OFFICIAL CGMINER mining software thread for linux/win/osx/mips/arm/r-pi 4.11.1

[-ck]

Ok, my findings so far:
5850 on SDK 2.5

2.2.7 ~400MH/s with phatk120213

2.3.0 ~320MH/s with phatk120222
2.3.0 ~400MH/s with phatk120213 *but* lots of HW errors!

So it seems the hashrate drop is in the kernel changes? 

The old kernel is simply not compatible with the current one.

Well this makes me fucking angry. All I can do is shake my fist even harder at ATI for the changes just SHOULD NOT CAUSE THIS as they're meant to be trivial changes. Well more fucking quick fucking releases to deal with fucking ATI fucking fail. Rollback phatk fucking time.

[1714055540]

1714055540

[1714055540]

1714055540

[1714055540]

1714055540

[Vbs]

OK, fixed it:
"bool result = any(W[117] == 0)" seems to be the slow op, changing back to the bitwise AND "W[117].x & W[117].y" (and the next if to !result), gets me back at ~400MH/s.

[-ck]

OK, fixed it:
"bool result = any(W[117] == 0)" seems to be the slow op, changing back to the bitwise AND "W[117].x & W[117].y" (and the next if to !result), gets me back at ~400MH/s.

Goddamn. Thanks for that. The any() function is meant to be faster than manually unrolling it the way I did in the old version. I guess I shouldn't put any faith in their hardware functions...

sigh...

[Vbs]

Btw, why check for 'result' in the first place? Isn't that a redundant check?

Using now,

Code:

#elif defined VECTORS2
          if (!W[117].x) {
              output[FOUND] = FOUND;
	      output[NFLAG & W[3].x] = W[3].x;
          }
	  if (!W[117].y) {
              output[FOUND] = FOUND;
	      output[NFLAG & W[3].y] = W[3].y;
          }

Seems to work!

[jamesg]

I have posted a bounty to get better documentation around the --scan-serial option for linux -> https://bitcointalk.org/index.php?topic=65879.0

If you have this knowledge, you can make some quick btc... 

[-ck]

Btw, why check for 'result' in the first place? Isn't that a redundant check?

Using now,

Code:

#elif defined VECTORS2
          if (!W[117].x) {
              output[FOUND] = FOUND;
	      output[NFLAG & W[3].x] = W[3].x;
          }
	  if (!W[117].y) {
              output[FOUND] = FOUND;
	      output[NFLAG & W[3].y] = W[3].y;
          }

Seems to work!

No, you're doing two branches on the common path now. We want as few branches as possible in the common path. It doesn't matter if we do 2 extra checks in the uncommon (i.e. found share) path.

[bulanula]

Ok, my findings so far:
5850 on SDK 2.5

2.2.7 ~400MH/s with phatk120213

2.3.0 ~320MH/s with phatk120222
2.3.0 ~400MH/s with phatk120213 *but* lots of HW errors!

So it seems the hashrate drop is in the kernel changes? 

The old kernel is simply not compatible with the current one.

Well this makes me fucking angry. All I can do is shake my fist even harder at ATI for the changes just SHOULD NOT CAUSE THIS as they're meant to be trivial changes. Well more fucking quick fucking releases to deal with fucking ATI fucking fail. Rollback phatk fucking time.

Yeah and people were bashing me when I was screaming Nvidia.

ATI will probably never fix these bugs in their SDK and drivers.

CPU bug, SDK 2.6, 8 GPU hardcoded limit, needs to be linked to xserver at all times.

If only Nvidia would wake up and see that we are waiting for the 4608 shader dual GPU.

Hopefully the integer performance is much better this time around and they implement a variation of BFI_INT or bitalign.

So when will you release the fixed cgminer ?

Thank you !

