Quote from: Wolf0 on October 02, 2014, 11:10:26 AM

Quote from: pallas on October 02, 2014, 10:50:45 AM

Quote from: Wolf0 on October 02, 2014, 08:59:25 AM

Quote from: SavageWS6 on October 02, 2014, 04:27:16 AM

Quote from: Wolf0 on October 02, 2014, 01:16:17 AM

Quote from: SavageWS6 on October 01, 2014, 08:00:30 PM

Can also confirm lost 2 MH/s total mining X11 combined with my 3x R9 270s and my 1x R9 280X on 14.9 drivers

All right, then I'm sure I fucked up my test somehow.

Did you delete the old .bin's from whatever drivers you had before, and had new ones generated for 14.9 drivers?

Of course I did. I'm gonna re-do the test now, though. 14.6 vs. 14.9 on both my X11 kernels.

Ooookay... this is odd. The hashrate changed - so it seems the driver did as well. Further, I fetched the date on fglrx each time before starting (14.6 being May 29 2014 and 14.9 being Sept 2 2014) so it DID change. As for the bin files, yes, I removed them; I double-checked. Anyway, Savage, this is what I've got, columns are GPU/clocks/mine/stock, rows are obviously GPUs.

Code:

Running driver version 14.6:
270X - 1155/1500 - 3.64 - 2.80
290X - 1045/1600 - 7.96 - 4.86
7950 - 1155/1500 - 5.03 - 4.01
280X - 1155/1500 - 5.77 - 4.54

Running driver version 14.9:
270X - 1155/1500 - 3.64 - 2.80
290X - 1045/1600 - 7.98 - 4.95
7950 - 1155/1500 - 5.03 - 4.05
280X - 1155/1500 - 5.75 - 4.60

changes are too small: probably just compiler variance.
compiling the same .cl file with the same settings may result in different hashrates (I believe a 2% range is common).
the file size changes too.
it doesn't happen with all the kernels but I'm sure it does on x11 as well as on my groestl, for example.

EDIT: so long for those fine optimizations which result in 1% improvement and you think it's good... and it's nothing instead ;-)

I don't understand your last comment. If you're saying my optimizations get 1% improvement, then you're reading the table wrong, because the third COLUMN is my kernel's hashrate, the fourth is the stock kernel's hashrate with the same GPU, driver, and clocks.

Example: On driver version 14.9, my 290X achieved a hashrate of 7.98MH/s with my kernel, and 4.95MH/s with the kernel that is on SGMiner's master branch right now. The clocks were the same for both runs, 1045/1600.

sorry, I didn't mean that.
I was just saying that to me and other developers I know, it often happened that an optimization which led to a 1% improvement, or even more, in the end turned out to be just that variance, so not an optimization at all. It probably happened to you as well.