OFFICIAL CGMINER mining software thread for linux/win/osx/mips/arm/r-pi 4.11.0 - page 687.

jake262144

full member

Activity: 210

Merit: 100

Quote from: DeathAndTaxes on January 28, 2012, 04:38:33 PM

Nice workaround but give how crappy 2.5 is I see no reason to keep it...

Dependency hell, DAT.
2.6 is extremely crappy but necessary for running 7xxx cards.
2.5 and 2.4 are less crappy and required for 69xx cards.
2.1 is the best SDK version unless you limnit yourself to VLIW5 architecture cards.

What miners should really do is organize their GPUs in groups by required SDK version.
Mixing VLIW4 and VLIW5 cards is a waste of hash rate, adding GCN to the bunch - a criminal waste

-ck

legendary

Activity: 4088

Merit: 1631

Ruu \o/

Quote from: The00Dustin on January 28, 2012, 09:46:13 AM

I'm not using cgminer with any nvidia cards, but if it is supposed to support them, you might want to do this in your next update:

Quote from: ufasoft on January 28, 2012, 09:09:27 AM

Quote from: Schleicher on January 28, 2012, 01:51:43 AM

I have not tested Ufasoft or cgminer, but there's no problem with Phoenix.
(using Nvidia driver 285.62 with Win 7 64Bit)

Edit: Ufasoft 64Bit doesn't see the graphics card, 32Bit is working fine.

Fixed, please redownload Windows binary.

Really it seems a bug in CUDA. by some reason x64 CUDA incorrectly interprets following construct:

Code:

__constant uint L = 0x98c7e2a2;

Replaced to :

Code:

__constant uint L[1] = { 0x98c7e2a2 };

And it works now!

Note that a few pots up someone points out that that nvidia driver version breaks it in cgminer as well, implying that it is supposed to support them, and I don't know if this is Win only or not.

Hmm thanks for the link, but that doesn't even appear in my code anywhere. Presumably that's ufa they're talking about. Meanwhile cgminer mines fine on my nvidia laptop so Huh

DeathAndTaxes

donator

Activity: 1218

Merit: 1079

Gerald Davis

Nice workaround but give how crappy 2.5 is I see no reason to keep it. Likely time will go by user forgets, installs new copy of cgminer (or different miner) and doesn't realize it is now re-using the worthless 2.5.

jake262144

full member

Activity: 210

Merit: 100

No need to remove anything.
Just copy the 2.1 SDK libraries into cgminer's directory and you're done!
You need the following files:

Code:

amdocl.dll
amdoclcl.dll
OpenCL.dll

A reboot might be necessary if Windows doesn't want to unload the previously-used 2.5 SDK libraries from memory.

DeathAndTaxes

donator

Activity: 1218

Merit: 1079

Gerald Davis

Quote from: BCMan on January 28, 2012, 02:59:57 PM

Quote from: jake262144 on January 28, 2012, 02:31:07 PM

Of course it is.
Add those to your config file:

Code:

"intensity" : "9,9,8,9",
"vectors" : "2",
"worksize" : "64",

Intensity is the same as Aggression in phoenix nomanclature. With cgminer you are use a per-card value as shown in the example code. Don't use values hight than 9 unless using a 7970 card.
You don't need to specify BFI_INT, it's on by default.

Also, RTFM - that info is in the readme file and in the very first post of this thread.

Thanks, but how to change platform? I need 2.1 instead of 2.5 sdk (getting cpu bug with 2.5). 2.5 installed as platform=0 (used by cgminer by default), 2.1 at 1.

I don't think you can. Simple solution it to removed 2.5 SDK. Wink

BCMan

hero member

Activity: 535

Merit: 500

Quote from: jake262144 on January 28, 2012, 02:31:07 PM

Of course it is.
Add those to your config file:

Code:

"intensity" : "9,9,8,9",
"vectors" : "2",
"worksize" : "64",

Intensity is the same as Aggression in phoenix nomanclature. With cgminer you are use a per-card value as shown in the example code. Don't use values hight than 9 unless using a 7970 card.
You don't need to specify BFI_INT, it's on by default.

Also, RTFM - that info is in the readme file and in the very first post of this thread.

Thanks, but how to change platform? I need 2.1 instead of 2.5 sdk (getting cpu bug with 2.5). 2.5 installed as platform=0 (used by cgminer by default), 2.1 at 1.

jake262144

full member

Activity: 210

Merit: 100

Of course it is.
Add those to your config file:

Code:

"intensity" : "9,9,8,9",
"vectors" : "2",
"worksize" : "64",

Intensity is the same as Aggression in phoenix nomanclature. With cgminer you are use a per-card value as shown in the example code. Don't use values hight than 9 unless using a 7970 card.
You don't need to specify BFI_INT, it's on by default.

