Pages:
Author

Topic: Genesis Mining Presents: SGMiner-GM - now with Zawawa's GG! [Updated 17/01/2017] - page 36. (Read 140422 times)

newbie
Activity: 49
Merit: 0
worksize: 192
rawintensity (rx 470): 1916928
rawintensity (rx 480): 2156544

You should try different intensities to optimize your hash rate

Thanks for response!
I'm under Windows and with above settings I have 9.8 mhs only (with Claymore: 22 mhs) per card ;(

This is what I put in the config file:

Code:
    "profiles": [{
        "worksize": "192",
        "name": "eth",
        "algorithm": "ethash",
        "gpu-threads": "2",
        "rawintensity": "1916928"
    }],
    "no-extranonce": "true",
    "default-profile": "eth"

Any thoughts?
TIA

I have no windows rig for testing, but I guess that 'gpu-threads: 2' might be the problem. I would start at 'gpu-threads: 1' and 'xintensity: 4620' because I'm not sure whether sgminer accepts rawintensity much larger than 2,000,000.
newbie
Activity: 50
Merit: 0
worksize: 192
rawintensity (rx 470): 1916928
rawintensity (rx 480): 2156544

You should try different intensities to optimize your hash rate

Thanks for response!
I'm under Windows and with above settings I have 9.8 mhs only (with Claymore: 22 mhs) per card ;(

This is what I put in the config file:

Code:
    "profiles": [{
        "worksize": "192",
        "name": "eth",
        "algorithm": "ethash",
        "gpu-threads": "2",
        "rawintensity": "1916928"
    }],
    "no-extranonce": "true",
    "default-profile": "eth"

Any thoughts?
TIA
newbie
Activity: 49
Merit: 0
Is here anyone who knows how to use it w/ Rx 480/470?



for me with 480s it resets the system
... the only problem is that the driver reports a wrong number of CUs. Therefore, you have to use a larger xintensity or just use rawintensity and the hashrate for those GPUs can be increased by using 2 threads per GPU.

Could you share proper values of xintensity or rawintensity for 480/470?
And is worksize of 256 correct?
TIA
worksize: 192
rawintensity (rx 470): 1916928
rawintensity (rx 480): 2156544

You should try different intensities to optimize your hash rate
newbie
Activity: 50
Merit: 0
Is here anyone who knows how to use it w/ Rx 480/470?



for me with 480s it resets the system
... the only problem is that the driver reports a wrong number of CUs. Therefore, you have to use a larger xintensity or just use rawintensity and the hashrate for those GPUs can be increased by using 2 threads per GPU.

Could you share proper values of xintensity or rawintensity for 480/470?
And is worksize of 256 correct?
TIA
newbie
Activity: 49
Merit: 0
Is here anyone who knows how to use it w/ Rx 480/470?



for me with 480s it resets the system
Could you give a little bit more details, please? RX 470/480 work without problems in combination with AMDGPU-PRO and the only problem is that the driver reports a wrong number of CUs. Therefore, you have to use a larger xintensity or just use rawintensity and the hashrate for those GPUs can be increased by using 2 threads per GPU. Of course, ADL is not supported by AMDGPU-PRO.
newbie
Activity: 50
Merit: 0
sr. member
Activity: 560
Merit: 250
Is here anyone who knows how to use it w/ Rx 480/470?



for me with 480s it resets the system
newbie
Activity: 50
Merit: 0
Is here anyone who knows how to use it w/ Rx 480/470?
newbie
Activity: 49
Merit: 0
hey guys whats the most stable combination of miner software on dwarfpool ? i'm having lots of stuck rigs right now...

Hi, I think that the best possibility is to use a stratum proxy to connect to dwarfpool at the moment because dwarfpool's stratum implementation isn't supported by sgminer. The reason for that are the weird JSON replies of dwarfpool, which make it impossible to decide whether requests of sgminer were successful or not and sgminer assumes that its requests failed. Perhaps the operators of dwarfpool should be notified so that the issue can be fixed.
sr. member
Activity: 588
Merit: 251
Nice 4-way dagger loop. I might steal this  Grin

I noticed that too.  Thought it would be faster than 8-way on Pitcairn, but it's actually slower (~10%).  Didn't dig any deeper to check VGPR use or try changing the unroll pragmas...