[kentrolla]

awesome update. the new kernel gives me a +2% increase in mh/s.  
i have a suggestion.
i like how you can set the intensity to d and it changes the intensity to maintain desktop interactivity. but its just a little bit too intense for me in windows but is perfect while playing world of tanks (weird, i thought gaming would be more intense)  so i keep having to change the intensity depending on what i'm doing.

my suggestion is that you make an option to set the intensity to d -2  or d -1  or something. so the intensity is still dynamic but is 1 or 2 less (or more) than d.
some people might want to use d +1 or d +2. who knows?  for me, intensity of d -1 would be perfect.

[-ck]

awesome update. the new kernel gives me a +2% increase in mh/s.  
i have a suggestion.
i like how you can set the intensity to d and it changes the intensity to maintain desktop interactivity. but its just a little bit too intense for me in windows but is perfect while playing world of tanks (weird, i thought gaming would be more intense)  so i keep having to change the intensity depending on what i'm doing.

my suggestion is that you make an option to set the intensity to d -2  or d -1  or something. so the intensity is still dynamic but is 1 or 2 less (or more) than d.
some people might want to use d +1 or d +2. who knows?  for me, intensity of d -1 would be perfect.

Read what it says at startup when you run dynamic intensity... sigh.

[kentrolla]

awesome update. the new kernel gives me a +2% increase in mh/s.  
i have a suggestion.
i like how you can set the intensity to d and it changes the intensity to maintain desktop interactivity. but its just a little bit too intense for me in windows but is perfect while playing world of tanks (weird, i thought gaming would be more intense)  so i keep having to change the intensity depending on what i'm doing.

my suggestion is that you make an option to set the intensity to d -2  or d -1  or something. so the intensity is still dynamic but is 1 or 2 less (or more) than d.
some people might want to use d +1 or d +2. who knows?  for me, intensity of d -1 would be perfect.

Read what it says at startup when your run dynamic intensity... sigh.

lol
ty

[-ck]

Updated the 2.3.0 package with -1 extension which includes a fixed phatk kernel. Links in top post. This shouldn't affect other kernels...

If it ends up affecting other kernels, I'll have to release a full new version tomorrow.

[Vbs]

Ok, one unrolled branch then! 

Code:

#elif defined VECTORS2
          if (!W[117].x) {
               output[FOUND] = FOUND;
	       output[NFLAG & W[3].x] = W[3].x;
	            if (!W[117].y)
                         output[NFLAG & W[3].y] = W[3].y;
          }
          else if (!W[117].y) {
               output[FOUND] = FOUND;
	       output[NFLAG & W[3].y] = W[3].y;
          }

[-ck]

Ok, one unrolled branch then! 

Code:

#elif defined VECTORS2
          if (!W[117].x) {
               output[FOUND] = FOUND;
	       output[NFLAG & W[3].x] = W[3].x;
	            if (!W[117].y)
                         output[NFLAG & W[3].y] = W[3].y;
          }
          else if (!W[117].y) {
               output[FOUND] = FOUND;
	       output[NFLAG & W[3].y] = W[3].y;
          }

Heh, you're not a coder are you? That's still two branches unless it's positive on the first branch.

[kentrolla]

awesome update. the new kernel gives me a +2% increase in mh/s.  
i have a suggestion.
i like how you can set the intensity to d and it changes the intensity to maintain desktop interactivity. but its just a little bit too intense for me in windows but is perfect while playing world of tanks (weird, i thought gaming would be more intense)  so i keep having to change the intensity depending on what i'm doing.

my suggestion is that you make an option to set the intensity to d -2  or d -1  or something. so the intensity is still dynamic but is 1 or 2 less (or more) than d.
some people might want to use d +1 or d +2. who knows?  for me, intensity of d -1 would be perfect.

Read what it says at startup when you run dynamic intensity... sigh.

--gpu-dyninterval <arg> Set the refresh interval in ms for GPUs using dynamic intensity

i've been playing with this and i guess i just don't understand how it works. when i set it to 20 it stays at 4 intensity and when i set it to 1 it goes back and forth between 3 and 4 intensity.  so there is a difference between --gpu-dyninterval 1 and --gpu-dyninterval 20 but its barely even noticeable and you can't use negative numbers.  i have a radeon 4850 btw

[DeathAndTaxes]

Ok, one unrolled branch then! 

