Bitcoin Forum

I wanted to dabble with btc a bit, so around a week ago I slipped a 5830 into a spare box that was lying about, started reading up, tweaking it when I had time, etc.

I'm happy to say that thanks to years of practice at overclocking-while-poor I am now able to achieve a sustained rate of 312 MHash/s (peak 312.71) courtesy of everyone on the forums who posts such useful info :)

My setup:

 Ubuntu 10.04.2 LTS x86_64
 fglrx 8.801
 stream 2.1
 phoenix 1.3
 gpu @ 1000
 mem @ 195
 v @ 1.163
 fan @ 100
 temp @ 74.5

I know this isn't the most efficient rig energy-wise, but I wanted to see how much bang-for-the-buck I could get off an entry-level card.

If there's reason to suspect a newer fglrx or stream package would bump it even higher, I'm probably game to push a little harder.

Oh, and if anyone wants the details on how I arrived at this number, I could probably go through my browser and command history and piece it together.

Cheers, and thanks again everybody!

You could write up a howto and post it on http://witcoin.com (http://witcoin.com) and earn even more bitcoins :)

That's REALLY nice. I'm pulling about 295 off of my XFX 5830 Windows 7 box, but this is my primary computer as well as my mining rig (because I am just THAT poor). I can't get the core clock much above 970 without losing system stability, and the whole computer crashes with the memory clock set to anything lower than 380. Can't seem to adjust the voltages on this card at all, either, even with an unlocked OC tool.

Seems like the 5830 is a decent card for an entry level hobbyist such as myself. I picked mine up for $109, and it was cheap enough that I got it in under the wife's radar. :P She's Bitcoin-friendly, anyway, but still...this little hobby can get expensive!

Wow those are some nice numbers for both of you! I have two Asus EAH5830's, seems the most I can pull is about 250Mhash/sec, and that's at 1000Mhz core, can't get any more out of it :( Win 7 box..

Any suggestions? It is just a dual core @ 3.2 and it's seeing 100% CPU utilization, even though I'm only mining on the GPU's, could that be my issue?

I've heard a lot about this issue...I'd be curious to see definitively what causes it.

Short of getting you to roll your Catalyst version back to something earlier and making sure you are running with SDK 2.1, I'm not sure what to tell you. What's your memory clock at?

...you don't have Generate Coins checked in the Bitcoin client, do you...?

Running catalyst suite version 11.4, no SDK kit as apparently this suite includes the directCU drivers, although that could be part of the issue I suppose..

I have the memory at stock 1000Mhz right now, I had tried dropping it down to around 400Mhz previously but it did not have any impact..

Double checked, Generate Coins is not checked off either. Just ran home from work to check on things during lunchtime here, when I'm back home tonight I might try a few different driver versions or the SDK kit and see what I can come up with. Both cards just cruising at 250mh/sec for now

Huh...yeah...that's pretty strange. Give the Catalyst rollback a shot...I went with 10.7, and if you have a 5830 in a Windows box, that's what I'd recommend for you, too.

Honestly the returns for going above say 975 aren't worth much.  I have it locked at 1000/90 basically just for fun (dedicated rig, test lab environment), and because I can :)  I want to try pushing the GPU further when I have some more time, but no hardmods.

WRT to the memory settings though, I saw definite gains when I moved downward from stock.

I would definitely run it that fast if I could. 965 is the fastest stable speed for my system given the current configuration and the fact that I actually use my computer for stuff while it's mining. Reducing the memory clock to below 400 definitely helped too...though too low and I was freezing up.

I also have an xfx 5830 and after seeing this thread switched from puddinpop's miner to phoenix 1.4 (see details below). At a core clock of 970MHz I saw an increase from ~220 MHash/s to ~285 MHash/s.

OS Windows 7 64bit
Catalyst driver: 11.4
Stream version 2.4 (presumably - it's whatever comes with the 11.4 drivers. Anything less won't work for me, at all)
Core clock 970Mhz (somewhere between here and 1000Mhz things get unstable)
Mem clock 560Mhz (can't set it any lower in MSI Afterburner)
voltage is default (unable to change?)

Miner Phoenix 1.4 -k poclbm device=0 platform=1 WORKSIZE=128 AGGRESSION=12 BFI_INT VECTORS

Any suggestions for getting it higher? I seem to be stuck with stream SDK 2.4 and can't raise the core clock much more. For reference, I'm also running a nvidia 260 in the same computer.

This is strange. What happens when you try and downgrade the drivers?

I've tried both 10.3 with stream SDK 2.1 and 2.2 and 10.4 with SDK 2.1 and 2.2. Regardless of combination I get the same problem - whenever I try and run any OpenCL code it just crashes (the command window glazes over and the not responding box pops up). This happens when trying to run rpcminer-opencl.exe on any device (I don't think it even gets as far as finding any usable platforms) and also when I try and run any of the sample code that comes with the SDKs.

Switching back to the 11.4 package fixes the problem and miners and the code examples run fine. Do you reckon it's worth the effort to try and fix this problem and use 2.1 or 2.2? How much faster are they?

Cheers

10.3 and 10.4 seem a little old...I'd personally try 10.7 and see what happens from there. :) SDK 2.1 is what I'm using with your same OS with quite a bit of success. I'm getting very similar speeds to you while using my computer.

C:\Phoenix\phoenix.exe -u http://REDACTED@deepbit.net:8332 -k poclbm VECTORS BFI_INT FASTLOOP AGGRESSION=7 DEVICE=1

Edit: Try turning the Worksize off...BFI_INT seems to somehow mitigate the need to process the work in smaller pieces. At least, it feels that way. No change in speed at all after the new miners were installed.

Forgot to mention, I kept seeing the MHash drop off when the screen would disable even though I configured Ubuntu to ignore all power saving settings; whenever I'd wake the screen up by jiggling the mouse or hitting a keystroke, it would be sub-300 but then on the next workunit jump right back up to 312+.

My solution, since this is a dedicated mining rig, was to set the session to xterm, and use AMDOverdrvCtrl with the -b flag to load a saved .ovdr profile followed by calling out Phoenix.  In this way I was able to eliminate the performance drop, and also since I have the system on UPS and set to auto-reboot / auto-logon, I basically have a bulletproof bitcoin dispenser :)

Thanks for the suggestion, is it possible to install 2.1 with the 11.4 drivers perhaps? I could try that and narrow down whether it's the SDK or the drivers causing the problem.

Just for reference, my current setup gets about 278 MHash/s with aggression=7.

Are you sure the hash rate was actually dropping while the screen was off? I've noticed in Windows 7 that waking the screen by moving the mouse causes a brief ~10-20% dip in GPU utilization (seems to be true for both ATI and nvidia cards). I have no idea about Linux really, but could it be possible it's just a blip as the screen comes back on? Another workaround would just be to set the screen to always on and turn it off yourself - something I have to do because one of my Dell screens audibly beeps while in standby mode (?!)

I haven't tried it, but it probably won't hurt. :P

RE: OpenCL versions, I see a definite decline when I move to 2.2 SDK or later.  It's possible I could re-optimize for a different version, but even though this is an enjoyable hobby I don't have THAT much free time :)

RE: MHash/s vs. screen, yes this was a measured difference of about 10 MHash/s apparently caused by GDM's method for 'sleeping' the display.  Switching to an xterm session caused it to go away.

Personal best on this 5830 is stable at
320 Mhash/s
still 2.1 SDK
phoenix+poclbm
1025/320
90%

Still need to play with phatk, overvolt, flexible proxy miner, etc.  Fun little toy project :)  Card will in theory try to go up to 1200 GPU core, so I'd imagine I might in theory be able to get up to what... 350?  I don't think I can do that with stock cooler and volts though.

How are you OCing the core above 875 in linux?  aticonfig is limited for me.

RBE

Meaning I'd have to read up on RBE haha.

Anyway are these on stock coolers? I'd be really happy to get 320 Mhash/s, what temperature are they running on? As long as it's long term sustainable @ above 300 Mhash/s + no additional components (cooling etc.) I'll say it's best bang for buck.

Thx for the reference. I'll be adding that to the Beginner's Guide if you don't mind. :) I was just starting a section on the more 'Advance' tweaking.

Also, is this 320 Mhash/s achievable on stock cards and what brand of 5830 should I be getting?

Thread starter, may I also ask that I link to this thread

Can you perhaps post a detailed guide or pm me one?

I'm planning on setting up a cluster of 40 (yes 40) Sapphire 5830 between 20 rigs in the near future.