5870 worker question, also pps/proportional

[The Koolio]

Hi guys,

Been mining now on a single 5870, i3, windows 7 for about 4 days now. Ive got an average of 300-330 Hash/sec (5870 set to 930 MHz and 300 MHz memory clock) about 3.7BTC as I write this. I dont use the cpu for mining as it seems to slow my GPU hash down a lil bit.

A) Now Im after any advice to speed my work rate up, will creating a second worker on the same GPU help push efficiency? I currently use POCLBM (aggression=13, w=128, frames=20). I want to push my GPU to 1GHz but i cant even get past 940 without a crash using AMD clock tool, im no overclocking expert so any guide or help would be appreciated. I restore to default clocks when using my PC so any advice can be given assuming this is a dedicated miner. Oh I also run my machine on very basic display settings

B) Secondly, I know PPS vs Proportional is a tiresome qn, what I am after though is someone who has a similar high end GPU to tell me their experiences of using either. I know what the 'definition' differences are, but I know real life situations can differ tremendously.

C) I might buy another GP, I was thinking another 5870 but as difficulty increases I doubt Ill see that money back. I have $50-$100 spare, you reckon trawling ebay for some cheap GPU's could provide any realistic value to my quest?

Thanks!!

[1714797964]

1714797964

[Genrobo]

$50-100 to spare?
Get a 5830, they're only about $100 now.
If you overclock a 5830 properly, you can easily pull 275Mh/s +

[grue]

your should be able to overclock your 5870 way more. i get 400 MH/s with my 5850. 1010 core, 250 mem

[innervisi0nn]

C) I might buy another GP, I was thinking another 5870 but as difficulty increases I doubt Ill see that money back. I have $50-$100 spare, you reckon trawling ebay for some cheap GPU's could provide any realistic value to my quest?

http://www.newegg.com/Product/Product.aspx?Item=N82E16814102878&Tpk=saphhire%205830 $109 is a steal
You can run 975 core clock / 340 mem clock and get about 280 mhas/s each, get 2, run about 570 mhash/s

Thats what im doing, best bang for the buck; note though, the 5830 is power hungry (slighty compared to higher models).

[innervisi0nn]

your should be able to overclock your 5870 way more. i get 400 MH/s with my 5850. 1010 core, 250 mem

Yea, MSI Afterburner does the trick Just keep an eye out for heat and stability.

[Dude65535]

A second worker is only likely to help if you see the GPU consistently running at less than 100% utilization or dipping in utilization when waiting for more work.

[The Koolio]

Cheers for the info guys, anyone help me find a decent guide on pushing my 5870 to above 1ghz?

[mewantsbitcoins]

Depending on how long you want to use your card, 1GHz might be not that good of an idea.

I have 5x5870. They are all running 412MH/s at 950MHz core clock and 220MHz memory clock. None of them exceed 75C temp. You can push it further to 420MH/s but it just consumes more electricity with no real gain in processing power

[The Koolio]

Oh and also, if i push to 1ghz, would it be advisable not to keep my memory clock at 300?

[Prze_koles]

Gigabyte Radeon 5870 980mhz/200mhz/stock voltage/temp 64-68     430Mhash/s

[The Koolio]

I did some changes, and now im getting errors in poclbm. can someone tell me what this errror means?

http://www.imagecabin.com/index.php?view=306853199afdcec01dbfa689a&html=1

I changed my clock to 980 and 200 and all i was getting was 350 hash/sec, but each share was cming up with that error in the pic

[MiningBuddy]

Why would you lower your clock more then 300?
I heard from many people that 300 is the idle state and any lower causes damage. 

[max in montreal]

I have the same card. Get me on msn messenger @ maxinmontreal at hotmail dot com, I am getting about 400. Get me on msn and I will walk you through it.

Max

[icaci]

Depending on how long you want to use your card, 1GHz might be not that good of an idea.

I have 5x5870. They are all running 412MH/s at 950MHz core clock and 220MHz memory clock. None of them exceed 75C temp. You can push it further to 420MH/s but it just consumes more electricity with no real gain in processing power

I'm getting 413 MH/s on each of 2x5870 at 935 MHz core and 319 MHz memory clock, fans at 100%. The lower card runs at 45C and the upper - at 75C (cards are too close - bad mobo).

