Problem with inconsistent reported hash rate

[dmoldovan]

Hello, everybody,

I have a rather strange problem here, I have 5 5850 cards all working in deepbit.net pool.
The pool reported hashrate changes over time. it varies from about 1000 MH/s to 1600 MH/s.
I use cgminer. cgminer reports about 320 MH/s for each card. this should be 1600 MH/s ALWAYS.

Can anybody give me a hint to what is/might be the problem.

Thanks all in advance.

[1714979593]

1714979593

[1714979593]

1714979593

[1714979593]

1714979593

[SaintFlow]

do not take this to serious.

What counts is that the correct hashes your mining software says it found is consistent with the
number of correct hashes deebpit says it recieved.

the MH/s deepbit shows you is a based on calculating the correct hashes send in relationship with
the current difficulty. I do not know the exact formula. Fact is your success against difficulty is
fluctuating somewhat in statisical parameters.

The MH/s rate shown on your miner is actual hashes run.



You only get 320 MH/s out of your 5850?

I get 312 MH/s out of 5830

You should be able to get more out of this!

[dmoldovan]

I have them at 890MHz, using cgminer on win7 and winxp.
This is what I get from them

Do you have any ideeas on how to improve te hashrate ?

Thanks

[SaintFlow]

https://en.bitcoin.it/wiki/Mining_hardware_comparison

this is the best reference i can give you

If you have any questions ask
i am shure the community will help

[jake262144]

Aren't you using the latest drivers with 2.6 sdk, OP?

So far, here's a list of what OP is doing wrong:
(1) mining at deepbit - any good reason to choose a pool with mining fees?
(2) using Windows for dedicated miners - why?
(3) most probably using the latest driver/sdk combo which is really bad for VLIW4 and VLIW5 cards.

OTOH, here's what OP is doing right:
(1) using cgminer - good choice!

A few suggestions:
make sure you specify intensity. Use intensity 8 or intensity 9 with those cards.
Use the default values for gpu-threads.
If you're planning to stick to 12.1/2.6 driver/sdk, vectors 4 and worksize 128 should help your hash rate a bit.

BTW, what temps and fan speeds you mining at?

[dmoldovan]

temp: approx 60 degrees celsius, at approx 3300 rpm, using standard reference design cards, made by Saphire.

Is Linux better for mining ? (can I get higher hashrates on linux ?)

PS: I just tried vectors 4 and workload 128, and it results into a 10% decrease in performance.
These machines are Asus  P35 an, P965 boards with Core2Duo CPUs.

Any ideas are welcome.

[SaintFlow]

oh and at what speed do you run the RAM?

Cause anything more then 300 Megaherz will just produce heat.

So if you downclock the RAM to 300
you might get the GPU to 950 or even 1000 without overheating
or instability

[jake262144]

what driver version? Can you confirm the drivers and sdk version?
That's the most important element affecting your performance.

With 12.1 / 2.6 your best bet is to remove the drivers and sdk (this step might be a little tricky) and installing some other combo, preferably 11.12 driver and 2.1 sdk.
2.5 sdk should do fine as well.

[dmoldovan]

Catalyst version 11.12 on one machine, and 12.1 on the other two.
Where do I find the sdk version ?
Anyway, I installed full package driver, including the sdk

Memory was already down-clocked to 300 MHz, but with GPU above 920 MHz, sometimes the display driver gets corrupted, and it re-initializes itself

[jake262144]

12.1 implies 2.6 SDK which sucks performance-wise for 5xxx or 6xxx cards.
How to check AMD SDK version

You can download any sdk version directly from AMD's site
When cleaning up the old sdk, make sure you remove the following files:
C:\Windows\System32\amdocl64.dll
C:\Windows\System32\amdoclcl64.dll
C:\Windows\System32\OpenCL.dll

Try dropping the clocks by 10 MHz, you seem to be pushing the cards (or just one less capable card) a bit too far, hence the driver issues.

[dmoldovan]

On the machine with 11.12 Catalyst, gpucaps says I have OpenCL 1.1, and AMD.APP 831.4

BTW, I use cgminer 2.1.1, and I noticed whan trying out cgminer 2.2.5, that it disables one thread per GPU (now it only uses one thread per GPU) and the hashrate lowers a little (like 10 MH/s lower)
I looked onto ckolivas changelog, but I wasn't able to identify the cause.
Does anybody know why is that ?

[jake262144]

Uhmmm... you didn't set the Intensity as you were supposed to. Without that you can forget about decent hash rate.
Latest cgminer versions revert to just a single thread per gpu if intensity is left at the default value.
Try adding "intensity" : "8" to the config file. You'll need one value per card so you can end up with something like "intensity" : "8,8,8,8,8"
You can always set the number of threads per gpu explicitly: "gpu-threads" : "2".

[DILLIGAF]

I have them at 890MHz, using cgminer on win7 and winxp.
This is what I get from them

Do you have any ideeas on how to improve te hashrate ?

Thanks

Make sure you are setting the intensity correctly if it is on auto then it will hash at a lesser rate. I have one 5850 running at 900/300 core/memory and it gets a rock solid 368MH/s with the intensity set to 8 in cgminer using Ubuntu 11.04. Oh it is nothing for my hash rate to vary by a GH/s +/- over what I run normally on deepbit they seem to have some strange reporting of it there for some reason.

[dmoldovan]

I chose windows because I am not very familiar with linux
but this might have to change if there is a big boost in hashrate.

Do you think that the hashrate difference between you and me (368MH/s vs 330 MH/s) is due to the operating system ?

Or it has to do with CPU and chipset too ?

I have 3 different systems now:
1. P35 with C2D E4500,
2. P965 with C2D E6300, and
3. P55 with i5 760

The i5 has smaller hashrate than E6300 (aggregate, on deepbit, about 100 SHARES less per day), although both have 2 * 5850 @900MHz