Pages:
Author

Topic: ethminer-0.9.41-genoil-1.1 - page 17. (Read 397363 times)

legendary
Activity: 1281
Merit: 1003
September 23, 2016, 02:52:56 PM
what speed you have on the rx470?  4gigas version?

21.9MH tops on stock clocks/bios (1260/1750), but it tends to fluctuate a lot more and averages out at about 21.3, whereas my kernel does a fairly steady 20.9 when using a high farm-recheck.

There's something wrong with my dropping his kernel into my code though, because it doesn't seem to find any shares Grin.   
Shocked only 21  so not much better than the 380, i sow somewhere people saying about 25mh
full member
Activity: 162
Merit: 100
September 23, 2016, 10:19:21 AM
i like the 119nr Smiley)
sr. member
Activity: 438
Merit: 250
September 23, 2016, 01:19:01 AM
what speed you have on the rx470?  4gigas version?

21.9MH tops on stock clocks/bios (1260/1750), but it tends to fluctuate a lot more and averages out at about 21.3, whereas my kernel does a fairly steady 20.9 when using a high farm-recheck.

There's something wrong with my dropping his kernel into my code though, because it doesn't seem to find any shares Grin.   
legendary
Activity: 1797
Merit: 1028
September 22, 2016, 06:44:43 PM
Also briefly loaded the wolf0 4-way kernel that he ships with sgminer-gm into this. Performance on RX470 seems just a bit better.

PLEASE EXPLAIN THE "4-WAY KERNEL"--

You may share things with developers, but please make a simple explanation as to what "4-way" means.  I've tried the sgminer that was released, it runs better on one of my 280X rigs and is more stable.  Likely, I will convert it to a RX 470 rig, but I'd like to know more about the software that I am running.      --scryptr

When hashing, a GPU uses many threads in parallel, where basically each thread tries to solve the same puzzle with a slightly different input (the nonce, a number that increases per adjacent thread). Some parts of some algorithms can be made to run faster when groups of threads temporarily join forces to do a bit of the work. You can compare it with hauling heavy boxes. If 4 people have to haul 4 very heavy boxes from A to B, they may actually be faster when carrying each box with 4 people at the same time compared to when they would only carry their own box.

Ethash was originally designed to haul boxes with 8 people at the same time (8-way) while Wolf0 made a 4-way variant. Of course Wolf's 4 threads have to do the same amount of work as the original kernel, but it can be a little bit more efficient. Ultimately (I think) he did this to prepare for his private ETH kernel, because that kernel uses a different way of sharing the work (coordinating the joint box hauling operation) between 4 adjacent threads, and 8 isn't supported.

If I recall well, Etar from the Etarminer (CUDA ETH miner) has a 16-way solution and who knows what Claymore did Smiley

8 can be supported under Tonga and up, but not for all GCN.

ETARMINER--

Well, I checked it out, but I did not download the miner.  He wrote for Kepler vintage CUDA cards, and there is a built-in 1% fee.  The miner works on newer CUDA cards with varying success.  Etar, the programmer, seems very sensitive.  I think he has a GTX 660, or something....       --scryptr
legendary
Activity: 1281
Merit: 1003
September 22, 2016, 05:47:09 PM
what speed you have on the rx470?  4gigas version?
legendary
Activity: 3248
Merit: 1070
September 22, 2016, 12:46:45 PM
Also briefly loaded the wolf0 4-way kernel that he ships with sgminer-gm into this. Performance on RX470 seems just a bit better.

PLEASE EXPLAIN THE "4-WAY KERNEL"--

You may share things with developers, but please make a simple explanation as to what "4-way" means.  I've tried the sgminer that was released, it runs better on one of my 280X rigs and is more stable.  Likely, I will convert it to a RX 470 rig, but I'd like to know more about the software that I am running.      --scryptr

When hashing, a GPU uses many threads in parallel, where basically each thread tries to solve the same puzzle with a slightly different input (the nonce, a number that increases per adjacent thread). Some parts of some algorithms can be made to run faster when groups of threads temporarily join forces to do a bit of the work. You can compare it with hauling heavy boxes. If 4 people have to haul 4 very heavy boxes from A to B, they may actually be faster when carrying each box with 4 people at the same time compared to when they would only carry their own box.

Ethash was originally designed to haul boxes with 8 people at the same time (8-way) while Wolf0 made a 4-way variant. Of course Wolf's 4 threads have to do the same amount of work as the original kernel, but it can be a little bit more efficient. Ultimately (I think) he did this to prepare for his private ETH kernel, because that kernel uses a different way of sharing the work (coordinating the joint box hauling operation) between 4 adjacent threads, and 8 isn't supported.

If I recall well, Etar from the Etarminer (CUDA ETH miner) has a 16-way solution and who knows what Claymore did Smiley

isn't this something like a single core vs quad core?
legendary
Activity: 1797
Merit: 1028
September 22, 2016, 09:20:26 AM
Also briefly loaded the wolf0 4-way kernel that he ships with sgminer-gm into this. Performance on RX470 seems just a bit better.