[The Koolio]

why am i only getting 350 at similar clocks? also why has my share generation turned into constant long polling errors?

[Prze_koles]

That poclbm error is Deepbit related. Nothing you can do about it. Also, for different cards, different memory clock rates are optimal. Try something between 200-400mhz.  If you want, you can check phoenix miner with phatk kernel. I'm assuming you're using windows with ATI SDK 2.4, then try this:
phoenix -u http://USER:PASSWORD@POOL_ADDRESS:8332/ -k phatk VECTORS BFI_INT FASTLOOP=false AGGRESSION=12 DEVICE=0 WORKSIZE=128

[The Koolio]

thanks, yeah i will try pheonix, and you are right with the SDK. The errors were once every 20-40 accepted shares. Now as of 30 mins ago its 10 errors for every 1 share accepted, something isnt right?! is anyone else getting this many errors now with deepbit, have they gone down or something

[icaci]

why am i only getting 350 at similar clocks? also why has my share generation turned into constant long polling errors?

There is a noticeable difference in hashing speeds between 319 MHz and 300 MHz. What miner are you using? What options? How about running Phoenix with phatk kernel that is specifically optimised for 5870 and SDK 2.4? My options are:
VECTORS BFI_INT FASTLOOP=false AGGRESSION=13 WORKSIZE=256
WORKSIZE=128 gives a little lower result with SDK 2.4. With SDK 2.1 worksize of 128 gives better hashing speed but still lower than worksize of 256 with SDK 2.4. Beware, that's not a general rule but specific to the phatk kernel!

[Prze_koles]

Just checked:
390Mhash/s with 256
430Mhash/s with 128
Not working for me I think 

[The Koolio]

my options are  AGGRESSION=12 WORKSIZE=128. i WAS USING POCLBM, however i think i might give pheonix a go, either way im getting these stupid long polling errors. can anyone confirm whether deepbit is down? i was using my miner for last 5 days no problems, then today i restarted and installed the latest 11.5 catalyist for my 5870 and now all i get are long polling errors, would that install have created them?

[max in montreal]

Phoenix is exactly what I am using. The best aggression for me is actually 7, and I am using 128 as worksize. at one point I had results better than 430, but now its at about 400.

When you guys mention SDK 2.4 do you mean the catalyst control center? Can you pleas post a link where I can download SDK 2.4, as of right now I am just using the latest Catalyst Control Center, and MSI afterburner, but my sliders will not go as low, or as high as mentioned in other posts above.

[Prze_koles]

my options are  AGGRESSION=12 WORKSIZE=128. i WAS USING POCLBM, however i think i might give pheonix a go, either way im getting these stupid long polling errors. can anyone confirm whether deepbit is down? i was using my miner for last 5 days no problems, then today i restarted and installed the latest 11.5 catalyist for my 5870 and now all i get are long polling errors, would that install have created them?

I don't think so, my friend is also getting those errors. That's why there's no fee at Deepbit atm.

[icaci]

When you guys mention SDK 2.4 do you mean the catalyst control center? Can you pleas post a link where I can download SDK 2.4, as of right now I am just using the latest Catalyst Control Center, and MSI afterburner, but my sliders will not go as low, or as high as mentioned in other posts above.

SDK 2.4 is the one bundled with the 11.5 drivers. Look at the downloaded file name. It should be something like:
11-5_vista32_win7_32_dd_ccc_ocl.exe
The "ocl" part means that the OpenCL SDK is included. Just (re-)install it.

[The Koolio]

i knew they had problems, even when announcd i was only getting the odd error, now it seems every worked share is an error!

[icaci]

Just checked:
390Mhash/s with 256
430Mhash/s with 128
Not working for me I think 

413 Mhash/s with WORKSIZE=256
406 Mhash/s with WORKSIZE=128

Phoenix 1.48 + phatk, Windows 7 32-bit, SDK 2.4, Phenom 140 CPU, dual 5870 at 935/319.

