Bitcoin Forum

So I am trying to squeeze the most of my Sapphire Radeon 5850, using Phoenix and with these flags
-v -k phatk2 VECTORS BFI_INT AGGRESSION=8 FASTLOOP=false WORKSIZE=64
those flags seem to work the best for my 5850, currently getting 305 Mhash/s
the gpu is at 750 Mhz core clock and 400 Mhz memory clock.
Is there anything else I could do (that isn't oc) for getting more Mhash/s?
Perhaps changing some flags or adding any?

What's wrong with overclocking ? I run mine at 860 and 850 MHz. That's not really hard for the cards. Some folks here run them over 900. You can lower memory down to 150 MHz.

I will overclock them but I also want to get the most out of the software

You can switch to cgminer but I doubt that this will speed up your cards. Phatk is the fastest kernel for my 5850s.

swissmate - are you running at stock voltage? If so, you should be able to do better. My Sapphire 5850s are getting about 308 Mh/s undervolted to 0.95. At stock voltage I can easily get 350+. My settings are similar to yours except I use worksize 128. I found that 64 throttled the hashrate quite a bit. But maybe it is different for you. Also, why are you using -v? That may not be hurting, but it's probably not helping either. I think you can raise the core clock quite a bit. I think mine will do at least 800 at stock voltage (and 760 undervolted). Try raising the core clock in little steps until the card freezes. Then back off the core clock that induced freeze by about 20 Mhz (or less if you want to run closer to the edge). You can also lower your memory clock to 300 - won't improve hashrate, but the card will run a little cooler and consume less power (but it won't be a big difference).

Another thing to keep in mind is that maximizing the hashrate that appears on your screen does not correlate tightly with your share submission rate (what actually pays). I've found that above an aggression level of 9, my share submission rate starts to fall off even though a higher hashrate is reported by phoenix. Since your aggression is 8, I assume this GPU is connected to a monitor and you want to be able to use the computer without lag? I have a GPU connected to a monitor, and for this one I use aggression 7 and turn fastloop on. I seem to recall reading somewhere that fastloop is irrelevant now, but I seem to get a better share submission rate with it turned on at lower aggression. At the very least, it doesn't seem to hurt.

5850 AMDdesign @ 850mhz @ 1.012V @ ~350mh/s @ cgminer 2.4.1

My stock core clock is 725 and worksize 64 seems to be better for mine.
Thanks for the help, without the -v I'm getting extra 7 Mhash/s
and I'm changing also the agression and fastloop parameters.
Thanks  :)

Sapphire 5850, BAMT, phoenix phatk2 VECTORS BFI_INT FASTLOOP=true AGGRESSION=11 WORKSIZE=128 MEMORY=300 CORE=865 VOLTAGE=1.0v FAN=75%Temp=64.5 c RATE=355.57 MH/s.

Testing right now on my rig.
--
ok won't work, resetting driver, GPU0 dont like.

I run mine at 1187 voltage, 980 core and 300 ram... get about 410 mhash

Try 1.088v I tweaked mine for weeks.

P.S. Check Giga's farm http://gigamining.com/mgpumon/ lots of info there.

I only get 347 Mhash/s at 900 MHZ. 

What software does everyone use?
 - SDK version
 -Catalyst version

Wuy i should go back to default voltage? They run now with 1.012 for a week. Just GPU0 dont like it with 1.000V, this cards is connected to my monitor.

The "sweet spot" for windows drivers and 5xxx cards will be 11.12 and 2.1 SDK. Underclock memory to about 3/16 of the core speed (800 core = 150MHz memory) and use a worksize of 128 if you are trying to optimize electricity efficiency. If you have free electricity, feel free to use stock or higher volts, 256 worksize, and memory to 4/11 of core speed (1000 core ~ 360 memory). Memory ratios are approximate, you'll likely need to fiddle around with them. The 256 worksize + higher memory is maybe 1-2% faster than the 128 worksize + lower memory, but comes with extra electricity cost (about 5 watts per card in my experience)

64 worksize is bad, and only starts to scale better than W256 with relatively high clocks starting at about 500MHz. At stock memory clocks, 128 = 64 > 256

Hardware
5850, Core=800MHz, Memory=300MHz, Volt=1,094, Intensity=4, Hashrate=320Mh/s
Five-day average.

Software
cgminer 2.4.1, kernel=diablo, vectors=2, worksize=128, threads=4
Catalyst 12.4, no additional SDKs

Undervolting and underclocking the memory, reduces power consumption.
Hope this helps.

I got a dud 5850 and can't go past 865MHz. Even tho, I still get ~335MH/s @ 300mem with CGMiner 2.4.1 (phatk) on Win7x64, 12.4 drivers, and 2.5 SDK.

Find the failure. ^^

I was under the impression that 2.5 worked just as well as 2.4. It was 2.6 that gave ~10% lower hash rates. Also, what does the SDK have to do with not being able to OC? My 5870 and 5830 OC to 960 and 940, respectively. My 5850 only goes to 875.

I just mean your hashrate buddy, can be 350mh/s @ 860mhz.

I'm okay with mine now

320 Mhash/s @775Mhz

0.96 volts(cgminer shows 0.959),  835 core, 150 mem, 344 Mhash ( a grand total of 80 watts per 5850)

This is what I have installed and it also gets rid of pesky 100% cpu bug.

https://bitcointalk.org/index.php?topic=52143.msg653435#msg653435 (https://bitcointalk.org/index.php?topic=52143.msg653435#msg653435)

Ya you're totally right. I was at work and got the numbers wrong. I just checked and it's doing ~354MH/s @ 865. Damn you were pretty close!

I just switched from BAMT (using Phoenix2) to cgminer (with the default kernel) running on the Peppermint Two 64 bit edition linux distro. I'm seeing ~2% increase in efficiency and I can clock my cards a bit higher than I could with BAMT. The difference could be due to switching from 32 bit to 64 bit, but whatever the reason, it's a nice little improvement for my 5850s. The desktop is also much more usable (much quicker) than it was with BAMT.

yap. ^^