CodeXL says on Ellesmere (Polaris), Tonga and Hawaii it does only 2 waves, Tahiti and Pitcairn 3 waves. But who cares, apparently it's faster  Grin

I used the linux binary distribution, which has pre-built kernels for worksize 64 and 192.  So I used worksize 128 to force it to use the opencl kernel source.   On Linux/fglrx it's 3 waves, so I'm not sure how much you should trust CodeXL.

Code:
_temp_0_Hawaii_GenerateDAG.isa:NumVgprs             = 100;
_temp_0_Hawaii_search.isa:NumVgprs             = 83;
_temp_1_Pitcairn_GenerateDAG.isa:NumVgprs             = 100;
_temp_1_Pitcairn_search.isa:NumVgprs             = 83;
_temp_2_Tonga_GenerateDAG.isa:NumVgprs             = 100;
_temp_2_Tonga_search.isa:NumVgprs             = 83;
sr. member
Activity: 438
Merit: 250
I poked around a little bit to also get Ellesmere/Tonga/Hawaii to 3 waves as well. Ultimately you only need to change line 255 into #pragma unroll 1 (saves 11 VGPR?!). But I don't know what effect this has. Really need to get me one of those cards..
legendary
Activity: 1050
Merit: 1293
Huh?
Nice 4-way dagger loop. I might steal this  Grin

I noticed that too.  Thought it would be faster than 8-way on Pitcairn, but it's actually slower (~10%).  Didn't dig any deeper to check VGPR use or try changing the unroll pragmas...

CodeXL says on Ellesmere (Polaris), Tonga and Hawaii it does only 2 waves, Tahiti and Pitcairn 3 waves. But who cares, apparently it's faster  Grin

Mm, weird though, it's not faster on Ellesmere but it is on Hawaii..

perhaps i need to do some more tests i guess Smiley
sr. member
Activity: 438
Merit: 250
Nice 4-way dagger loop. I might steal this  Grin

I noticed that too.  Thought it would be faster than 8-way on Pitcairn, but it's actually slower (~10%).  Didn't dig any deeper to check VGPR use or try changing the unroll pragmas...

CodeXL says on Ellesmere (Polaris), Tonga and Hawaii it does only 2 waves, Tahiti and Pitcairn 3 waves. But who cares, apparently it's faster  Grin
full member
Activity: 199
Merit: 108
Look, I'm really not that interesting. Promise.
Who is mad?
Me nope I can live without your miner..no need to be mad.

Not you mate, user peteycamey. Sorry, the quote didn't tag him.
sr. member
Activity: 588
Merit: 251
Nice 4-way dagger loop. I might steal this  Grin

I noticed that too.  Thought it would be faster than 8-way on Pitcairn, but it's actually slower (~10%).  Didn't dig any deeper to check VGPR use or try changing the unroll pragmas...
sr. member
Activity: 438
Merit: 250
Gave it a spin on a Tahiti (7950). Performance seems just about equal as my 1.1.7. Hard to really compare because of the different way of measuring. I'll rip the kernel out and dump it into ethminer to see if there's a noticeable difference.

One minor issue though is that the fans don't come back down to normal RPM when I exit the program (windows). 
sr. member
Activity: 574
Merit: 250
Fighting mob law and inquisition in this forum
Who is mad?
Me nope I can live without your miner..no need to be mad.
full member
Activity: 199
Merit: 108
Look, I'm really not that interesting. Promise.


Aye, the kernels aren't included. They were sold exclusively, after all. We may make a public kernel in the future.

But still - there is no nvidia support even nvidia does OpenCL.
They using AMD APP Framework for detection of OpenCL Device..so any word from the dev?=


He's mad because we won't sell him the private Ethereum kernel.

No means no, 'petey'.
hero member
Activity: 906
Merit: 507
claymores miner have dual-mining capability by that it justifies the fee.

sgminer-gm performs worse than genoils miner which has no fee either.

either make the kernel public or get lost.

Dual mining, no problem, start sgminer, start another instance with another algo, yup, you're dual mining..

performs worse? Nah! On Hawaii even better! Just gotta set the correct intensity and work size.
what do u recommend for worksize and intensity for Hawaii
member
Activity: 81
Merit: 1002
It was only the wind.


Aye, the kernels aren't included. They were sold exclusively, after all. We may make a public kernel in the future.

But still - there is no nvidia support even nvidia does OpenCL.
They using AMD APP Framework for detection of OpenCL Device..so any word from the dev?=


MY PERSONAL kernels weren't - there's an ethash.cl in there, if you'd look.
Pages:
Jump to: