Quote from: DiabloD3 on March 15, 2011, 05:44:25 AM

Quote from: Cerebrum on March 14, 2011, 01:00:11 PM

Quote from: DiabloD3 on March 14, 2011, 08:39:07 AM

Cerebrum, I think you've overlooked -d. It tells you lots of information, including when it submits a hash, and if the hash is accepted or not.

Also, my miner double checks all solutions before submitting... if there is a hw error, it bitches very loudly without -d.

And I don't believe poclbm's hash meter is correct. Mine lists two numbers, 15 second average, forever average. poclbm does not have a meter that is comparable to either. My miner is also faster because poclbm does not try to keep work queued for the hardware.

In addition, since I queue multiple kernels in parallel, network activity for remote pools costs much less than it does on poclbm.

Wonderful. Thanks for these clarifications. It is never complaining about invalid or stale hashes so I guess I'm never making any.

I was noticing that when I run a single instance of DiabloMiner with all 4 graphics cores, I get worse performance than running 4 seperate instances with one for each core. Do you have any idea why that is? The performance difference isn't huge (about 4-5%) but it's definitely present. Any idea as to why it works better with 4 independent instances than with all 4 in the same instance?

EDIT: Forgot to ask. Is there any way to start up a fake X session from the command line so that I can start the miner via SSH? Or do I have to look into getting VNC installed on my desktop so I can log into the graphical environment remotely?

You shouldn't be getting less. Are you sure you're doing your math right? Run with -f 1 connecting to a local, and let it run for 5-10 minutes and check the second number.

If you're on Radeon hardware, you require a real X session running. You don't have to use VNC or use X for anything at all, just start your miner with DISPLAY=:0 ./DiabloMiner over ssh as the user that started X.

Thanks for pointing out my obvious error. It turns out that I wasn't waiting long enough when doing that test on single vs. multiple instances, and I'm actually getting about 500KH/s more per core from the combined instance.