Clock Speed, Memory, Voltage, SDK and CCC Version, Flags, etc.

[JJG]

very strange 

Not really, that's just how hardware works. Each chip will be stable at a different speed/voltage as it comes off of the line. The stock clocks are set such that 100% of the chips that are actually used in production cards will be stable in almost all cases at those speeds. Some have more headroom than others.

The hardcore overclockers (the type who compete in this stuff) will buy numerous chips/cards and test each one individually to find the fastest unit. Variations can be huge.

[1714166237]

1714166237

[Dyaheon]

Anyone else have a Gigabyte GV-R587UD-1GD?  I've only managed to get keep it stable at 950mhz clock, and it appears that other 5870's have been able to go considerably higher.  I think it could go higher if I could increase the voltage a bit but I haven't found a software yet that will do it and I don't want to tinker with the BIOS unless someone else has had success.  Just wondering if anyone has gone higher on that card and what suggestions/settings they might have.

I have 3x Gigabyte 5870 OC models (870MHz default core) that look like yours, and they are stable at 975 except one of them needs a little lower (5-10MHz) to be stable. Though I guess it's possible that these OC models have increased voltage from defaults, 1.163V according to GPU-Z.

I believe you can overvolt these with gigabyte software, which you can probably find from the driver CD. Haven't tried it myself and probably won't, as I doubt it would be worth it. Most 5870s really don't go much higher than that, 1GHz+ is very rare afaik and most will be 950-1000MHz at manageable voltage.

Mine do 433MH/s in Ubuntu, 975/300 clocks, phoenix w/ poclbm kernel, aggression 12 worksize 256 fastloop=false.
427 in Windows at same settings, except phatk kernel as it seems to be faster in Windows.

[zapeta]

Anyone else have a Gigabyte GV-R587UD-1GD?  I've only managed to get keep it stable at 950mhz clock, and it appears that other 5870's have been able to go considerably higher.  I think it could go higher if I could increase the voltage a bit but I haven't found a software yet that will do it and I don't want to tinker with the BIOS unless someone else has had success.  Just wondering if anyone has gone higher on that card and what suggestions/settings they might have.

I have 3x Gigabyte 5870 OC models (870MHz default core) that look like yours, and they are stable at 975 except one of them needs a little lower (5-10MHz) to be stable. Though I guess it's possible that these OC models have increased voltage from defaults, 1.163V according to GPU-Z.

I believe you can overvolt these with gigabyte software, which you can probably find from the driver CD. Haven't tried it myself and probably won't, as I doubt it would be worth it. Most 5870s really don't go much higher than that, 1GHz+ is very rare afaik and most will be 950-1000MHz at manageable voltage.

Mine do 433MH/s in Ubuntu, 975/300 clocks, phoenix w/ poclbm kernel, aggression 12 worksize 256 fastloop=false.
427 in Windows at same settings, except phatk kernel as it seems to be faster in Windows.

Thanks for the info.  The stock voltage on mine is also 1.163 but it is stable at 950 and unstable at 955.  I wasn't sure if the gigabyte software would adjust voltages but based on your results it isn't worth my trouble.  I run mine aggression 9, worksize 128, and phatk asince I do use the card for video output and need it to be somewhat responsive and I get about 408MH/s.   

[icaci]

I have 2x Sapphire Vapor-X 5870 OC running at stock voltage (1.16 V if I remember correctly), 935 MHz core and 319 MHz mem clocks. No CrossFireX bridges. Windows 7 32-bit, Catalyst 11.5 w/ bundled SDK 2.4. Running Phoenix 1.48 + phatk. Options are: VECTORS BFI_INT WORKSIZE=256 FASTLOOP=false AGGRESSION=13. Getting 413 Mhash/s from each card. The rig is located in a conditioned room alongside other IT equipment. With both fans at 100% and with case's side covers removed core temperatures are 45C and 75C (the second card sits too close to the first because of bad PCI-E slots separation).

[Seraphim401]

I edited the file in question,but still can't tweak the voltage.

[tiberiandusk]

I'm getting 435 with mine with core at 990 and mem at 185. Stock voltage and phoenix with phatk. Sitting around 70c with fan at 85%.

[The Koolio]

getting a nice 440-460 Hash with my 5870, clocked to 1Ghz and 185 mem clock using AMD clock tool. Using msi afterbruner to run gpu fans at 100% after 50c, currently stable at 74c, been running for 5 days now, averaging 1BTC-1.1BTC per day. Whenever i try and take my mem clock to 200-250 or below 180, as well as taking gpu to over 1Ghz i get heavy artifacting, I end up having to go into safe mode, use driver sweeper then reinstall CCC 11.5. Also running SDK 2.4

My flags are:

phoenix -u http://username_1:pass@btcguild.com:8332/ -k phatk VECTORS BFI_INT FASTLOOP=false AGGRESSION=13 DEVICE=0 WORKSIZE=128

BTW

Ive ran deepbit, slush and btcguild each for at least 24 hours, in my honest opinion BTCguild is bringing some nice speeds, although i do get periods where theres no work, but this last for a max 4 seconds, hardly any refused shares and take into account the 0% fee!

Ive also ran Phoenix and POCLBM, again, my best speeds were gained using phoenix and the phatik flag.

Now planning to buy either another 5870, or a 5870x2 (drooling at the 850 hash rate!)

[tiberiandusk]

getting a nice 440-460 Hash with my 5870, clocked to 1Ghz and 185 mem clock using AMD clock tool. Using msi afterbruner to run gpu fans at 100% after 50c, currently stable at 74c, been running for 5 days now, averaging 1BTC-1.1BTC per day. Whenever i try and take my mem clock to 200-250 or below 180, as well as taking gpu to over 1Ghz i get heavy artifacting, I end up having to go into safe mode, use driver sweeper then reinstall CCC 11.5. Also running SDK 2.4

My flags are:

phoenix -u http://username_1:pass@btcguild.com:8332/ -k phatk VECTORS BFI_INT FASTLOOP=false AGGRESSION=13 DEVICE=0 WORKSIZE=128

BTW

Ive ran deepbit, slush and btcguild each for at least 24 hours, in my honest opinion BTCguild is bringing some nice speeds, although i do get periods where theres no work, but this last for a max 4 seconds, hardly any refused shares and take into account the 0% fee!

Ive also ran Phoenix and POCLBM, again, my best speeds were gained using phoenix and the phatik flag.

Now planning to buy either another 5870, or a 5870x2 (drooling at the 850 hash rate!)

Good luck finding them. Every site I have checked is sold out and there are no stores like Fry's anywhere near where I live.

[The Koolio]

ive found someone who has a faulty 5870x2, judging from what he diagnosed it with i reckon 10 mins in the oven should bring it bk to life

[50Million]

Update:
Removed the card, cleaned the TIM off the chip (looked like someone was trying to fill a lake with the stuff) and put some arctic s 5 on there. Re mounted the cooler, tightened up the back holder plate thing, and cranked it up.
Started off pulling 380M at default clock speeds for the card (900 and 1200), and was fine at this for about 3 mins, but the temp was slowly going up and up... When it hit 90, the card then dropped the clocks (mem and main), as if throttled. But, in saying that, it now pulls 320 instead... so slight improvement, but still not right.

[PaZer]

Hi,

I have 3 XFX HD5870 Cards. My max MHash/s is approx. 380. I cannot go over 400 :-(

I'm using poclbm (according to this instruction http://forum.bitcoin.org/index.php?topic=7514.0) on Ubuntu 11.04 64 Bit with SDK2.1.
I overclocked the GPU Core to 900 MHz, and decreased the Mem Speed to 900 as the command "aticonfig --od-setclocks=900,900 --adapter=all" don't let me go under 900 MHz.

How did you set the GPU Mem Speed to 300 MHz. And can somebody explain it to me why this increases the MHash/s rate?

What else can I do to get some kind of 400 MHash/s?

Thanks!

[The Koolio]

Hi,

I have 3 XFX HD5870 Cards. My max MHash/s is approx. 380. I cannot go over 400 :-(

I'm using poclbm (according to this instruction http://forum.bitcoin.org/index.php?topic=7514.0) on Ubuntu 11.04 64 Bit with SDK2.1.
I overclocked the GPU Core to 900 MHz, and decreased the Mem Speed to 900 as the command "aticonfig --od-setclocks=900,900 --adapter=all" don't let me go under 900 MHz.

How did you set the GPU Mem Speed to 300 MHz. And can somebody explain it to me why this increases the MHash/s rate?

What else can I do to get some kind of 400 MHash/s?

Thanks!

use the amd clock tool, input figures in small incriments. I got 1000 mhz and 185 mem getting 440 on my 5870

[PaZer]

Hi The Koolio,

thanks for your answer. Where can I get the amd clock tool? Sounds like a win tool, doesn't it? Where can I got it for Ubuntu?

Another question: Is there also a fan and temperature GUI tool for Ubuntu to monitor ATI Cards?

I'm still intersted in the relationship of low GPU mem speed to higher hash rates 

[icaci]

AMD clock tool is for Windows only. Try the AMDOverdriveCtrl tool on Linux.

[The Koolio]

Hi The Koolio,

thanks for your answer. Where can I get the amd clock tool? Sounds like a win tool, doesn't it? Where can I got it for Ubuntu?

Another question: Is there also a fan and temperature GUI tool for Ubuntu to monitor ATI Cards?

I'm still intersted in the relationship of low GPU mem speed to higher hash rates 

yeh read the above comment, i can only talk for windows. Have you tried flashing your card?

[PaZer]

OK, will check AMDOverdriveCtrl in Linux. Will report later.

Flashin the graphics card? No. Never heard about that. What the advantage? How can I do it?