Code:

#elif defined VECTORS2
          if (!W[117].x) {
               output[FOUND] = FOUND;
	       output[NFLAG & W[3].x] = W[3].x;
	            if (!W[117].y)
                         output[NFLAG & W[3].y] = W[3].y;
          }
          else if (!W[117].y) {
               output[FOUND] = FOUND;
	       output[NFLAG & W[3].y] = W[3].y;
          }

Heh, you're not a coder are you? That's still two branches unless it's positive on the first branch.

To save ckolivas from more frustration maybe I can help.

vbs:  AMD (and I assume Nvidia) GPU take a horrible hit on branches.  The number of total checks is irrelevant.  What matters is the number of branches on the main path.

Only one in 4.3 billion hashes will be a share thus 99.999999976716935634613037109375% of the time any subsequent share checks are never executed.  Optimizing the path which occurs one in 4.3 billion executions is silly right?

We want to make the one that occurs 4.29999999999 billlion out of 4.3 billion attempts as fast as possible.  Given the massive (and I do mean massive forget what you think you know about C++ compilers on x86 hardware) hit that AMD GPU take when it comes to branches that means making the main path have as few branches as possible. 

Neither of your code snippets do that.

[-ck]

awesome update. the new kernel gives me a +2% increase in mh/s.  
i have a suggestion.
i like how you can set the intensity to d and it changes the intensity to maintain desktop interactivity. but its just a little bit too intense for me in windows but is perfect while playing world of tanks (weird, i thought gaming would be more intense)  so i keep having to change the intensity depending on what i'm doing.

my suggestion is that you make an option to set the intensity to d -2  or d -1  or something. so the intensity is still dynamic but is 1 or 2 less (or more) than d.
some people might want to use d +1 or d +2. who knows?  for me, intensity of d -1 would be perfect.

Read what it says at startup when you run dynamic intensity... sigh.

--gpu-dyninterval <arg> Set the refresh interval in ms for GPUs using dynamic intensity

i've been playing with this and i guess i just don't understand how it works. when i set it to 20 it stays at 4 intensity and when i set it to 1 it goes back and forth between 3 and 4 intensity.  so there is a difference between --gpu-dyninterval 1 and --gpu-dyninterval 20 but its barely even noticeable and you can't use negative numbers.  i have a radeon 4850 btw

It's a fairly sensitive tuning tool. The difference between the default 7 and 1 should be quite significant to the interface. Don't get too hung up on what the intensity setting reported is since that's done once every 5 seconds but at dyninterval of 1 the value is updated 1000 times per second. 1 is very interactive desktop with lower hashrate, 1000 for example is higher hashrate less dynamic dekstop.

[Diapolo]

Btw, why check for 'result' in the first place? Isn't that a redundant check?

Using now,

Code:

#elif defined VECTORS2
          if (!W[117].x) {
              output[FOUND] = FOUND;
	      output[NFLAG & W[3].x] = W[3].x;
          }
	  if (!W[117].y) {
              output[FOUND] = FOUND;
	      output[NFLAG & W[3].y] = W[3].y;
          }

Seems to work!

No, you're doing two branches on the common path now. We want as few branches as possible in the common path. It doesn't matter if we do 2 extra checks in the uncommon (i.e. found share) path.

That is true, I thought it would be better the above way, but think about it ... most of the time only 1 branch is checked and therefore processed. If you always check 2 branches, this will slow things down. It has to be mentioned, that if in a work-group, let's
say of size 256, a positive nonce is found, all 3 branches are checked and this slows down execution of ALL work-items in that particular work-group. So because a positive nonce is seldom, Cons current approach is the right one.

I found out in internal tests with diakgcn, that any() is indeed slower :-/.