[Prze_koles]

Ok, clocked memory from 200mhz to 350mhz and got 432Mhash/s (2Mhash/s increase). Not sure if it's worth it (more power and temperature, I guess?)

[LegitBit]

When you guys mention SDK 2.4 do you mean the catalyst control center? Can you pleas post a link where I can download SDK 2.4, as of right now I am just using the latest Catalyst Control Center, and MSI afterburner, but my sliders will not go as low, or as high as mentioned in other posts above.

Sort of, AMD currently bundles the opencl driver with the Catalyst driver, so just make sure you have a clean install of the newest version, 10.5, At the moment. You do not need to install the SDK unless you have issues. (At least on a windows based-machine.)


Be careful with over-clocking, I recommended using "MSI Afterburner" like others have said, or "Saphhire TRIXX" if you have a Sapphire card. Use those with HWINFO http://www.hwinfo.com/download32.html to check the temperatures and utilization rate.

The most important thing to remember is to set the fan up high, some utilities will allow you to modify the rates at which the fan turns up, others just specify a static percentage.

Keep the GPU under 60 C if you are running this all the time. Using HWINFO you can see the Shader temp, use this as your scale instead of the core temp. Hashing (mining) uses the shaders almost exclusively. Other utilities will give you an overall temp, but may be underscoring the shader highs.

Tweak slowly, bump the core up at 20 a time then test using either the built-in AMD tool, or Kombuster which comes with MSI Afterburner, as you run into issues, lower by 5 or slightly increase the voltage, be very careful with voltage, you can easily fry the card and this is not covered by warantee.

EDIT: I mention keeping the temps under 60, although higher temps won't necessarily fry the machine, (Most cards are rated to around 90C) Long term this can cause heat damage. Ambient temperature in the room can also rise over time if not well ventilated. You don't want to come back from work and find the room at 85F with a burnt our card is all I am saying.

[The Koolio]

ahh its ok now, both deepbit and slush pools working with poclbm and phoenix. I seem to be getting around 10 m/hash more in phoenix, just a query though, the displayed hash, are they true or average, reason is ask is poclbm often shows me at 330~ but then deepbit will display my average in last 5 mins at 350 or something like that

[icaci]

Ok, clocked memory from 200mhz to 350mhz and got 432Mhash/s (2Mhash/s increase). Not sure if it's worth it (more power and temperature, I guess?)

There are "sweet spots" in core/memory clock ratio that lead to great performance. Their existence is a consequence of the fact that GPU core must access data in memory from time to time and if core and memory are out of clock sync, core wait cycles are added and that leads to performance decrease. If one has enough time he can experiment with varying the memory clock speed in search for those sweet spots.

[LegitBit]

Ok, clocked memory from 200mhz to 350mhz and got 432Mhash/s (2Mhash/s increase). Not sure if it's worth it (more power and temperature, I guess?)

Memory does hardly anything and is not worth the temp increase. Hashing uses the Shaders in the core mostly,

In fact, many miners underclock the memory to a very low value, though be careful when doing this as you may crash the machine if you are trying to use the machine at the same time. If you are using Windows, I'd be sure to set your desktop theme to the most basic, as in never versions (vista and up) the desktop is actually GFX accelerated, and could interfere with mining or cause a crash when coming back from screen lock.

[The Koolio]

ok now ive managed to get 410 on phoenix using 128 and aggr=13. I used the same settings on poclbm but never got anywhere near that many hash/sec.

Ok now that im fine tuning my speeds can i ask whether slush or deepbit would be better for me?

[max in montreal]

Slush takes 2% deepbit takes 3%.

Slush's website has a few problems when I try connecting, but I think its worth the 1%...

[Prze_koles]

Ok, clocked memory from 200mhz to 350mhz and got 432Mhash/s (2Mhash/s increase). Not sure if it's worth it (more power and temperature, I guess?)

Memory does hardly anything and is not worth the temp increase. Hashing uses the Shaders in the core mostly,

In fact, many miners underclock the memory to a very low value, though be careful when doing this as you may crash the machine if you are trying to use the machine at the same time. If you are using Windows, I'd be sure to set your desktop theme to the most basic, as in never versions (vista and up) the desktop is actually GFX accelerated, and could interfere with mining or cause a crash when coming back from screen lock.

