Gateless Gate Sharp 1.3.8: 30Mh/s (Ethash) on RX 480!

[laik2]

Well looks like the linux build?
I couldn't find the windows build on the release page ... no he mentioned it that way.
his last post sounds like its not yet done.

Anyway good to know it works for you

True, its linux. I don't have windows machine to test but it should work too.

[1714883188]

1714883188

[1714883188]

1714883188

[1714883188]

1714883188

[1714883188]

1714883188

[ioglnx]

To bad i don't have cygwin installed anymore. Since  Silentarmy lost against the cuda miner for my nvidia i removed all that stuff from system

[FFI2013]

Nope, not yet. Almost there, though.

for some reason Hawaii is only getting a bin in GC and I tried genoil eth miner it sees all gpus put only build a dag or bin in GC for my hawaii do you think I should reinstall drivers

Have you tried this?

http://forum.ethereum.org/discussion/2737/some-cards-need-export-gpu-max-alloc-percent-95

yes I dont understand why Im having trouble with just eth the weird thing is I uninstalled drivers with amd and DD reinstalled one card at a time and its still choosing hawaii and hawaii is not my main card my 270 is both my 290 and 390 are on 1x maybe its time to go back to win7 I never had any trouble with win 7 win 10 just updates drives whenever it wants without asking unless u do bunch of extra crap thanks though I no your busy just didnt no if you had any advice

[zawawa]

I suppose I wasn't very clear. Support for PASC is already in the repo.
(I just ported the suprnova stuff. There were some really nasty gotchas.)
I will prepare a next point release with a precompiled Windows binary today or tomorrow.

I'm building llvm now. So far, so good...

[ioglnx]

Can you tell me witch gotchas was it in the cl file?

[ioglnx]

Could someone please be so kind to provide a windows build?

[joaocha]

Could someone please be so kind to provide a windows build?

Just calm down, it ill be ready very soon

[ioglnx]

What calm down? I didn't even get bad or unpatient. I just asked..maybe if you get cold water its hot for you right?

[rjg]

Could someone please be so kind to provide a windows build?

Please read 2 posts above yours...

[kilo17]

zawawa - PM sent 

[zawawa]

Looks like liblcl, the OpenCL library for LLVM, is a major pain to build as well.
I just hope the inline assembly would really work...

[zawawa]

Here is the PASC support:

https://github.com/zawawawa/gatelessgate/releases/tag/v0.1.3-pre1

I'm still working on the assembly version. Enjoy!

[reb0rn21]

Here is the PASC support:

https://github.com/zawawawa/gatelessgate/releases/tag/v0.1.3-pre1

I'm still working on the assembly version. Enjoy!

can you fix CPU usage on Nvidia and pascal cards in opencl?

there is no way to run 5 cards rig on any CPU, each card use 30% from i5 CPU

NV 1070 works fine with -w 1024, speed is 1Ghs, but as I sad CPU usage is problem with opencl and NV

zcash did not work on NV

ethash works fine with gpu thread 1, but only HW errors, need to play with settings

[toptek]

my 470 4 gb is getting 680 mh Pascal mining, does that sound about right ? . with  Gateless

[twan69666]

Zawawa, I've always been a lurker, but had to register to say thank you! Very flexible miner and great work you're doing.

I'm an AMD guy, so I just sent you a few $ as a thank you. Keep up the great work!

[zawawa]

Here is the PASC support:

https://github.com/zawawawa/gatelessgate/releases/tag/v0.1.3-pre1

I'm still working on the assembly version. Enjoy!

can you fix CPU usage on Nvidia and pascal cards in opencl?

there is no way to run 5 cards rig on any CPU, each card use 30% from i5 CPU

NV 1070 works fine with -w 1024, speed is 1Ghs, but as I sad CPU usage is problem with opencl and NV

zcash did not work on NV

ethash works fine with gpu thread 1, but only HW errors, need to play with settings

That's right. I totally forgot about the NVIDIA CPU usage problem.
I will look into a permanent solution.

[zawawa]

my 470 4 gb is getting 680 mh Pascal mining, does that sound about right ? . with  Gateless

That looks about right.

[zawawa]

Zawawa, I've always been a lurker, but had to register to say thank you! Very flexible miner and great work you're doing.

I'm an AMD guy, so I just sent you a few $ as a thank you. Keep up the great work!

Thanks! It's still a work in progress, but I'm glad you liked it.

[nerdralph]

I've been crunching more numbers on the GDS performance, and I think for GPUs with more than 4 memory controllers (like Tahiti & Hawaii) that the best performance may be from splitting the row counters between L2 and GDS. With all the row counters in GDS, the GDS bandwidth should get maxed out before the core/L2 bandwidth limit is reached.

[zawawa]

I've been crunching more numbers on the GDS performance, and I think for GPUs with more than 4 memory controllers (like Tahiti & Hawaii) that the best performance may be from splitting the row counters between L2 and GDS. With all the row counters in GDS, the GDS bandwidth should get maxed out before the core/L2 bandwidth limit is reached.

That recommendation is consistent with my somewhat informal observations. Thank you so much!
Just to let you know, I was able to link the object file of the kernel with libclc. The only remaining step is to convert one big LLVM Bit Code (.bc) file into an OpenCL binary. We will see.