V1.7.1
- Added new algo FreeHaven
- Small OpenCL optimisations
- Added 3 different modes for heavy algos (Heavy, Haven, BitTube, Italo, FreeHaven)
- Added value '9999' for fragments, that sets internally to the value it was in previous 1.6.8 version
- Fixed benchmarking mode + startup monitor bug
- Fragments, aes, bralock and heavy_mode can be set now outside gpu_conf, so it applies to every GPU


- Freehaven algo is here, a lot of you were waiting for this
- Added new parameter "heavy_mode", it can be used in gpu_conf, or just in the config, it is applied if not in auto mode (intensity = 0).
It can be 1,2,3.
1: This is the default method for doing heavy calculations now, it's more stable and should produce less compute errors than in the previous 1.7.0 version. Note that worksize 9 isn't the optimal worksize anymore for Vegas if using this mode. Worksize 8 now gives better results, you just need to find a good value for intensity.

2: This is the calculation method from the previous 1.7.0 version, use it if you can get better results with it than with mode 1. Note for Vegas : worksize 9 is optimal here.

3: Experimental mode, it is the fastest but tends to produce the most compute errors, if your GPU's are tuned maximally to the edge. I put it here because maybe someone can get the best results with it.

- Ellesmere 8gb cards just love to throttle hashrate on heavy algos, lot of you said you had better results on previous versions, so if you set fragments to 9999 it uses the same internal settings for it used in 1.6.8.
I am sure with the new mode for heavy calculations higher hashrates can be achieved, but a lot of users just copy over the old settings and guess what, does are not optimal anymore.
With the new method you should lower the intensity a little bit than before, and find the right fragments setting for your GPU's. Some users reported they have higher hashrates after playing a little bit with these settings.

- I had to add parameters 'fragments', 'aes', 'bralock, and now 'heavy_mode' to the config file outside the 'gpu_conf' because.. well, nobody reads the first page or readme.txt anymore

You are missing a , after "double_threads" : true

Don't repeat, but read back. Try various values for 'fragments' parameter. Benchmark uses same config as normal mining.

It looks like that at the end 1.7.0 is better also with Ellesmere 8G cards than older SRB versions, just needs some fine tuning, as few users who played a little bit with fragments reported.
This is good to know.

So after setting the fragments in gpu_conf, you get higher hashrate than on 1.6.7 ?

There is a DOC maintained by users here : https://docs.google.com/spreadsheets/d/1pH326fwnYWC5Jef2FdF4KnJ4CLj0cwDDtGjrlgekk1A/edit#gid=514064850 it contains settings for various cards.
It can be found on the first page , under the Tips section.

No problem i got used to angry people 
What is really making me confused sometimes is that some complain on XY gpu new version is a lot worse, while others for that same XY gpu say the new version is a lot better.

I know something got 'lost' for older cards on the road, in the next version i will again add a 'useless extra tuning gadget', which will maybe bring back what got lost for those older cards.

Well if windows hangs then you can't do anything else but restart the machine physically.
Maybe you should lower a little bit your overclocking settings so it won't hang ?

API in SRB changed on every release?

4gb ellesmere cards are better on 170, only the 8Gb cards do that magical hashrate game.
On previous versions you had sometimes to wait for a few hours to get all cards hashing at for example 1100hs, and even then sometimes they never reached to that maximum.
For some the gpu-z tricks worked, for some not. I managed to make that hashrate 90% stable on 170, so no need to start the miner X times to maybe reach a maximum hashrate, and no need for gpu-z tricks.
So you get a little lower hashing speed, but it is much more stable.

which algo is best for you on 1.68 , and what gpu's ?

Why do i always have to repeat myself : i don't have every amd card to test against, so yes, it may happen that an optimisation for one is not working out for the other, especially this is related to older cards.
The extra gadgets don't take off from your hashrate, but they can gain you if 10hs then 10hs, but they won't screw 50% of gpu's as you say.

I agree with dingdongtobias, if there is an older version that works good for you, then why not use that instead?
Further optimisations will be all towards newer cards in the future, those i got and can test with.

--setcomputemode has to be run as admin

yup, it is a bug, it's going to be fixed

the question is which algo was without errors on 1.6.4, and on which algo you have the catastrophic hash drop?
because you talk about stellite which is a v7 variant, then you say you use 2 miners for v8.

I will introduce a few modes for heavy algos in next version, so everyone can find the one that works for him best.

1. Optimal intensity can sometimes differ even for the same brand cards, so i don't really know how to calculate the best value. Unfortunately its a try-fail game until you find the best one. It's not always the highest intensity that gives the best results.

2. Intensity = number of global threads (in other miners that value is SRB intensity x 16). Now when you do scratchpad size * global threads you get the approx size of memory that will be used.
example for heavy : 56 * 16 * 4 = 3584mb.

3. There is, check the first page (--enableduplicategpuid (enables the usage of the same gpu id in gpu_conf multiple times))

FINALLY!

you won't believe how glad i am that you finally resolved that issue.

are you sure your psu has enough juice?
V8 needs more power than V7

which card specifically, and what was the hashrate on 1.6.8 and what it is on 1.7.0 ?
I want to try to fix this issue.

Need more info, like is it stellite only or EVERY algo? same settings? what have you tried beside running new version with old config ?

I dont have every possible gpu to try out when coding, sometimes what pushes one card, takes off another, so dont be immediately mad bros 

play with fragments (its enough to try 0,1,2,4,8,16) and see which one gives the best results for that particular card.
also try aes 0,1 or 2 .
Fast algo is V7 variant nothing has changed in code for v7 and variants
my guess is that the default fragment value isn't optimal for your cards.

also in the next version i will also return the 'old' code for heavy, that was in 1.6.7, and some of you say it was better for you than the one now.
So with a new parameter (yes, again), you can choose which mode for heavy you want to use.

That way everyone will be happy