Genesis Mining Presents: SGMiner-GM - now with Zawawa's GG! [Updated 17/01/2017] - page 31.

nerdralph

sr. member

Activity: 588

Merit: 251

Looks like there might be a bug handling the DAG change:

Code:

[23:55:33] Error -2: Setting args for the DAG kernel and/or executing it.
[23:55:33] Error: clSetKernelArg of all params failed.
[23:55:33] GPU 0 failure, disabling!

I hit q(uit), started sgminer-gm, and it is once again working fine. I was testing with a single GPU (290x).

Kevinatin

sr. member

Activity: 612

Merit: 250

Quote from: scryptr on September 30, 2016, 07:01:55 AM

Quote from: nerdralph on September 29, 2016, 04:10:03 PM

Quote from: scryptr on September 29, 2016, 11:08:05 AM

PROGRAMMERS CAN WORK MIRACLES--

But all I have heard for a year is that there is not much juice to squeeze from the DAG. Generating it on-the-fly was a big thing, saving start-up minutes.

Likely, card optimization will be the way to go. 29MH/s on the RX 480 is top speed reported. Newcards (480x, 490, etc) are still top secret, not expected until next year. I search, and se the same old tech blogs. --scryptr

Unless you can find a flaw in the ethash algorithm, it is impossible to get more than 32Mh/s from a card like the RX480 with 256GB/s of memory bandwidth (i.e. memory clocked at 2Ghz). This is because each hash requires 8MB of random memory reads.

THANK YOU FOR THE EXPLANATION--

I think that you have explained the theoretical maximum hashrate for an RX 480. Looking at posted data from various sources, a well tuned RX 480 Ethereum rig will mine at 29MH/s, and an RX 470 rig will mine at 27MH/s per card. It may take BIOS mods and a carefully adjusted OS to get there. Out of the box, the cards will mine at 22-24MH/s completely stock. RX 460 cards will reportedly mine at 10-11MH/s stock, similar to the performance of a previous generation R7 370 card, but at as little as 1/2 the wattage.

I have seen posts at higher hash rates for single-card rigs running a short duration, but none more than the theoretical maximum as you explained it.

The next generation of cards (Vega) may be a game changer, or may not. The top cards will be using a new memory type, and that does not always translate to better mining performance. I keep looking for new information, but it has not surfaced yet. --scryptr

That is true. that is the reason why the R9 390 is faster than the RX 480 as its memory bandwidth is double that of the RX 480 if running at the same memory speed.

sammy007

legendary

Activity: 1904

Merit: 1003

Is it possible to make this miner work with https://github.com/sammy007/open-ethereum-pool/blob/master/docs/STRATUM.md in near future?

scryptr

legendary

Activity: 1797

Merit: 1028

Quote from: nerdralph on September 29, 2016, 04:10:03 PM

Quote from: scryptr on September 29, 2016, 11:08:05 AM

PROGRAMMERS CAN WORK MIRACLES--

But all I have heard for a year is that there is not much juice to squeeze from the DAG. Generating it on-the-fly was a big thing, saving start-up minutes.

Likely, card optimization will be the way to go. 29MH/s on the RX 480 is top speed reported. Newcards (480x, 490, etc) are still top secret, not expected until next year. I search, and se the same old tech blogs. --scryptr

Unless you can find a flaw in the ethash algorithm, it is impossible to get more than 32Mh/s from a card like the RX480 with 256GB/s of memory bandwidth (i.e. memory clocked at 2Ghz). This is because each hash requires 8MB of random memory reads.

THANK YOU FOR THE EXPLANATION--

I think that you have explained the theoretical maximum hashrate for an RX 480. Looking at posted data from various sources, a well tuned RX 480 Ethereum rig will mine at 29MH/s, and an RX 470 rig will mine at 27MH/s per card. It may take BIOS mods and a carefully adjusted OS to get there. Out of the box, the cards will mine at 22-24MH/s completely stock. RX 460 cards will reportedly mine at 10-11MH/s stock, similar to the performance of a previous generation R7 370 card, but at as little as 1/2 the wattage.

I have seen posts at higher hash rates for single-card rigs running a short duration, but none more than the theoretical maximum as you explained it.

The next generation of cards (Vega) may be a game changer, or may not. The top cards will be using a new memory type, and that does not always translate to better mining performance. I keep looking for new information, but it has not surfaced yet. --scryptr

nerdralph

sr. member

Activity: 588

Merit: 251

Quote from: scryptr on September 29, 2016, 11:08:05 AM

PROGRAMMERS CAN WORK MIRACLES--

But all I have heard for a year is that there is not much juice to squeeze from the DAG. Generating it on-the-fly was a big thing, saving start-up minutes.

Likely, card optimization will be the way to go. 29MH/s on the RX 480 is top speed reported. Newcards (480x, 490, etc) are still top secret, not expected until next year. I search, and se the same old tech blogs. --scryptr

Unless you can find a flaw in the ethash algorithm, it is impossible to get more than 32Mh/s from a card like the RX480 with 256GB/s of memory bandwidth (i.e. memory clocked at 2Ghz). This is because each hash requires 8MB of random memory reads.

Inkdatar

hero member

Activity: 1582

Merit: 523

Quote from: yoyo1 on September 29, 2016, 07:36:13 AM

According to:
RTFM
https://github.com/genesismining/sgminer-gm/blob/master/doc/configuration.md

Thank you for the guide.

I'd like to summarize some findings:
There are several ways of loading GPU via manipulating following parameters:
*xintensity* put in config: "xintensity":"" value=1~9999
*intensity* put in config: "intensity":"" value=8~31
*rawintensity* put in config: "rawintensity":"" value=1~2147483647

Keep in mind that only one parameter will work.

So the config side related to GPU tweaking will look like:
   "gpu-powertune": "0",
   "worksize": "192",
   "name": "eth",
   "algorithm": "ethash",
   "gpu-threads": "1",
   "xintensity": "4096"

*gpu-threads* is a number of mining threads per GPU
I don't how effective can be this parameter changing but default is "1"

I noticed your xintensity value is 4096. If I put that to be more than 1024, there are loads of hardware errors.

scryptr

legendary

Activity: 1797

Merit: 1028

PROGRAMMERS CAN WORK MIRACLES--

But all I have heard for a year is that there is not much juice to squeeze from the DAG. Generating it on-the-fly was a big thing, saving start-up minutes.

Likely, card optimization will be the way to go. 29MH/s on the RX 480 is top speed reported. New cards (480x, 490, etc) are still top secret, not expected until next year. I search, and see the same old tech blogs. --scryptr

yoyo1

newbie

Activity: 47

Merit: 0

Thank you very much for the explanation.
Can community expect to get some extra juice from the sgminer-gm?

And are there any news about AMD ADL and amdgpu-pro updates?
How can it be possible to overclock AMD cards without flashing (i'm not so risky actually).

scryptr

legendary

Activity: 1797

Merit: 1028

Quote from: yoyo1 on September 29, 2016, 08:24:55 AM

I wonder what options should I specify for dual mining?
Is it possible?
May I use "ethash" and mine ETH & ETC at the same time for example?

ETHEREUM (ETH) AND ETHEREUM CLASSIC (ETC) ARE ON DIFFERENT EPOCHS--

The DAG files woulkd be different in size and composition. Hash from one would be invalid on the other. Dual mining is done with a memory intensive algo (ETH) and a compute intensive algo like SiaCoin (SC). A dual miner is specially coded to manage two different algos without generating conflicts. SGminer is not a dual miner. --scryptr

yoyo1

newbie

Activity: 47

Merit: 0

I wonder what options should I specify for dual mining?
Is it possible?
May I use "ethash" and mine ETH & ETC at the same time for example?

yoyo1

newbie

Activity: 47

Merit: 0

According to:
RTFM
https://github.com/genesismining/sgminer-gm/blob/master/doc/configuration.md

Thank you for the guide.

I'd like to summarize some findings:
There are several ways of loading GPU via manipulating following parameters:
*xintensity* put in config: "xintensity":"" value=1~9999
*intensity* put in config: "intensity":"" value=8~31
*rawintensity* put in config: "rawintensity":"" value=1~2147483647

Keep in mind that only one parameter will work.

So the config side related to GPU tweaking will look like:
   "gpu-powertune": "0",
   "worksize": "192",
   "name": "eth",
   "algorithm": "ethash",
   "gpu-threads": "1",
   "xintensity": "4096"

*gpu-threads* is a number of mining threads per GPU
I don't how effective can be this parameter changing but default is "1"

nerdralph

sr. member

Activity: 588

Merit: 251

Quote from: Eliovp on September 28, 2016, 01:43:07 PM

Quote from: rednoW on September 28, 2016, 11:45:48 AM

Quote from: nerdralph on September 28, 2016, 06:46:01 AM

Just noticed this miner automatically determines the eth stratum protocol mode of the pool it is connecting to, at least for dwarf vs coinotron mode. Something I may try to add to Genoil's...

Strange, when I tried it did not want to connect to dwarfpool's 8080 port, only http 80 port was working.

Ditto with any pools based on sammy's open-ethereum-pool.

I tested with MPH port 20535 and Alpereum port 4001. Maybe it's just the pools auto-detecting the client stratum protocol. I'm still going to look at doing auto-detect on my fork of Genoil's client.
When connecting to MPH, it sends an unsolicited mining.notify. Alpereum sends an unsolicited jsonrpc result. Dwarfpool requires the client to make a request before it responds. So it may be possible to auto-detect on the client.

Eliovp

legendary

Activity: 1050

Merit: 1293

Huh?

Quote from: rednoW on September 28, 2016, 11:45:48 AM

Quote from: nerdralph on September 28, 2016, 06:46:01 AM

Just noticed this miner automatically determines the eth stratum protocol mode of the pool it is connecting to, at least for dwarf vs coinotron mode. Something I may try to add to Genoil's...

Strange, when I tried it did not want to connect to dwarfpool's 8080 port, only http 80 port was working.

Ditto with any pools based on sammy's open-ethereum-pool.

rednoW

legendary

Activity: 1510

Merit: 1003

Quote from: nerdralph on September 28, 2016, 06:46:01 AM

Just noticed this miner automatically determines the eth stratum protocol mode of the pool it is connecting to, at least for dwarf vs coinotron mode. Something I may try to add to Genoil's...

Strange, when I tried it did not want to connect to dwarfpool's ~~8080~~ 8008 port, only http 80 port was working.

Jacknife

newbie

Activity: 64

Merit: 0

Quote from: nerdralph on September 28, 2016, 06:46:01 AM

Just noticed this miner automatically determines the eth stratum protocol mode of the pool it is connecting to, at least for dwarf vs coinotron mode. Something I may try to add to Genoil's...

That could be true. I just connected to the ethpool.org. I did not change anything in the mode for the stratum.

nerdralph

sr. member

Activity: 588

Merit: 251

Just noticed this miner automatically determines the eth stratum protocol mode of the pool it is connecting to, at least for dwarf vs coinotron mode. Something I may try to add to Genoil's...

nerdralph

sr. member

Activity: 588

Merit: 251

Quote from: yoyo1 on September 27, 2016, 05:30:44 AM

Thank you for your investigations guys!
So here is my piece of:

   "worksize": "192",
   "name": "eth",
   "algorithm": "ethash",
   "gpu-threads": "2",
   "xintensity": "4608"

Gives me robust 24 Mh/s at Dwarf at *stock clocks* via HIS RX 480 8Gb version Roaring

We need some improvements guys you know.

And options explanation for dummies like me.

RTFM
https://github.com/genesismining/sgminer-gm/blob/master/doc/configuration.md

sanas

hero member

Activity: 546

Merit: 500

Quote from: Eliovp on September 26, 2016, 03:11:40 PM

Quote from: sanas on September 26, 2016, 02:47:43 PM

Quote from: nerdralph on September 26, 2016, 12:54:15 PM

Quote from: Qastessa on September 26, 2016, 12:37:30 PM

Quote from: nerdralph on September 26, 2016, 11:49:35 AM

I've done some more testing, trying different xI and rI values, and can say the following:
1) Hawaii performance is great; as good or better than Claymore & Genoil.
2) Tonga performance is about the same as Claymore & Genoil.
3) Pitcairn performance sucks; ~10% slower than Claymore & Genoil.

@Wolf0, any plans to tune this up for Pitcairn?

What are your work size, thread concurrency, intensity values for the R9390? What is your 390 setting?

Where did I say I have an R9 390?

You mentioned Hawaii cards.

There are only two Hawaii cards, the R9 390 or the 390X graphics cards.

You forgot the 290, 290x, 295x2(Dual Hawaii "Vesuvius") Wink

There is little difference between the 290 and 290x or the 390 and 390x. I think we use similar settings for them.

yoyo1

newbie

Activity: 47

Merit: 0

Thank you for your investigations guys!
So here is my piece of:

   "worksize": "192",
   "name": "eth",
   "algorithm": "ethash",
   "gpu-threads": "2",
   "xintensity": "4608"

Gives me robust 24 Mh/s at Dwarf at *stock clocks* via HIS RX 480 8Gb version Roaring

We need some improvements guys you know.

And options explanation for dummies like me.

nerdralph

sr. member

Activity: 588

Merit: 251

After a lot of trial and error, I'm using 64/xI:896 on Hawaii and Tonga. That also gives me the best results on Pitcairn (though still not as good as Genoil's).
xI:1024 isn't any better, and since I think worksize * shaders * xI gives OpenCL work-items per kernel itteration, larger xI values will increase the stale share rate.

And don't forget to turn off auto-gpu if you are already controlling your GPU clock rate.

Topic: Genesis Mining Presents: SGMiner-GM - now with Zawawa's GG! [Updated 17/01/2017] - page 31. (Read 140421 times)