Again, why does it help to increase hash rate when decreasing the mem speed? What's the dependency?

Thanks

[cablepair]

Just looking for some discussion on optimizing the 5870 and pushing it to its limit. I'll try and update my first post with really useful information for those running 5870's as well. Also looking for advice on mine.

My setup is is two 5870's. I'm using MSI Afterburner to configure both of my cards independently. To adjust voltages and lower memory speeds really low you have to edit the MSIafterburner.cfg file like so

Change "UnlockVoltageControl      = 1"
and
[ATIADLHAL]
EnableUnofficialOverclocking   = 1
UnofficialOverclockingEULA   = I confirm that I am aware of unofficial overclocking limitations and fully understand that MSI will not provide me any support on it
UnofficialOverclockingMode   = 2
AccessibilityCheckingPeriod   = 0

I have one at 980MHz core 180MHz memory and 1.2v and another at 960MHz core 180MHz memory and 1.2v core. Temps stay bellow 80 depending on how much I crank up the fan.
My .bat file to start the latest phoenix miner using slush's pool looks like this.

start /DC:\Bitcoin\phoenix phoenix -v -u http://xxxxxxxxxxxxx@mining.bitcoin.cz:8332/;askrate=10 -k poclbm device=0 WORKSIZE=128 VECTORS BFI_INT AGGRESSION=7 FASTLOOP 

I'm using CCC 11.4 and SDK 2.4.
Each card pulls ~400Mhash/s at these settings in windows 7. I have a Core i7 at 3.8GHz and 6GB DDR3 RAM.

What works best for you guys? Mine clearly isn't ideal and I need to adjust it, I'd also like some input on that..
I've heard using -f 1 or -f 0 can increase your output quite a bit, how's that work for everyone?
What voltages are people running at? I had a hard time getting this stable and went up to 1.2v, but I see people with a lot less. I can't even get my aggression above 7 atm, above that it is likely to crash. I've heard SDK 2.1 is the best, but how does it compare to 2.4? I removed 2.4 and installed 2.1 and my client wouldn't pick up both my cards. Does anyone use their crossfire bridge while mining? I did a short test and couldn't find a difference in productivity, and it's easier for me since this is my gaming system as well.
If anyone needs dummy plugs here's the guide to make them http://www.overclock.net/folding-home-guides-tutorials/384733-30-second-dummy-plug.html.

ive extensively tested different techniques

I have two 5870's
1 - Saphire Vapor-x 1 GB
1 - Saphire EyeFinity 6 Version - 2 GB

Your Optimial Clocking...
980/180 (no need to mess with voltage) 75% Fan
poclbm Guiminer actually is faster than phoenix with this card
You will Get 412-420 MH/s

You can overclock it a little more and get about 10 MH/s more out of it but it wont be stable
these cards will run at 980/180 for years as long as there is sufficient airflow/cooling

[PaZer]

Hi,

done some tests:

AMDOverdriveCtrl
Nice tool, showing temp and fan and much more. You can use to over- and downclock the ATI devices. I could downclock the Mem speed to 300 successfully. Going further down to 250 and below crashed the screen, using ubuntu 11.04 with ubuntu classic GUI.
But there is no way to overclock the GPU over 900 MHz. How this can be done under linux? Is there any way?

btw: downclocking the Mem speed didn't chnge any MHash/s rate. It's 380 MHash @ 1200 MHZ or 900 MHz or 300 MHz. So why downclocking here? Changing the VCore didn't changy anything. In my case only overclocking of the GPU speeds up the hash rate.

phoenix
using phoenix with phatk kernel instead of poclbm pushes my hashrate near to 392 MHash/s but only with Agrresion=13. Problem, the GUI is nearly not useable afterwards, it's extremely slow!

./phoenix.py -u http://xxxxxx/ -k phatk VECTORS BFI_INT FASTLOOP=false AGGRESSION=13 DEVICE=1 WORKSIZE=128

So, in my opinion to get over 400 MHash/s with a 5870 device you have to overclock the GPU to at least 950-980 MHz. But how can I do this?

Maybe I should switch to Windows, how is stability?

Thanks

[icaci]

I'm getting 413 Mhash/s on both my 5870s clocked at 935 MHz core and 319 MHz memory. The point in downclocking the memory is to conserve power and to reduce the heat output, thus lowering card temperature and displacing the inevitable failure further in time. GDDR5 uses lots of power even when idle and lowering it's clock speed two- or threefold reduces the memory power usage accordingly.

Finding the optimal clock speeds is a tricky optimization task. CMOS logic power consumption is proportional to the clock speed and to the square of the voltage. Hashing still uses the card's memory for some storage and time misalignment between GPU's and memory's clocks can lead to wait cycles when incommensurable frequencies are set.

[PaZer]

Hi icaci,

are you running linux or windows? if linux, how did you get the 935 MHz on the GPU?