30+ Mhs drop adding a second card

[conspirosphere.tk]

One of my rigs with a single 5870 @965/300 used to have an output of 460+ Mhs (with phoenix 1.7 phatk2)
Since I added a 5830 the performance of the 5870 dropped to barely 440Mhs even if I increased the oc to 970 (keeping the miner in foreground. In background it drops much further to 404 Mhs).
Even the 5830 is not satisfying: @940/300 just 310 Mhs.

Anyone has any explanation/suggestion?  
I use this string on the 5870:
phatk2 DEVICE=0 VECTORS FASTLOOP=false AGGRESSION=8 WORKSIZE=256 -a 1000
Before with aggression=7 I had higher performance, and I cannot increase it more in order to keep my puter responsive.
My string on the 5830 is:
phatk2 DEVICE=1 VECTORS FASTLOOP=false AGGRESSION=7 WORKSIZE=256 -a 1000
With these strings I seem to have the best performance. I have catalyst 11.6.
Board is an Asrock Alivexfire e-sata 2 with an amd 5600 x2 cpu running Xp.

[1714148831]

1714148831

[1714148831]

1714148831

[1714148831]

1714148831

[Kluge]

Are they set up in xfire? If not, they should be running independently and have no perfrmance decrease unless the mobo's PCI bandwidth scales way, way down with 2+ cards.

[conspirosphere.tk]

Are they set up in xfire? If not, they should be running independently and have no perfrmance decrease unless the mobo's PCI bandwidth scales way, way down with 2+ cards.

No Xfire. But I found that the 2 PCIEx16 slots on this crappy board can run at x16 speed only when there is just one card inserted in one of them. Using both slots with two cards decrease their speed to PCIEx8. This may explain my performance drop? 

[Kluge]

Are they set up in xfire? If not, they should be running independently and have no perfrmance decrease unless the mobo's PCI bandwidth scales way, way down with 2+ cards.

No Xfire. But I found that the 2 PCIEx16 slots on this crappy board can run at x16 speed only when there is just one card inserted in one of them. Using both slots with two cards decrease their speed to PCIEx8. This may explain my performance drop? 

Probably not. I've read there shouldn't be any bandwidth issues unless it scales down to 2x or under. I can't think of what the issue would be. Maybe when you set the 5830 clock @ 940, it also sets the 5870 @ 940? If you are using something like MSI AB, make sure "Synchronize clocks" is unchecked in options. otherwise, I'm stumped.

[P4man]

Can you try another miner to rule that out as possible cause? Cgminer or bitminter?

[conspirosphere.tk]

Thanks for your help. I found something interesting: I use to run a dynamical clocking utility for my cpu (CrystalCPUID) which underclocks and undervolts the cpu when is not utilized, so that most of the time it runs at 220*6=1320Mhz.
Well, if I set the cpu at minimum (220*6)=1320Mhz my cards produce:
5830: 300 Mhs
5870: 430 Mhs
If I set the cpu at middle range (220*10)=2200 Mhz I get:
5830: 310 Mhs
5870: 440 Mhs
And If I set the cpu at max (220*14)=3080 Mhz I get:
5830: 316 Mhs
5870: 446 Mhs
Strange thing is that I do not have the cpu bug. Phoenix needs just 2-5% of my cpu, so I wonder why it affects at all my hashing.
Giving Above Normal priority to Phoenix does not seem to improve anything.

I tried yesterday Bitminter beta which seems to give slightly lower output than Phoenix.

[RyNinDaCleM]

You can run on 1x pcie extenders. So, it's not a bandwidth issue.

[jake262144]

I use to run a dynamical clocking utility for my cpu (CrystalCPUID) which underclocks and undervolts the cpu when is not utilized

You shouldn't have to resort to third-party programs for CPU housekeeping. It's just wrong and now you see why.
Can't your mobo take care of CPU power saving? I don't believe I've ever seen one that can't.

Search the BIOS for options like C1E, AMD Cool'n'Quiet, Intel SpeedStep, Enhanced Intel SpeedStep Technology (EIST, EISST) and make sure they're on.
Perhaps upgrade your BIOS to the latest version while you're messing with it?

Giving Above Normal priority to Phoenix does not seem to improve anything.

If that machine is your dedicated miner I don't see any advantage in running Windows on it. Linux is where you probably should go.

[mokahless]

So did you manage to figure out the problem?
I recently encountered this in windows 8 under cgminer 2.9.7
6950 went from ~360 to 338 when I added a 5770 in the second slot. Note also that I used to have a 6450 hashing in the second slot before. At first I suspected heat, but the rate doesn't hit what it did before even when it's still warming up. I haven't tested mining on just the 6950 yet.

[zvs]

change the priority of that process to real-time

it should be fine with that.  if you're, say, running four 5870's on a single core sempron, then you wouldn't want to use real-time, just high...

and drop the 5870 to 965-170 at 128 worksize    (or maybe 965-195 in phoenix)

[mokahless]

Sorry, I forgot to mention my first card is a 6950

[krackajak]

Any luck with this?

I'm experiencing the same thing. I have a 4 card setup, 3 5830's and a 6970. All are down by 20-30 Mhash when together. Solo they each perform fine. I've tried cgminer, bitminter, and phoenix with same results. Mobo is a MSI 890-FXA GD70. Cgminer only uses 2-3% cpu.

[mokahless]

I've looked at mine and determined it is a combination of heat and AMD power control.
When I up the AMD power control to +20, I can get the old rates but of course both heat and power increase so it isn't sustainable.
I've also determined that the thermal throttling of the 6950 is 102.5*C. It will clock down about every 2 minutes, for about a second, which is why I didn't notice the throttling before. At a glance, I will likely still be looking at 830MHz.
I can sustain under the throttle limit probably until the outside temperature hits 10*C, where I am. At which point, I hope BFL finally delivers...
Don't worry about my card's longevity people, It has double lifetime warranty XFX has already replaced my fan once because it broke.

So try cranking up the AMD power control to +20 if heat is not an issue and you are using windows.

[krackajak]

I've looked at mine and determined it is a combination of heat and AMD power control.
When I up the AMD power control to +20, I can get the old rates but of course both heat and power increase so it isn't sustainable.
I've also determined that the thermal throttling of the 6950 is 102.5*C. It will clock down about every 2 minutes, for about a second, which is why I didn't notice the throttling before. At a glance, I will likely still be looking at 830MHz.
I can sustain under the throttle limit probably until the outside temperature hits 10*C, where I am. At which point, I hope BFL finally delivers...
Don't worry about my card's longevity people, It has double lifetime warranty XFX has already replaced my fan once because it broke.

So try cranking up the AMD power control to +20 if heat is not an issue and you are using windows.

That sounds like a setting that is only found in the Catylst Control Center. I don't have that installed on my machines. I found that I could get 20 of the 30 Mhash back by using SDK 2.1 instead of 2.4

I find it weird that 2.4 would perform differently on 1 card vs 4 cards. It seems like the SDK just gets bogged down.

[mokahless]

That sounds like a setting that is only found in the Catylst Control Center. I don't have that installed on my machines. I found that I could get 20 of the 30 Mhash back by using SDK 2.1 instead of 2.4

I find it weird that 2.4 would perform differently on 1 card vs 4 cards. It seems like the SDK just gets bogged down.

Here's a thread about it:
https://bitcointalk.org/index.php?topic=8384.0

That said, I forgot that AMD only introduced it with their 6xxx series cards so it definitely wouldn't account for your 5830s. What are your temps by the way?

[skeeterskeeter]

I have this very same problem.  (FULL Description https://bitcointalk.org/index.php?topic=146839.0)

I have a 7770 and a 610GT and soon to be two more 5830's.

When I run the 7770 and 610 under the same pool they work fine. When I run them by themselves under a pool they work fine. But If I run one in a pool and one solo they both go way down in Mhash/s. If I run them both solo they both go to kHash/s range!!!!!! I have tried different flags, the whole shabang but nothing ever comes from it.

Maybe a new miner here tonight? (Cg or Phoenix) Though I am not sure how to set these up to run in a pool and solo simultaneously. Not so happy using the command line, but hey I am a CS I should like it.


I wish there was a fix for this.


Using GUIminer, poclbllbmmbll (whatever the acronym is), and OpenCLMiners. Bitcoin-qt as wallet server.

Evga Z77 FTW mobo, 3570K cpu.

[krackajak]

5830: 78
5830: 78
5830: 84
5830: 75
6970: 82

I am running warm.

Just upgraded from 2.4 -> 2.5 No considerable gains/losses. If anything I gained 1-2 Mhash

[skeeterskeeter]

My temps are the same as the posts above mine. Hot.

Changed to Phoenix lsat night and it seemed to not make a difference.

[mokahless]

But neither of you are even close to the thermal limit. I know from experimentation that a 6950 will thermal throttle at 102-103*C. Blame XFX's crappy coolers (their single fan ones anyway) for my knowledge about this.
Even that said, 78-84*C is normal. It is not hot. Especially considering the load Bitcoin mining puts on the cards. When you are approaching 90*C, that is when you should maybe start to be concerned about the longevity of the card (Anyone remember the 4870 single slot? Those things certainly had electromigration after a few years. A lot of cards 88-100*C load.)

[skeeterskeeter]

There is thermal limit, the longevity limit, and there's the I feel l ok with these temps limit. I've blown through them all at some point lol.