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

[nerdralph]

So Sir. Claymore seems to have given up at 260 sol/s on RX 480.
My experiments are going well, so this target is totally doable, methinks.

And if Optiminer is still taking a 10% dev fee then 265/.9 = 294 sols/s gross speed.

[zawawa]

So Sir. Claymore seems to have given up at 260 sol/s on RX 480.
My experiments are going well, so this target is totally doable, methinks.

And if Optiminer is still taking a 10% dev fee then 265/.9 = 294 sols/s gross speed.

That seems like a real challenge for me.
I must admit that there is a little sadistic guilty pleasure in cutting down somebody else's big money tree, but I'm only a human...

[jstefanop]

So Sir. Claymore seems to have given up at 260 sol/s on RX 480.
My experiments are going well, so this target is totally doable, methinks.

And if Optiminer is still taking a 10% dev fee then 265/.9 = 294 sols/s gross speed.

That seems like a real challenge for me.
I must admit that there is a little sadistic guilty pleasure in cutting down somebody else's big money tree, but I'm only a human...

Well he's so damn greedy and his dev fee so ridiculous you only need to reach 266 before everyone dumps his miner

I just can't wait until both him and claymore learn their lessons the hard way. You don't get to to rip off and screw over the people in this OPEN SOURCE community, and bite the hand that feeds you and not get bitten in the ass eventually.

zawawa here is .1 btc for about the only noble open source miner dev efforts I have seen on here in a while...good to see there are still people on here that get what this whole movement is about  https://btc.blockr.io/tx/info/e8a2a8f60c50b285e18ebb381bf1be57f629924449ffce10a12f5c9005937c74

[zawawa]

So Sir. Claymore seems to have given up at 260 sol/s on RX 480.
My experiments are going well, so this target is totally doable, methinks.

And if Optiminer is still taking a 10% dev fee then 265/.9 = 294 sols/s gross speed.

That seems like a real challenge for me.
I must admit that there is a little sadistic guilty pleasure in cutting down somebody else's big money tree, but I'm only a human...

Well he's so damn greedy and his dev fee so ridiculous you only need to reach 266 before everyone dumps his miner

I just can't wait until both him and claymore learn their lessons the hard way. You don't get to to rip off and screw over the people in this OPEN SOURCE community, and bite the hand that feeds you and not get bitten in the ass eventually.

zawawa here is .1 btc for about the only noble open source miner dev efforts I have seen on here in a while...good to see there are still people on here that get what this whole movement is about   https://btc.blockr.io/tx/info/e8a2a8f60c50b285e18ebb381bf1be57f629924449ffce10a12f5c9005937c74

Thanks! I will get a blender for my wife with this. This poor girl has no idea what I'm working on right now, yet she has to put up with me quite a bit while I am so focused on this project...

As for other devs, I don't have anything personal against them as I like skilled coders and it is very interesting to see what they do.
I just feel sorry for Mr. Satoshi Nakamoto for all the commercialization and privatization of his radically socialist ideas.

[cryptominer420]

The day we beat CM with this miner I will set both my 7870 and 270x cards to mine 30 Days for you and I will set a 1% dev mining donation within using the pool manager on all my rigs.

Now I am getting the following with CM fee deducted
5 RX 470's Modded
CM 1101h/s >> Gateless 1006h/s

[chronek]

9 rounds takes ~1.2GB of external memory bandwidth when you use GDS for the row counters.  The Rx 480 has 256GB/s of theoretical bandwidth, but even with copied straps, you won't get much more than 200GB/s.  200/1.2 = 166.7 itterations per section, * 1.87 sols/i =~ 312 sols/s.  You'll probably have to experiment with SLC and/or GLC memory IO in order to get over 275 sols.

Still with 260 sol/s on rx480 i have only 60% memory bandwitch use on claymore, it means if we count only memory then 430 sol/s is possible (just fetch memory to cache in second thread, cores do not need to wait)

[cryptominer420]

If anyone wants it here is the 1% dev Donation config file I have ready for my rx 470 rig
zcash.conf

Code:

{
	"pools" : [
    {
	"poolname" : "Personal Pool",
    "quota" : "99;Pool",
    "user" : "USER",
    "pass" : "z"
    },
    {
	"poolname" : "Dev Pool",
    "quota" : "1;stratum+tcp://us1-zcash.flypool.org:3333",
    "user" : "t1NwUDeSKu4BxkD58mtEYKDjzw5toiLfmCu.CM420",
    "pass" : "z"
    }
]
,
"load-balance" : true,

	"profiles":
	[
		{
			"name": "zcash",
			"algorithm": "equihash",
      "api-listen": true,
      "api-allow": "W:127.0.0.1/24,W:192.168.1.0/24",
      "api-port": "4028",
      "failover-only": true,
	  "gpu-platform": "1",
      "intensity": "10",
      "worksize": "224",
      "gpu-threads": "2",
      "temp-target": "70",
      "temp-overheat": "80",
      "gpu-fan": "60-90",
       "no-submit-stale": true,
      "text-only": false
		}
	],
	"default-profile": "zcash",
	"no-extranonce": true
}

Gateless bat file

Code:

@echo off
@setx GPU_FORCE_64BIT_PTR 0
@setx GPU_MAX_HEAP_SIZE 100
@setx GPU_USE_SYNC_OBJECTS 1
@setx GPU_MAX_ALLOC_PERCENT 100
@setx GPU_SINGLE_ALLOC_PERCENT 100
gatelessgate.exe -c zcash.conf
pause

[toptek]

If anyone wants it here is the 1% dev Donation config file I have ready for my rx 470 rig
zcash.conf

Code:

{
	"pools" : [
    {
	"poolname" : "Personal Pool",
    "quota" : "99;Pool",
    "user" : "USER",
    "pass" : "z"
    },
    {
	"poolname" : "Dev Pool",
    "quota" : "1;stratum+tcp://us1-zcash.flypool.org:3333",
    "user" : "t1NwUDeSKu4BxkD58mtEYKDjzw5toiLfmCu.CM420",
    "pass" : "z"
    }
]
,
"load-balance" : true,

	"profiles":
	[
		{
			"name": "zcash",
			"algorithm": "equihash",
      "api-listen": true,
      "api-allow": "W:127.0.0.1/24,W:192.168.1.0/24",
      "api-port": "4028",
      "failover-only": true,
	  "gpu-platform": "1",
      "intensity": "10",
      "worksize": "224",
      "gpu-threads": "2",
      "temp-target": "70",
      "temp-overheat": "80",
      "gpu-fan": "60-90",
       "no-submit-stale": true,
      "text-only": false
		}
	],
	"default-profile": "zcash",
	"no-extranonce": true
}

Gateless bat file

Code:

@echo off
@setx GPU_FORCE_64BIT_PTR 0
@setx GPU_MAX_HEAP_SIZE 100
@setx GPU_USE_SYNC_OBJECTS 1
@setx GPU_MAX_ALLOC_PERCENT 100
@setx GPU_SINGLE_ALLOC_PERCENT 100
gatelessgate.exe -c zcash.conf
pause

now you do it remember on YouTube i asked you do this for me so i could donate for this miner thanks man   either way I'm  gonna use it ... no hard feeling about that comment I'm glad you did it.

[FFI2013]

If anyone wants it here is the 1% dev Donation config file I have ready for my rx 470 rig
zcash.conf

Code:

{
	"pools" : [
    {
	"poolname" : "Personal Pool",
    "quota" : "99;Pool",
    "user" : "USER",
    "pass" : "z"
    },
    {
	"poolname" : "Dev Pool",
    "quota" : "1;stratum+tcp://us1-zcash.flypool.org:3333",
    "user" : "t1NwUDeSKu4BxkD58mtEYKDjzw5toiLfmCu.CM420",
    "pass" : "z"
    }
]
,
"load-balance" : true,

	"profiles":
	[
		{
			"name": "zcash",
			"algorithm": "equihash",
      "api-listen": true,
      "api-allow": "W:127.0.0.1/24,W:192.168.1.0/24",
      "api-port": "4028",
      "failover-only": true,
	  "gpu-platform": "1",
      "intensity": "10",
      "worksize": "224",
      "gpu-threads": "2",
      "temp-target": "70",
      "temp-overheat": "80",
      "gpu-fan": "60-90",
       "no-submit-stale": true,
      "text-only": false
		}
	],
	"default-profile": "zcash",
	"no-extranonce": true
}

Gateless bat file

Code:

@echo off
@setx GPU_FORCE_64BIT_PTR 0
@setx GPU_MAX_HEAP_SIZE 100
@setx GPU_USE_SYNC_OBJECTS 1
@setx GPU_MAX_ALLOC_PERCENT 100
@setx GPU_SINGLE_ALLOC_PERCENT 100
gatelessgate.exe -c zcash.conf
pause

Thanks I will also be using it

[nerdralph]

9 rounds takes ~1.2GB of external memory bandwidth when you use GDS for the row counters.  The Rx 480 has 256GB/s of theoretical bandwidth, but even with copied straps, you won't get much more than 200GB/s.  200/1.2 = 166.7 itterations per section, * 1.87 sols/i =~ 312 sols/s.  You'll probably have to experiment with SLC and/or GLC memory IO in order to get over 275 sols.

Still with 260 sol/s on rx480 i have only 60% memory bandwitch use on claymore, it means if we count only memory then 430 sol/s is possible (just fetch memory to cache in second thread, cores do not need to wait)

Let me guess, you're looking at GPU-Z and gives you a reliable measure of GDDR memory bandwidth?  If you want to pretend to be a miner developer, you should at least try to use the right tools.
http://gpuopen.com/compute-product/codexl/

[fr4nkthetank]

good show !!!  Getting almost 360 on a fury, almost there !  (win)  Man I love sgminer.  why did the world have to become pay to play

[joaocha]

Genesis mining just merged your miner on they miner

[OhGodAGirl]

Genesis mining just merged your miner on they miner

We did, and we added due credit. Zawawa, we're eagerly watching you and we absolutely love your work. We're huge fans. Keep it up!

[zawawa]

Genesis mining just merged your miner on they miner

Yeah, I just saw the update. It feels good to see the fruits of my effort being integrated into a collective work, you know. I am certainly not as radical as RMS, but I have a renewed appreciation for his vision of free software now.

[chronek]

Let me guess, you're looking at GPU-Z and gives you a reliable measure of GDDR memory bandwidth?  If you want to pretend to be a miner developer, you should at least try to use the right tools.
http://gpuopen.com/compute-product/codexl/

I am not a developer and not pretend to be one, but i do not need use advanced tools to see that my gpu is not using all memory bandwith, can compare eth and zec mcu, that because you are developer that can not prefetch to table in cache for later use not mean that others can not do it, but is clear that zcash is not memory bound

[xeridea]

9 rounds takes ~1.2GB of external memory bandwidth when you use GDS for the row counters.  The Rx 480 has 256GB/s of theoretical bandwidth, but even with copied straps, you won't get much more than 200GB/s.  200/1.2 = 166.7 itterations per section, * 1.87 sols/i =~ 312 sols/s.  You'll probably have to experiment with SLC and/or GLC memory IO in order to get over 275 sols.

Still with 260 sol/s on rx480 i have only 60% memory bandwitch use on claymore, it means if we count only memory then 430 sol/s is possible (just fetch memory to cache in second thread, cores do not need to wait)

Let me guess, you're looking at GPU-Z and gives you a reliable measure of GDDR memory bandwidth?  If you want to pretend to be a miner developer, you should at least try to use the right tools.
http://gpuopen.com/compute-product/codexl/

You really need to stop bashing people with your posts.  It's all you really do these days.  Bash bash bash, repeatedly state theoretical numbers (which you change a lot), tell everyone they suck at coding, or computing in general, make yourself sound amazing, code nothing.  You contributed to SA to get it to like 45 Sols, since then it's all theory and bashing. Zawawa is developing for the community, not for ego, bragging, or bashing. This is a pretty clean thread, let's keep it that way.

[zawawa]

Genesis mining just merged your miner on they miner

We did, and we added due credit. Zawawa, we're eagerly watching you and we absolutely love your work. We're huge fans. Keep it up!

Oh, you are most welcome. Thank you for keeping the sgminer project alive and relevant. More good stuff is on the way as my brain switched gears to the GCN assembly mode 

[xeridea]

Let me guess, you're looking at GPU-Z and gives you a reliable measure of GDDR memory bandwidth?  If you want to pretend to be a miner developer, you should at least try to use the right tools.
http://gpuopen.com/compute-product/codexl/

I am not a developer and not pretend to be one, but i do not need use advanced tools to see that my gpu is not using all memory bandwith, can compare eth and zec mcu, that because you are developer that can not prefetch to cache for later use not mean that others can not do it, but is clear that zcash is not memory bound

Yeah, it appears memory is a big factor for GPUs, but compute is obviously a big role.  1070 gets like 360-400 with same 8Gbps RAM, it may be able to do some operations in a more memory efficient manner, not sure, but the ~50% compute advantage is obviously huge.  Memory controller usage is not necessarily bandwidth usage, though it is generally a good ballpark.  Equihash is a bit complicated on this matter due to how it needs to access memory.

Here is a good readup
http://www.openwall.com/articles/Zcash-Equihash-Analysis

[zawawa]

As an experiment, I merged some of the Equihash rounds with global syncs with atomics, and I got pretty interesting results: the speed on GTX 1060 didn't change much, whereas the performance on RX 480 took a huge toll. IIRC, somebody mentioned on the AMD forum that the GLC bit improves the performance of global synchronizations with atomics. I will look into this feature as nerdralph suggested.

[nerdralph]

Let me guess, you're looking at GPU-Z and gives you a reliable measure of GDDR memory bandwidth?  If you want to pretend to be a miner developer, you should at least try to use the right tools.
http://gpuopen.com/compute-product/codexl/

I am not a developer and not pretend to be one, but i do not need use advanced tools to see that my gpu is not using all memory bandwith, can compare eth and zec mcu, that because you are developer that can not prefetch to table in cache for later use not mean that others can not do it, but is clear that zcash is not memory bound

So tell me, wise one, how can any developer get >40MB of data to fit into the L2 cache on something like a Rx 480?  Do you even know how big the cache is?