Extra Flags, general tweaking, and different miners discussion

[Fhqwhgads6680]

Hey Guys, I thought since this is XS we should be experimenting and finding ways to get the most out of our hardware. First thing I have noticed is a new version of guiminer is out: http://forum.bitcoin.org/?topic=3878.0 That adds the phoenix miner to Windows. I have tested this on all 4 of my cards: hd6870's hd6950, hd5850 and received about a 15-20 mhash boost using the following Flags: -k phatk platform=0 device=1 VECTORS BFI_INT AGGRESSION=11 worksize=128 FASTLOOP=false

Also of note here is platform refers to the device number (if 0-0 then use 0 if 0-1 then use 1)

Also any suggestions on how we can organize this thread with tweaks and tips?

[OldChap]

Also any suggestions on how we can organize this thread with tweaks and tips?

Copied from the other thread...This is a better place.

Phoenix arguments

-v (verbose) - Logs additional debug messages to the console. Default is disabled.
-q (queue size) - Sets the size of the internal work queue. Default is 1. This shouldn't need to be changed for most GPU miners.
-a (average samples) - Sets the number of samples to use for hashrate averaging. Default is 10. You might want to lower this for longer kernel execution times. (high aggression)
-u (URL) - Sets the URL of the work server. The correct protocol is selected based on the prefix (RPC for http://, MMP for mmp://)
-k (kernel) - Selects which kernel to load. Default is poclbm. All other arguments MUST come before specifying a kernel. Any arguments after -k are sent to the kernel.

Poclbm kernel arguments

PLATFORM=ID - Sets the OpenCL platform to use. This isn't needed if you only have a single platform.
DEVICE=ID - Sets the OpenCL device to use. This isn't needed if you only have a single device.
VECTORS - Enables 2-way vectors. This may improve hashrate if enabled, but it can be slower on some hardware. Default is disabled.
AGGRESSION=LEVEL - Sets the aggression. This allows you to control the kernel execution time to improve hashrate or reduce interface lag. Default is 4.
WORKSIZE=SIZE - Sets the worksize. Tweaking this setting may improve performance similar to poclbm's -w flag. Default is the maximum supported by the device.
FASTLOOP - Enables fast internal loop. This improves hashrate at lower aggression levels without introducing any additional interface lag. Default is enabled.
BFI_INT - Enables the BFI_INT instruction on newer AMD/ATI GPUs. This significantly improves hashrate. Default is disabled.

NOTE 1: Using FASTLOOP at higher AGGRESSION won't improve performance. However, it no longer results in stale shares. To disable FASTLOOP use FASTLOOP=false.
NOTE 2: The phatk kernel uses the same command line options

[p2501]

I have noticed that the best agression setting for 6990 cards is 11. Also the difference between kernels seems to be negligible, worksize optimum is 128, above the production drops off.

[KaptainBlaZzed]

I have been using the Phoenix Rising GUI for the phoenix miner and really like it.

The main feature l like is it has an auto restart feature after a set time of inactivity. currently mine is set to 30 seconds.

[BeepBeep2]

I got 7 more Mhash/s from my 5770's using the phatk kernel...running APP 2.4
Both cards are at 1000 core, getting 219 each.

-k phatk VECTORS DEVICE=0 and 1 AGGRESSION=10 FASTLOOP BFI_INT WORKSIZE=128

AGGRESSION=10 seems to give less rejected results. I've tried up to 13 but just got more results rejected and lower efficiency in my pool. No load/mhash benefit for me after 9/10.

My 4870X2 gets 1 Mhash benefit with 64 worksize over 128, 256 is worse than both. sort of negligible I guess but with a GPU like that every one counts.