CCminer(SP-MOD) Modded GPU kernels.

[sp_]

maybe we could set a global flag when, say, 3 shares submission fails, then add that flag to the check you outlined.

yes, sounds good. We also need a failover function if one gpu is failing. Now ccminer will just exit if one of the gpu's in the rigs are failing. one flag per gpu, and a sleep and restart method.

[pallas]

maybe we could set a global flag when, say, 3 shares submission fails, then add that flag to the check you outlined.

yes, sounds good. We also need a failover function if one gpu is failing. Now ccminer will just exit if one of the gpu's in the rigs are failing. one flag per gpu, and a sleep and restart method.

Do you want me to try doing that flag thing or are you going to do it yourself?
I'd do it right now but I can't test network downtime 'cause I'm remotely connected to the rig ;-)

[sp_]

I prefer to spend time on faster kernals. If you can do fork my fork and make a pullrequest when you are done, I will add it.

[sp_]

Release 56 has finally made be break 100 mh/s on quark. Virtual beers for everyone!

Aliman,
Your 980ti is beating the Titan-X  

The titans in this picture are to close to eachother. They will heat up and trottle.
Bether to use 70cm usb risers and spread the cards.
The titans are not good for mining. 12GB of ram doesn't help, and the prise is skyhigh.

[djm34]

Glad to do it. Please point me to the piece of code that does the check and I can look into it.
Some more info: ubuntu 15.04, network card and link are up but the connection to the gateway is down.
From the power meter I can see a slight reduction in power usage but the output, while including the connection issues, it also includes the normal hashing of the cards.

take a look at this commit:

From the tpruvot branch: prevent gpu scans before a job is recieved:

https://github.com/sp-hash/ccminer/commit/8f7f6dfd8f627fefd2f5a2fbda4612827588ef02

This will prevent the miner to start if no connection is there in the startup. But if the connection has timed out, there is no code that stops the gpu to mine

maybe we could set a global flag when, say, 3 shares submission fails, then add that flag to the check you outlined.

actually it is a behavior which appeared "recently", the first versions of ccminer didn't behave like that (and yes it is boring...)

[myagui]

[...] The titans in this picture are to close to eachother. They will heat up and trottle.
Bether to use 70cm usb risers and spread the cards.
The titans are not good for mining. 12GB of ram doesn't help, and the prise is skyhigh.

Hm, cuda porn!!! 

The TitanX's should be awesome for high n-factor scrypt-n and scrypt-jane. The 12GB of VRAM will be entirely useful in Linux, and the cards will run much cooler with a properly selected lookup gap, trading a little bit of hashpower, for a lot of thermal and power savings. For every other algorithm, meh, waste of money 

[fenomenhaa]

From my side +50-65 k/hash for 750Ti's and 200k/hash for 970  with built 56. Good work sp... )))

More is coming... If you want a higher hashrate, Please donate some beers.

7.05 MHASH per 750ti.

With djm34's changes we perhaps get 7,5MHASH per 750ti ??

Sounds good ! ))) 750 ti @7.5mhash 

[Slava_K]

C11 solomine errore when try yay or booo....

[crsminer]

From my side +50-65 k/hash for 750Ti's and 200k/hash for 970  with built 56. Good work sp... )))

More is coming... If you want a higher hashrate, Please donate some beers.

7.05 MHASH per 750ti.

With djm34's changes we perhaps get 7,5MHASH per 750ti ??

Small donation here, more to come -- 0.1 BTC - 9adb5d9cb8d84a5d01ad0c6f11eca8eaaeadcc268091050e2e1daf4c9847258e

Thanks!

[AliMan]

How's the power consumption on 980ti , compared to 970 ?

Lets see, this does 268 - 280W at the max, running at 1500Mhz. My MSI GTX 970 4G consumed something like 220W, given the custom PCB and heftier power connectors.

[AliMan]

Release 56 has finally made be break 100 mh/s on quark. Virtual beers for everyone!

Aliman,
Your 980ti is beating the Titan-X  

The titans in this picture are to close to eachother. They will heat up and trottle.
Bether to use 70cm usb risers and spread the cards.
The titans are not good for mining. 12GB of ram doesn't help, and the prise is skyhigh.

Wrong about the heat; servers tend to be cooled differently. Right about everything else.

Cooled differently by magic I presume?

[AliMan]

Release 56 has finally made be break 100 mh/s on quark. Virtual beers for everyone!

Aliman,
Your 980ti is beating the Titan-X  

The titans in this picture are to close to eachother. They will heat up and trottle.
Bether to use 70cm usb risers and spread the cards.
The titans are not good for mining. 12GB of ram doesn't help, and the prise is skyhigh.

Wrong about the heat; servers tend to be cooled differently. Right about everything else.

Cooled differently by magic I presume?

Forced cool air in the front and out the back.

And those are the characteristics of the blower type heatsink, it's got nothing to do with being a server. Right now, it's actually like a test bench. And SP is right about the risers and all.

[AliMan]

Release 56 has finally made be break 100 mh/s on quark. Virtual beers for everyone!

Aliman,
Your 980ti is beating the Titan-X  

The titans in this picture are to close to eachother. They will heat up and trottle.
Bether to use 70cm usb risers and spread the cards.
The titans are not good for mining. 12GB of ram doesn't help, and the prise is skyhigh.

Wrong about the heat; servers tend to be cooled differently. Right about everything else.

Cooled differently by magic I presume?

Forced cool air in the front and out the back.

And those are the characteristics of the blower type heatsink, it's got nothing to do with being a server. Right now, it's actually like a test bench. And SP is right about the risers and all.

I meant, rack mounted systems tend to have cold air forced in the front - you could actually have them not overheat.

But unfortunately, they still do over heat. One card absorbs the heat of the other, and it moves all the way to the last card.

[dga]

Release 56 has finally made be break 100 mh/s on quark. Virtual beers for everyone!

Aliman,
Your 980ti is beating the Titan-X  

The titans in this picture are to close to eachother. They will heat up and trottle.
Bether to use 70cm usb risers and spread the cards.
The titans are not good for mining. 12GB of ram doesn't help, and the prise is skyhigh.

Wrong about the heat; servers tend to be cooled differently. Right about everything else.

Cooled differently by magic I presume?

Forced cool air in the front and out the back.

And those are the characteristics of the blower type heatsink, it's got nothing to do with being a server. Right now, it's actually like a test bench. And SP is right about the risers and all.

I meant, rack mounted systems tend to have cold air forced in the front - you could actually have them not overheat.

But unfortunately, they still do over heat. One card absorbs the heat of the other, and it moves all the way to the last card.

This is also one of the differences between the consumer line and the Tesla line.  The k80, for example, doesn't ship with a fan:  It's designed to be cooled by the server's airflow, where the air flows directly through the card from front to back (exiting the chassis out the back).  The GTX 980 and Titan draw air from *next to them* - a 90 degree angle from the card - and so stacking them in a dense configuration as in the photo means that their airflow is restricted and they're drawing in air that's already been heated a bit by the back of the card next to them.  The k80 design requires more careful integration with the chassis design, but with big fans, can be stuffed more densely than the consumer lines.

[zTheWolfz]

New release.


-Faster quark (optimized bmw-512)
-Added the credit algorithm (from the djm34 branch)
-Added the c11 algorithm(from the tprovot branch, but all kernals are replaced)

1.5.56(sp-MOD) is available here: (22-07-2015)

https://github.com/sp-hash/ccminer/releases/

The sourcecode is available here:

https://github.com/sp-hash/ccminer

Not sure if it was to be or a problem but I noticed that 1.5.5.56 no longer shows IP algo block numbers in blue like it did in .54 & .55
Also seeing could not validate on CPU a good bit now in .56? My Nvida cards are in another PC or I'd do SS so show what I talking about.
Maybe later if you need more info I can login from that rig to post pictures for more info. I do only solo mining some scrypt and neoscrypt "Phoenix Coin" at this time.
EVGA GTX960 SSC 2GB seeing 567 speed on scrypt at this time -i 9 -l T16x24. Forces -i 9 doesn't matter what you try to use on scrypt.

[sambiohazard]

Release 56 has finally made be break 100 mh/s on quark. Virtual beers for everyone!

Aliman,
Your 980ti is beating the Titan-X   

The titans in this picture are to close to eachother. They will heat up and trottle.
Bether to use 70cm usb risers and spread the cards.
The titans are not good for mining. 12GB of ram doesn't help, and the prise is skyhigh.

Wrong about the heat; servers tend to be cooled differently. Right about everything else.

Cooled differently by magic I presume?

Forced cool air in the front and out the back.

And those are the characteristics of the blower type heatsink, it's got nothing to do with being a server. Right now, it's actually like a test bench. And SP is right about the risers and all.

I meant, rack mounted systems tend to have cold air forced in the front - you could actually have them not overheat.

But unfortunately, they still do over heat. One card absorbs the heat of the other, and it moves all the way to the last card.

This is also one of the differences between the consumer line and the Tesla line.  The k80, for example, doesn't ship with a fan:  It's designed to be cooled by the server's airflow, where the air flows directly through the card from front to back (exiting the chassis out the back).  The GTX 980 and Titan draw air from *next to them* - a 90 degree angle from the card - and so stacking them in a dense configuration as in the photo means that their airflow is restricted and they're drawing in air that's already been heated a bit by the back of the card next to them.  The k80 design requires more careful integration with the chassis design, but with big fans, can be stuffed more densely than the consumer lines.

Just to make this last time someone quotes this, have you read the article from which this pic comes. They had fans forced at 100% and yet temp were at 80C. Here is a quote from the article explaining what the machine is

"Here is something interesting that is not originally intended to be used as a GPU mining rig, but for a powerful compute oriented machine – an 8x GPU AsRock Rack Barebone populated with eight Nvidia GeForce GTX TITAN X video cards and two 14-core Intel Xeon processors. These powerful rackmount systems are designed for use with Nvidia Tesla cards for high-performance CUDA applications, but you can put in GeForce cards as well such as the TITAN X and have them all running at PCI-E 3.0 x16 speeds (no SLI support is available)."

Article link: http://cryptomining-blog.com/5263-mining-with-a-8x-gpu-geforce-gtx-titan-x-system/

[zjy]

From my side +50-65 k/hash for 750Ti's and 200k/hash for 970  with built 56. Good work sp... )))

More is coming... If you want a higher hashrate, Please donate some beers.

7.05 MHASH per 750ti.

With djm34's changes we perhaps get 7,5MHASH per 750ti ??


https://ip.bitcointalk.org/?u=http%3A%2F%2Fi57.tinypic.com%2F30nd8hd.jpg&t=554&c=JlfEDP8Vdrvb6A

Hi SP beers here 0.1188--98a07c7e87ce4ec654cd2db0d58b68301c4ebb4c1439d34de029f13edad75b60

tyvm for u work

[pallas]

I had a look at the code; here is a quick fix to avoid mining when network fails:

diff --git a/ccminer.cpp b/ccminer.cpp
index c13d617..6c11757 100644
--- a/ccminer.cpp
+++ b/ccminer.cpp
@@ -215,6 +215,7 @@ bool autotune = true;
 bool opt_autotune = true;
 
 bool abort_flag = false;
+bool network_fail_flag = false;
 char *jane_params = NULL;
 
 char *rpc_user = NULL;
@@ -1356,7 +1357,7 @@ static void *miner_thread(void *userdata)
                pthread_mutex_unlock(&g_work_lock);
 
                /* prevent gpu scans before a job is received */
-               if (have_stratum && loopcnt>0 && work.data[0] == 0)
+               if (have_stratum && ((loopcnt>0 && work.data[0] == 0) || network_fail_flag))
                {
                        sleep(1);
                        continue;       
@@ -1903,6 +1904,7 @@ static void *stratum_thread(void *userdata)
                                if (!opt_benchmark)
                                        applog(LOG_ERR, "...retry after %d seconds", opt_fail_pause);
                                sleep(opt_fail_pause);
+                               network_fail_flag = true;
                        }
                }
 
@@ -1913,6 +1915,7 @@ static void *stratum_thread(void *userdata)
                        g_work_time = time(NULL);
                        if (stratum.job.clean)
                        {
+                               network_fail_flag = false;
                                if (!opt_quiet)
                                        applog(LOG_BLUE, "%s %s block %d", short_url, algo_names[opt_algo],
                                                stratum.job.height);

I tested it by creating a REJECT iptable rule to the pool and it worked fine, even though it submitted a lot of expired jobs on resume (maybe the jobs should be cleaned, but it doesn't hurt....).
On a side note, I think the cpu-miner.c file is unused (maybe others as well?).

[djm34]

I had a look at the code; here is a quick fix to avoid mining when network fails:

diff --git a/ccminer.cpp b/ccminer.cpp
index c13d617..6c11757 100644
--- a/ccminer.cpp
+++ b/ccminer.cpp
@@ -215,6 +215,7 @@ bool autotune = true;
 bool opt_autotune = true;
 
 bool abort_flag = false;
+bool network_fail_flag = false;
 char *jane_params = NULL;
 
 char *rpc_user = NULL;
@@ -1356,7 +1357,7 @@ static void *miner_thread(void *userdata)
                pthread_mutex_unlock(&g_work_lock);
 
                /* prevent gpu scans before a job is received */
-               if (have_stratum && loopcnt>0 && work.data[0] == 0)
+               if (have_stratum && ((loopcnt>0 && work.data[0] == 0) || network_fail_flag))
                {
                        sleep(1);
                        continue;       
@@ -1903,6 +1904,7 @@ static void *stratum_thread(void *userdata)
                                if (!opt_benchmark)
                                        applog(LOG_ERR, "...retry after %d seconds", opt_fail_pause);
                                sleep(opt_fail_pause);
+                               network_fail_flag = true;
                        }
                }
 
@@ -1913,6 +1915,7 @@ static void *stratum_thread(void *userdata)
                        g_work_time = time(NULL);
                        if (stratum.job.clean)
                        {
+                               network_fail_flag = false;
                                if (!opt_quiet)
                                        applog(LOG_BLUE, "%s %s block %d", short_url, algo_names[opt_algo],
                                                stratum.job.height);

I tested it by creating a REJECT iptable rule to the pool and it worked fine, even though it submitted a lot of expired jobs on resume (maybe the jobs should be cleaned, but it doesn't hurt....).
On a side note, I think the cpu-miner.c file is unused (maybe others as well?).

yes, it isn't used, not sure why it is still there...

[Epsylon3]

its from me except you changed the code...

in mine i only use :
ccminer.cpp:1669:               if (have_stratum && work.data[0] == 0 && !opt_benchmark) {

loopcnt > 0 do the reverse of what you want