Dia

[kentrolla]

awesome update. the new kernel gives me a +2% increase in mh/s.  
i have a suggestion.
i like how you can set the intensity to d and it changes the intensity to maintain desktop interactivity. but its just a little bit too intense for me in windows but is perfect while playing world of tanks (weird, i thought gaming would be more intense)  so i keep having to change the intensity depending on what i'm doing.

my suggestion is that you make an option to set the intensity to d -2  or d -1  or something. so the intensity is still dynamic but is 1 or 2 less (or more) than d.
some people might want to use d +1 or d +2. who knows?  for me, intensity of d -1 would be perfect.

Read what it says at startup when you run dynamic intensity... sigh.

--gpu-dyninterval <arg> Set the refresh interval in ms for GPUs using dynamic intensity

i've been playing with this and i guess i just don't understand how it works. when i set it to 20 it stays at 4 intensity and when i set it to 1 it goes back and forth between 3 and 4 intensity.  so there is a difference between --gpu-dyninterval 1 and --gpu-dyninterval 20 but its barely even noticeable and you can't use negative numbers.  i have a radeon 4850 btw

It's a fairly sensitive tuning tool. The difference between the default 7 and 1 should be quite significant to the interface. Don't get too hung up on what the intensity setting reported is since that's done once every 5 seconds but at dyninterval of 1 the value is updated 1000 times per second. 1 is very interactive desktop with lower hashrate, 1000 for example is higher hashrate less dynamic dekstop.

hmm, my friend with a 5870 is saying that --gpu-dyninterval does effect his intensity. but for me, it seems to do nothing at all. with gpu-dyninterval 1 i still have lag in windows but if i use intensity -2 the lag goes away and i only lose about 5mh/s.  i guess ill stick to just using static intensity for now

[DeathAndTaxes]

awesome update. the new kernel gives me a +2% increase in mh/s. 
i have a suggestion.
i like how you can set the intensity to d and it changes the intensity to maintain desktop interactivity. but its just a little bit too intense for me in windows but is perfect while playing world of tanks (weird, i thought gaming would be more intense)  so i keep having to change the intensity depending on what i'm doing.

my suggestion is that you make an option to set the intensity to d -2  or d -1  or something. so the intensity is still dynamic but is 1 or 2 less (or more) than d.
some people might want to use d +1 or d +2. who knows?  for me, intensity of d -1 would be perfect.

Read what it says at startup when you run dynamic intensity... sigh.

--gpu-dyninterval <arg> Set the refresh interval in ms for GPUs using dynamic intensity

i've been playing with this and i guess i just don't understand how it works. when i set it to 20 it stays at 4 intensity and when i set it to 1 it goes back and forth between 3 and 4 intensity.  so there is a difference between --gpu-dyninterval 1 and --gpu-dyninterval 20 but its barely even noticeable and you can't use negative numbers.  i have a radeon 4850 btw

It's a fairly sensitive tuning tool. The difference between the default 7 and 1 should be quite significant to the interface. Don't get too hung up on what the intensity setting reported is since that's done once every 5 seconds but at dyninterval of 1 the value is updated 1000 times per second. 1 is very interactive desktop with lower hashrate, 1000 for example is higher hashrate less dynamic dekstop.

hmm, my friend with a 5870 is saying that --gpu-dyninterval does effect his intensity. but for me, it seems to do nothing at all. with gpu-dyninterval 1 i still have lag in windows but if i use intensity -2 the lag goes away and i only lose about 5mh/s.  i guess ill stick to just using static intensity for now

I just noticed you are using a 4850 to mine and drive the desktop.  That is kinda an edge case.  The computational power on the 4850 is so low (relatively speaking) that it is likely outside the bounds of any adjustment calculations. 

[Vbs]

Thanks for the explanations guys. I'll play a bit with the kernel when I have some free time, if I get any improvement I'll tell. I have experience with C but not much with OpenCL, but I'll probably have to get into it for my job...