Also, RTFM - that info is in the readme file and in the very first post of this thread.

BCMan

hero member

Activity: 535

Merit: 500

It's possible to set additional phatk2 commands like PLATFORM=1 BFI_INT WORKSIZE=256 AGGRESSION=12 FASTLOOP=False, etc? I lose performance without them , with Phoenix I've 223 mhash/s, with cgminer - 216 mhash/s.

The00Dustin

hero member

Activity: 807

Merit: 500

I'm not using cgminer with any nvidia cards, but if it is supposed to support them, you might want to do this in your next update:

Quote from: ufasoft on January 28, 2012, 09:09:27 AM

Quote from: Schleicher on January 28, 2012, 01:51:43 AM

I have not tested Ufasoft or cgminer, but there's no problem with Phoenix.
(using Nvidia driver 285.62 with Win 7 64Bit)

Edit: Ufasoft 64Bit doesn't see the graphics card, 32Bit is working fine.

Fixed, please redownload Windows binary.

Really it seems a bug in CUDA. by some reason x64 CUDA incorrectly interprets following construct:

Code:

__constant uint L = 0x98c7e2a2;

Replaced to :

Code:

__constant uint L[1] = { 0x98c7e2a2 };

And it works now!

Note that a few pots up someone points out that that nvidia driver version breaks it in cgminer as well, implying that it is supposed to support them, and I don't know if this is Win only or not.

-ck

legendary

Activity: 4088

Merit: 1631

Ruu \o/

Quote from: freakfantom on January 28, 2012, 03:40:29 AM

Quote from: -ck on January 28, 2012, 01:13:51 AM

Okay, here is the first windows executable with PCI bus based reordering of GPU devices. Put it into an existing 2.1.2 installation.

Please test it out to see if it works and what it does to your advertised card order!

http://ck.kolivas.org/apps/cgminer/temp/cgminer.exe

Yes, it works fine on windows x64 3x6990. It seems to me the order of gpus is reversed, but I don't have any proof

Just want to confirm that new exe is working.

Thanks very much. Yes the order will often be reversed due to mobo+os+drivr discrepancies.. The order should now most accurately represent what is happening physically with the cards.

kano

legendary

Activity: 4634

Merit: 1851

Linux since 1997 RedHat 4

Quote from: jjiimm_64 on January 27, 2012, 11:18:32 PM

Quote from: sveetsnelda on January 27, 2012, 11:08:40 PM

Quote from: DeathAndTaxes on January 27, 2012, 10:49:21 PM

That was covered in the readme. Simple solution use intensity 9 (or possibly Cool

.

Pff. Do you think I read manuals? Wink

Yeah, I never saw that. The problem is that the hash rate on the 7970s seem to still increase significantly when you turn the intensity above 9. Other cards barely budge.

do I have to get older bin files to get the hashrate your getting ?

I have never used poclbm before, so I dont have any Sad

GPU 1: 54.0C 3188RPM 50% | 557.5/559.0Mh/s | 1100Mhz 1020Mhz 1.17V A:21 R:0 HW:0 U:7.15/m I: 9

edit: I set one of the cards to 10, getting only 1 Mh more then the 9's.

btw, gpu-vddc did not change the voltage on the card. were you able to change voltage.? and please someone answer the 'older bin' question?

The bin files will of course be related to the opencl version you were running at the time the bin file was generated.
So it's not really older or newer, the performance figures anyone quotes will also be directly related to the opencl version they were using to generate the bin.

freakfantom

newbie

Activity: 73

Merit: 0

Quote from: -ck on January 28, 2012, 01:13:51 AM

Okay, here is the first windows executable with PCI bus based reordering of GPU devices. Put it into an existing 2.1.2 installation.

Please test it out to see if it works and what it does to your advertised card order!

http://ck.kolivas.org/apps/cgminer/temp/cgminer.exe

Yes, it works fine on windows x64 3x6990. It seems to me the order of gpus is reversed, but I don't have any proof

Just want to confirm that new exe is working.

-ck

legendary

Activity: 4088

Merit: 1631

Ruu \o/

Okay, here is the first windows executable with PCI bus based reordering of GPU devices. Put it into an existing 2.1.2 installation.

Please test it out to see if it works and what it does to your advertised card order!

http://ck.kolivas.org/apps/cgminer/temp/cgminer.exe

Linux build works from "reorder" branch on git.

jjiimm_64

legendary

Activity: 1876

Merit: 1000

Quote from: sveetsnelda on January 27, 2012, 11:08:40 PM

Quote from: DeathAndTaxes on January 27, 2012, 10:49:21 PM

That was covered in the readme. Simple solution use intensity 9 (or possibly Cool

.

Pff. Do you think I read manuals? Wink

Yeah, I never saw that. The problem is that the hash rate on the 7970s seem to still increase significantly when you turn the intensity above 9. Other cards barely budge.

do I have to get older bin files to get the hashrate your getting ?

I have never used poclbm before, so I dont have any Sad

GPU 1: 54.0C 3188RPM 50% | 557.5/559.0Mh/s | 1100Mhz 1020Mhz 1.17V A:21 R:0 HW:0 U:7.15/m I: 9

edit: I set one of the cards to 10, getting only 1 Mh more then the 9's.

btw, gpu-vddc did not change the voltage on the card. were you able to change voltage.? and please someone answer the 'older bin' question?

sveetsnelda

hero member

Activity: 642

Merit: 500

Quote from: DeathAndTaxes on January 27, 2012, 10:49:21 PM

That was covered in the readme. Simple solution use intensity 9 (or possibly Cool

.

Pff. Do you think I read manuals? Wink

Yeah, I never saw that. The problem is that the hash rate on the 7970s seem to still increase significantly when you turn the intensity above 9. Other cards barely budge.

jjiimm_64

legendary

Activity: 1876

Merit: 1000

Quote from: DeathAndTaxes on January 27, 2012, 10:49:21 PM

Quote from: sveetsnelda on January 27, 2012, 08:37:27 PM

Turning intensity down to 9 makes CPU usage sink to almost nothing. Increasing it to 10 (or above) completely saturates 1 core...

That was covered in the readme. Simple solution use intensity 9 (or possibly Cool

.

tru, but doesn't 79xx like the 10 ?

I am about to try the poclbm for the first time with the 79....

edit: YAAAAAHHHH

Code:

GPU 0:	70.0C	1763RPM 	31% | 	526.1/519.4Mh/s | 	1025Mhz 	1.17V 	A:24	R:0	HW:0	U:5.48/m 	I: 7
GPU 1:	67.0C	1448RPM 	25% | 	526.2/519.7Mh/s | 	1025Mhz 	1.17V 	A:33	R:1	HW:0	U:7.54/m 	I: 7
GPU 2:	68.0C	1332RPM 	25% | 	526.2/519.7Mh/s | 	1025Mhz 	1.17V 	A:35	R:2	HW:0	U:8.00/m 	I: 7
GPU 3:	72.0C	1763RPM 	33% | 	526.3/519.8Mh/s | 	1025Mhz 	1.17V 	A:25	R:1	HW:0	U:5.71/m 	I: 7

edit2: looks like I need to add mem now to the webmon

DeathAndTaxes

donator

Activity: 1218

Merit: 1079

Gerald Davis

Quote from: sveetsnelda on January 27, 2012, 08:37:27 PM

Turning intensity down to 9 makes CPU usage sink to almost nothing. Increasing it to 10 (or above) completely saturates 1 core...

That was covered in the readme. Simple solution use intensity 9 (or possibly Cool

.

sveetsnelda

hero member

Activity: 642

Merit: 500

Turning intensity down to 9 makes CPU usage sink to almost nothing. Increasing it to 10 (or above) completely saturates 1 core...

sveetsnelda

hero member

Activity: 642

Merit: 500

(I had to use 13 for intensity to get this, BTW)

sveetsnelda

hero member

Activity: 642

Merit: 500

Good news and bad news, ckolivas:

All of my previous 7970 testing was getting bottlenecked by CPU usage. This is the reason why the hashrate fluctuated so heavily. I enabled the other 2 cores on this rig and I'm getting much better results. I usually leave 2 cores disabled and undervolt the CPU to save power.

Now that the other 2 cores are enabled, I can get around 633Mhash/sec with the poclbm kernel and the same clocks used in the other testing. I say "around 630Mhash/sec" because the hash rate bounces around quite a bit due to the CPU still being bottlenecked. If I only run CGMiner against 1 card, the hash rate starts around 610Mhash, gradually increases to 633, falls down again, etc etc. Running CGMiner like this averaged 612Mhash/sec over 5 minutes.

I'll toy around with Phatk and the modified version you posted, and then report back.

Topic: OFFICIAL CGMINER mining software thread for linux/win/osx/mips/arm/r-pi 4.11.0 - page 687. (Read 5805728 times)