PLEASE EXPLAIN THE "4-WAY KERNEL"--

You may share things with developers, but please make a simple explanation as to what "4-way" means.  I've tried the sgminer that was released, it runs better on one of my 280X rigs and is more stable.  Likely, I will convert it to a RX 470 rig, but I'd like to know more about the software that I am running.      --scryptr

When hashing, a GPU uses many threads in parallel, where basically each thread tries to solve the same puzzle with a slightly different input (the nonce, a number that increases per adjacent thread). Some parts of some algorithms can be made to run faster when groups of threads temporarily join forces to do a bit of the work. You can compare it with hauling heavy boxes. If 4 people have to haul 4 very heavy boxes from A to B, they may actually be faster when carrying each box with 4 people at the same time compared to when they would only carry their own box.

Ethash was originally designed to haul boxes with 8 people at the same time (8-way) while Wolf0 made a 4-way variant. Of course Wolf's 4 threads have to do the same amount of work as the original kernel, but it can be a little bit more efficient. Ultimately (I think) he did this to prepare for his private ETH kernel, because that kernel uses a different way of sharing the work (coordinating the joint box hauling operation) between 4 adjacent threads, and 8 isn't supported.

If I recall well, Etar from the Etarminer (CUDA ETH miner) has a 16-way solution and who knows what Claymore did Smiley

WOW!  THANKS, I AM TAKING NOTES--

Hopefully, I will be able to put this to good use.  I certainly learned something.  I will look up this "Etarminer", also.  Thank you!       --scryptr
sr. member
Activity: 438
Merit: 250
September 22, 2016, 06:15:48 AM
Also briefly loaded the wolf0 4-way kernel that he ships with sgminer-gm into this. Performance on RX470 seems just a bit better.

PLEASE EXPLAIN THE "4-WAY KERNEL"--

You may share things with developers, but please make a simple explanation as to what "4-way" means.  I've tried the sgminer that was released, it runs better on one of my 280X rigs and is more stable.  Likely, I will convert it to a RX 470 rig, but I'd like to know more about the software that I am running.      --scryptr

When hashing, a GPU uses many threads in parallel, where basically each thread tries to solve the same puzzle with a slightly different input (the nonce, a number that increases per adjacent thread). Some parts of some algorithms can be made to run faster when groups of threads temporarily join forces to do a bit of the work. You can compare it with hauling heavy boxes. If 4 people have to haul 4 very heavy boxes from A to B, they may actually be faster when carrying each box with 4 people at the same time compared to when they would only carry their own box.

Ethash was originally designed to haul boxes with 8 people at the same time (8-way) while Wolf0 made a 4-way variant. Of course Wolf's 4 threads have to do the same amount of work as the original kernel, but it can be a little bit more efficient. Ultimately (I think) he did this to prepare for his private ETH kernel, because that kernel uses a different way of sharing the work (coordinating the joint box hauling operation) between 4 adjacent threads, and 8 isn't supported.

If I recall well, Etar from the Etarminer (CUDA ETH miner) has a 16-way solution and who knows what Claymore did Smiley
legendary
Activity: 1797
Merit: 1028
September 22, 2016, 02:12:10 AM
Also briefly loaded the wolf0 4-way kernel that he ships with sgminer-gm into this. Performance on RX470 seems just a bit better.

PLEASE EXPLAIN THE "4-WAY KERNEL"--

You may share things with developers, but please make a simple explanation as to what "4-way" means.  I've tried the sgminer that was released, it runs better on one of my 280X rigs and is more stable.  Likely, I will convert it to a RX 470 rig, but I'd like to know more about the software that I am running.      --scryptr
sr. member
Activity: 438
Merit: 250
September 21, 2016, 11:46:45 AM
Also briefly loaded the wolf0 4-way kernel that he ships with sgminer-gm into this. Performance on RX470 seems just a bit better.
sr. member
Activity: 588
Merit: 251
September 21, 2016, 09:45:34 AM
1.1.9 is out under the 110 branch. Meanwhile, promoted 1.1.7 to stable in the master branch. If you like DAG files, you can still find the old version under the 108 branch.

1.1.9 is mostly the work of others, notably @nerdralph. Lots of code cleanups, compilation fixes and some tweaks to the console log output, including per-GPU hashrate display. -SC option is gone, ethminer now defaults to the synchronous V2 client.

Executable is built with CUDA 8.0RC, don't have a CUDA GPU momentarily so I hope it works. 

Thanks for the plug.  I see you added a couple tweaks of your own.  Adding the DAG size to the creating time is nice.
Code:
GPU0  1.59GB of DAG data generated in 4448 ms.
sr. member
Activity: 438
Merit: 250
September 21, 2016, 08:53:40 AM
1.1.9 is out under the 110 branch. Meanwhile, promoted 1.1.7 to stable in the master branch. If you like DAG files, you can still find the old version under the 108 branch.

1.1.9 is mostly the work of others, notably @nerdralph. Lots of code cleanups, compilation fixes and some tweaks to the console log output, including per-GPU hashrate display. -SC option is gone, ethminer now defaults to the synchronous V2 client.