Actually I was comparing using Worksize=128 (430Mhash with 200mhz memory) and worksize=256 (432Mhash with 350mhz).  And I installed classic menu shell, because I hate Windows 7 start menu

I'm using BTC Guild, 0% fees, and no problems since new server started.

[The Koolio]

many thanks max, btw i added u on msn. When you get a chance walk me through o/c. I did download msi afterburner, but its worth noting that it will not accept my changes in clock values. I changed the allow overclocking setting in the conf file but still continues to not accept my changes. For the moment ive been using amd clock tool, do you think i should just change the settings to 1000, 300 memory clock straight away or work my way up? also clock tool has me at 1.1 voltage atm, i cant change this to anything else (like 1.35) only 0.9, 1 and 1.1 are available for selection

[LegitBit]

Actually I was comparing using Worksize=128 (430Mhash with 200mhz memory) and worksize=256 (432Mhash with 350mhz).  And I installed classic menu shell, because I hate Windows 7 start menu

I'm using BTC Guild, 0% fees, and no problems since new server started.

Ahh, my bad.

Are you getting better yields with BTC Guild?
I currently use Eligius and get about 1btc a day on average with similar speeds.

[max in montreal]

Windows 7 32 bit 5870 using Phoenix Miner, latest Catalyst Control Center using TRIXX Sapphire Tweak Utility setting core clock at 1000, memory clock at 1200, default voltage at 1125, temperature seems steady at 68 degrees Celcius. Whole Tower is using about 275 Watts of electricity. My 300 Watt PSU could not handle it as sometimes it spikes to 305 and the whole PC reboots, so I purchased a 750 Watt PSU. The MB fits one graphic card only and uses an Intel Quadcore 6600.

A batch file placed in the startup gets everything going automatically after a reboot.

Hash rate fluctuates between 455 and 465

Time to clone the drive, set up the second system...and then I have to get 2 more cards for 2 other identical systems.

this was something I posted in another forum when I first started.

[The Koolio]

ive got mine running at 440 hash now fairly stable. I only ever really mine when not using my pc, so would you suggest i keep my 185/300 memory clock (sometimes i change between the two but 185 is fastest so far) rather than your 1200?

[The Koolio]

running at 440 hash according to phoenix (1ghz, 185 mem) however btcguild is saying my average is 520 hash, which is the more accurate or true? I think btcguild is giving me way better results than either slush or deepbit cud. I have no invalid shares or errors...! I think this is much better than i cud have hoped, anyone got any reasons why i shudnt be with btcguild?

[max in montreal]

I cannot change the clocks by that much, i think the max on one is 900 and the min on the other is 600

[icaci]

running at 440 hash according to phoenix (1ghz, 185 mem) however btcguild is saying my average is 520 hash, which is the more accurate or true? I think btcguild is giving me way better results than either slush or deepbit cud. I have no invalid shares or errors...! I think this is much better than i cud have hoped, anyone got any reasons why i shudnt be with btcguild?

Nope, pools can't measure your real hash rate. All they do is compute an estimate of your speed based on the number of low difficulty blocks (a.k.a. shares) you submit in a given timeframe. It goes like that: 1 pool share equals one solved block at difficulty of 1 which takes about 4 Ghashes on average to solve. If you submit 10 shares for 1 minute that is 40 Ghashes/min or 667 Mhash/s on average. But one should consider that part of that minute is network latency associated with getting work and submitting back each result. It also depends on how lucky you are - it could take 1 hash to solve the share (very unlikely) or 400 Ghashes (once again very unlikely). It's all simple statistics - look at "geometric distribution" in Wikipedia or some other math book if you would like to understand the theory.

Your miner software measures the actual number of hashing operations done in a unit time and thus provides much more accurate estimate on the hashing power of your rig.

On average it shouldn't matter if you mine solo or in a pool. It should not matter if you mine at deepbit, slush or BTCGuild, besides the different taxes, number of staled shares, server latency and whether you are paid for invalid blocks or not.