Executable is built with CUDA 8.0RC, don't have a CUDA GPU momentarily so I hope it works. 
legendary
Activity: 1281
Merit: 1003
September 20, 2016, 08:36:34 PM
ok but it s fine too under windows and 4gigas

i m under win7 & 10  with 4 gigas it s run ok
just need a bigger pagefile otherwise it s hang on dag load

few weeks ago some miners where unstable specially on eth less on etc
but now they running fine , no crash last week


legendary
Activity: 1797
Merit: 1028
September 19, 2016, 11:01:30 AM
I have in one linux rig only 2gb of RAM and it shows only 500mb use
I have there only 2cards but I think 1gb a card is maximum neccesary Smiley

LINUX RUNS IN 2GB MAX RAM--

A proper Linux setup can run in 1GB RAM, if you can find 512MB DDR3 RAM sticks.  All of my Linux/AMD rigs run stable with 2GB RAM, 4 and 6 cards each.         --scryptr
hero member
Activity: 1246
Merit: 708
September 18, 2016, 11:08:26 PM
I have in one linux rig only 2gb of RAM and it shows only 500mb use
I have there only 2cards but I think 1gb a card is maximum neccesary Smiley
hero member
Activity: 710
Merit: 502
September 18, 2016, 10:51:01 PM
THE COMMAND IN WINDOWS IS "SETX"--

The environment variables in Windows are set with "setx GPU_USE_SYNC_OBJECTS 1".  There is no equals sign "=" in the Windows syntax.  The equals sign is used in Linux, and "EXPORT" is used in Linux rather than "setx".  The user has to provide the correct syntax for his operating system.       --scryptr


Thanx for reply, edit to SETX, but the same problem.

How much ram do you have?, for 5 of those cards i will use AT LEAST 12GB, and use a 24GB paging file in windows.
try removing all the absolutely not necessary options (--farm-recheck 200 --cl-local-work 256 --cl-global-work 16384 -SP1 ? -SC 1?), start with the minimum necessary and build from there.

are you sure all the cards are detected properly?, can you see them all active in the MSI Afterburner?

I only use:
setx GPU_MAX_ALLOC_PERCENT 100
setx GPU_USE_SYNC_OBJECTS 1

and add a:
timeout 120

before starting the miner, because i have a slow hard drive in that miner, so i need to give it time to load.....whatever it loads  Grin or it will fuck the miner up.

from where did you get  that you need 12giga of memory for 5 cards ??
you wrong 4 giga is enough, my miners have 4giga of ram only and ruining 5 - 6 gpu's

In windows I needed a lot of ram to prevent hangups or program crashes, 8GB did the trick with 4 cards, maybe it is not a requirement with latest versions, but with version 107 with 4GB freezes, need 8GB minimum, 5 cards i needed another 4GB stick.
in the linux boxes i always use 8GB anyway but never tested with less
hero member
Activity: 710
Merit: 502
September 18, 2016, 10:48:39 PM
How many hashes I should expect from a single 7950 GPU ? I want to mine ETH and ETC... please help me

I am getting about 14.5MH/s from a Powercolor HD7950.
same here, I am getting 14 MH with a sapphire 7950 dualx oc flex (stock clock)
and the R9 280X MSI is around 16.5 MH stock clock also.
legendary
Activity: 1281
Merit: 1003
September 18, 2016, 06:45:49 PM
THE COMMAND IN WINDOWS IS "SETX"--

The environment variables in Windows are set with "setx GPU_USE_SYNC_OBJECTS 1".  There is no equals sign "=" in the Windows syntax.  The equals sign is used in Linux, and "EXPORT" is used in Linux rather than "setx".  The user has to provide the correct syntax for his operating system.       --scryptr


Thanx for reply, edit to SETX, but the same problem.

How much ram do you have?, for 5 of those cards i will use AT LEAST 12GB, and use a 24GB paging file in windows.
try removing all the absolutely not necessary options (--farm-recheck 200 --cl-local-work 256 --cl-global-work 16384 -SP1 ? -SC 1?), start with the minimum necessary and build from there.

are you sure all the cards are detected properly?, can you see them all active in the MSI Afterburner?

I only use:
setx GPU_MAX_ALLOC_PERCENT 100
setx GPU_USE_SYNC_OBJECTS 1

and add a:
timeout 120

before starting the miner, because i have a slow hard drive in that miner, so i need to give it time to load.....whatever it loads  Grin or it will fuck the miner up.

from where did you get  that you need 12giga of memory for 5 cards ??
you wrong 4 giga is enough, my miners have 4giga of ram only and ruining 5 - 6 gpu's
legendary
Activity: 1281
Merit: 1003
September 18, 2016, 06:42:54 PM
How many hashes I should expect from a single 7950 GPU ? I want to mine ETH and ETC... please help me

I am getting about 14.5MH/s from a Powercolor HD7950.

that's a little better than r9 270
newbie
Activity: 59
Merit: 0
September 18, 2016, 07:34:51 AM
How many hashes I should expect from a single 7950 GPU ? I want to mine ETH and ETC... please help me

I am getting about 14.5MH/s from a Powercolor HD7950.
Pages:
Jump to: