CCminer(SP-MOD) Modded GPU kernels.

[bensam1231]

I really hope Ethereum never goes PoS because when it does GPU mining is going to completely fall apart. Eth is gaining about 1TH per month and it doesn't look like it's slowing down.

We're going to be completely fucked when it does.

PoS doesn't really matter because there's a difficulty bomb in place so mining will get harder and harder until it's not profitable anymore:

Difficulty adjustment scheme

A lot of you have been wondering how we would implement a switch from PoW to PoS in time for Serenity. This will be handled by the newly introduced difficulty adjustment scheme, which elegantly guarantees a hard-fork point in the next 16 months.

It works as follow: starting from block 200,000 the difficulty will undergo an exponential increase which will only become noticeable in about a year. At that point (just around the release of the Serenity milestone), we’ll see a significant increase in difficulty which will start pushing the block resolution time upwards.

So, a year on, the network will continue to be useful for roughly 3-4 months, but eventually will reach an ‘Ice Age’ of sorts: the difficulty will simply be too high for anyone to find a block. This will allow us to introduce PoS, perhaps via Casper, if it proves itself.

The difficulty bomb might get changed to take longer to set in though if PoS/Serenity is not ready in time.

That's good to know... So about a year from when that post was written... Which is ~3 months from now.

Not sure why they went with increasing difficulty instead of reducing the block reward, they're essentially the same thing. Also rather disheartening that it's logarithmic with a steep drop off instead of a linear increase in difficulty or decrease in block reward. This means the market is currently deceptive and we're quickly approaching Armageddon. It wont be long before everything explodes.

Three months and we're back to earning $.50 per 1070 per day... if that. GPU mining never had this much hash on it, even back during the boom in 2013/14. Pretty sure everything is going negative. 3TH is a lot of mining power to displace.

I really hope they reconsider their current position in light of other alt coins and substantial growth of their coin, if only changing it to a linear increase in difficulty before going PoS. A lot of miners, the market, currently depends on Ethereum. The spice must flow.

[Amph]

I really hope Ethereum never goes PoS because when it does GPU mining is going to completely fall apart. Eth is gaining about 1TH per month and it doesn't look like it's slowing down.

We're going to be completely fucked when it does.

PoS doesn't really matter because there's a difficulty bomb in place so mining will get harder and harder until it's not profitable anymore:

Difficulty adjustment scheme

A lot of you have been wondering how we would implement a switch from PoW to PoS in time for Serenity. This will be handled by the newly introduced difficulty adjustment scheme, which elegantly guarantees a hard-fork point in the next 16 months.

It works as follow: starting from block 200,000 the difficulty will undergo an exponential increase which will only become noticeable in about a year. At that point (just around the release of the Serenity milestone), we’ll see a significant increase in difficulty which will start pushing the block resolution time upwards.

So, a year on, the network will continue to be useful for roughly 3-4 months, but eventually will reach an ‘Ice Age’ of sorts: the difficulty will simply be too high for anyone to find a block. This will allow us to introduce PoS, perhaps via Casper, if it proves itself.

The difficulty bomb might get changed to take longer to set in though if PoS/Serenity is not ready in time.

That's good to know... So about a year from when that post was written... Which is ~3 months from now.

Not sure why they went with increasing difficulty instead of reducing the block reward, they're essentially the same thing. Also rather disheartening that it's logarithmic with a steep drop off instead of a linear increase in difficulty or decrease in block reward. This means the market is currently deceptive and we're quickly approaching Armageddon. It wont be long before everything explodes.

Three months and we're back to earning $.50 per 1070 per day... if that. GPU mining never had this much hash on it, even back during the boom in 2013/14. Pretty sure everything is going negative. 3TH is a lot of mining power to displace.

I really hope they reconsider their current position in light of other alt coins and substantial growth of their coin, if only changing it to a linear increase in difficulty before going PoS. A lot of miners, the market, currently depends on Ethereum. The spice must flow.

1070 should give us more life in mining the same with 480, much more, the earning with their spec would be 2.5 x so even if the diff double, we would still make more than what we are making right now

not to mention that the global diff would decrease if the overall market will tank due to etheruem going out of the scene

but in my case i might be forced to shut down the rig, and wait for better time, one thing is sure for me... i will not sell the rig no matter what will be the profit, i'll simply turn it off...

[pallas]

yeah don't sell it, a new coin may come and you need to be ready.
it happened so many times during the last 3 years that I've learnt the lesson

[Genoil]

My fork has a --cuda-schedule flag that allows changing the blocking /syncing behaviour. It may help when running it alongside ccminer.

[scryptr]

My fork has a --cuda-schedule flag that allows changing the blocking /syncing behaviour. It may help when running it alongside ccminer.

BLOCKING / SYNCING BEHAVIOR--

I 'd like to ask you for a better explanation.  How does a miner set and adjust this feature?  Is there documentation that explains this feature?       --scryptr

[scryptr]

You have to realize that Nvidia chips up until Pascal, have penalty when switching contexts of computing - or between computing and gaming
afaik, Pascal does have a penalty but way smaller.
AMD chips can switch almost for free between tasks - because GCN was deveoped with such scenarios in mind...
Too bad AMD never capitalised on this, except in cryptomining... and funny thing is, they were not targeting this niche at all


So "native" dual-miner will give way better results for NV. On AMD combining 2 non-native miners is more likely to give satisfying results

IMO

ABORIGINES DANCE IN MY 970 RIG?

OK, there is voodoo magic in my nVidia rig and it makes things all better?  WTF?  What precisely are you saying here?       --scryptr

That AMD architecture is more friendly to "dual-mining"
Simply put, its easier to do make it work.
Hence for Nvidia, a dedicated effort to make dual miner will give better more consistent results than simply ruinning 2 different miners and playing with intensities.
And what runs on one MB+video+driver combo may not work on other

FOR NVIDIA A DUAL MINER WOULD WORK BETTER--

But it might be harder to code the miner.  That is what I am getting, at least.

Genoil says that his Ethminer has a "--cuda-schedule" command line flag that may make it easier to run 2 miners (ethminer/ccminer) concurrently.  I don't understand it well, but the code involved may be a key to an nVidia dual-miner.

Thanks for your response.         --scryptr

[Genoil]

Ethminer.exe --help lists the options for --cuda-schedule.

[bensam1231]

I really hope Ethereum never goes PoS because when it does GPU mining is going to completely fall apart. Eth is gaining about 1TH per month and it doesn't look like it's slowing down.

We're going to be completely fucked when it does.

PoS doesn't really matter because there's a difficulty bomb in place so mining will get harder and harder until it's not profitable anymore:

Difficulty adjustment scheme

A lot of you have been wondering how we would implement a switch from PoW to PoS in time for Serenity. This will be handled by the newly introduced difficulty adjustment scheme, which elegantly guarantees a hard-fork point in the next 16 months.

It works as follow: starting from block 200,000 the difficulty will undergo an exponential increase which will only become noticeable in about a year. At that point (just around the release of the Serenity milestone), we’ll see a significant increase in difficulty which will start pushing the block resolution time upwards.

So, a year on, the network will continue to be useful for roughly 3-4 months, but eventually will reach an ‘Ice Age’ of sorts: the difficulty will simply be too high for anyone to find a block. This will allow us to introduce PoS, perhaps via Casper, if it proves itself.

The difficulty bomb might get changed to take longer to set in though if PoS/Serenity is not ready in time.

That's good to know... So about a year from when that post was written... Which is ~3 months from now.

Not sure why they went with increasing difficulty instead of reducing the block reward, they're essentially the same thing. Also rather disheartening that it's logarithmic with a steep drop off instead of a linear increase in difficulty or decrease in block reward. This means the market is currently deceptive and we're quickly approaching Armageddon. It wont be long before everything explodes.

Three months and we're back to earning $.50 per 1070 per day... if that. GPU mining never had this much hash on it, even back during the boom in 2013/14. Pretty sure everything is going negative. 3TH is a lot of mining power to displace.

I really hope they reconsider their current position in light of other alt coins and substantial growth of their coin, if only changing it to a linear increase in difficulty before going PoS. A lot of miners, the market, currently depends on Ethereum. The spice must flow.

1070 should give us more life in mining the same with 480, much more, the earning with their spec would be 2.5 x so even if the diff double, we would still make more than what we are making right now

not to mention that the global diff would decrease if the overall market will tank due to etheruem going out of the scene

but in my case i might be forced to shut down the rig, and wait for better time, one thing is sure for me... i will not sell the rig no matter what will be the profit, i'll simply turn it off...

Everyone is upgrading to 1070s... AMD still gets more hashrate according to current reports on 1080/1070s. Once hardware normalizes the difficulty goes down by people going out of business who have the most expensive power.

Sitting on hardware is a opportunity cost. It doesn't pay to have a $2000 piece of equipment laying around that depreciates over time. You essentially lose money by hanging onto it.

This all has happened before.

[Amph]

i don't think everyone will upgrade to 1070, big farm with 1k gpu can not surely, afford to buy 400k worth of gpu just to have some more hash and less power

right now to double the diff you need to add something like 100k or more gpu, this is around $200(counting the 480 here not even the 1070) x 100k, which is $20M...no way it can be done when the new gpu will come out and not even some months after it

i guess there is a small time frame on which, with new gpu, small farm can have a huge profit, those month are july and august

[nicehash]

We're giving away 100% speedup on Lyra2REv2: https://www.nicehash.com/?p=news&id=80

Best regards,
NiceHash team.

[pallas]

We're giving away 100% speedup on Lyra2REv2: https://www.nicehash.com/?p=news&id=80

Best regards,
NiceHash team.

Sources?

[myagui]

We're giving away 100% speedup on Lyra2REv2: https://www.nicehash.com/?p=news&id=80

Best regards,
NiceHash team.

Sources?

https://github.com/nicehash/NiceHashMiner/releases/tag/1.5.1.0

 

Edit: no clue if the ccminer sources are there btw, just pointing out their git location!

[joblo]

We're giving away 100% speedup on Lyra2REv2: https://www.nicehash.com/?p=news&id=80

Best regards,
NiceHash team.

Where did the lyra2v2 kernel come from? Is it open source?

[joblo]

We're giving away 100% speedup on Lyra2REv2: https://www.nicehash.com/?p=news&id=80

Best regards,
NiceHash team.

Sources?

https://github.com/nicehash/NiceHashMiner/releases/tag/1.5.1.0

 

Edit: no clue if the ccminer sources are there btw, just pointing out their git location!

They aren't.

[sp_]

The lyra2v2 kernal has been split into three passes instead of one pass. I guess to reduce the register pressure.

I just did a comparison on the 750ti with sp-mod private #6 (-i 18)

My cubehash.blakekeccak256,skein,bmw kernals are all faster, but the lyra2v2 kernal is 44% slower.

sp-mod private: 4675
Lyra2v2-Nicehash: 6100 (30% faster)

Not bad, I think the work has been done by djm34.

Code:

==3632== Profiling application: ccminer_sp_lyra2v2.exe -a lyra2v2 -i 18
==3632== Profiling result:

Time(%)      Time     Calls       Avg       Min       Max  Name
 46.87%  2.75600s       143  19.273ms  19.181ms  21.679ms  lyra2v2_gpu_hash_32_2(unsigned int, unsigned int, __int64*)
 35.42%  2.08255s       286  7.2817ms  7.2457ms  8.2117ms  cubehash256_gpu_hash_32(unsigned int, unsigned int, uint2*)
  6.10%  358.73ms       143  2.5086ms  2.4916ms  2.8360ms  blakeKeccak256_gpu_hash_80(unsigned int, unsigned int, unsigned int*)
  4.33%  254.58ms       143  1.7803ms  1.7564ms  2.0057ms  skein256_gpu_hash_32(unsigned int, unsigned int, __int64*)
  3.66%  215.44ms       143  1.5066ms  1.4988ms  1.6974ms  lyra2v2_gpu_hash_32_1(unsigned int, unsigned int, uint2*)
  1.89%  111.38ms       143  778.90us  773.77us  877.74us  lyra2v2_gpu_hash_32_3(unsigned int, unsigned int, uint2*)
  1.71%  100.41ms       143  702.18us  698.09us  792.84us  bmw256_gpu_hash_32(unsigned int, unsigned int, uint2*, unsigned int*, unsigned int)
  0.01%  703.33us       143  4.9180us  4.5440us  7.1050us  [CUDA memcpy DtoH]
  0.01%  470.55us       143  3.2900us  3.1040us  7.6490us  [CUDA memset]
  0.00%  11.744us         7  1.6770us  1.3120us  3.7440us  [CUDA memcpy HtoD]

sp-mod private #6:

==820== Profiling application: ccminer.exe -a lyra2v2 --benchmark -i 18
==820== Profiling result:
Time(%)      Time     Calls       Avg       Min       Max  Name
 68.70%  5.65170s       147  38.447ms  38.297ms  43.242ms  lyra2v2_gpu_hash_32(unsigned int, unsigned int, uint2*)
 22.94%  1.88714s       294  6.4188ms  6.4047ms  7.2165ms  cubehash256_gpu_hash_32(unsigned int, unsigned int, uint2*)
  3.96%  325.75ms       147  2.2160ms  2.2060ms  2.4968ms  blakeKeccak256_gpu_hash_80(unsigned int, unsigned int, unsigned int*)
  3.21%  263.95ms       147  1.7956ms  1.7774ms  2.0206ms  skein256_gpu_hash_32(unsigned int, unsigned int, __int64*)
  1.18%  97.455ms       147  662.96us  659.58us  745.59us  bmw256_gpu_hash_32(unsigned int, unsigned int, uint2*, unsigned int*, unsigned int)
  0.00%  385.64us       147  2.6230us  2.5270us  3.7110us  [CUDA memset]
  0.00%  260.56us       147  1.7720us  1.5680us  3.1360us  [CUDA memcpy DtoH]
  0.00%  6.2390us         7     891ns     704ns  1.3440us  [CUDA memcpy HtoD]

[scryptr]

The lyra2v2 kernal has been split into three passes instead of one pass. I guess to reduce the register pressure.

I just did a comparison on the 750ti with sp-mod private #6 (-i 18)

My cubehash.blakekeccak256,skein,bmw kernals are all faster, but the lyra2v2 kernal is 44% slower.

sp-mod private: 4675
Lyra2v2-Nicehash: 6100 (30% faster)

Not bad, I think the work has been done by djm34.

Code:

==3632== Profiling application: ccminer_sp_lyra2v2.exe -a lyra2v2 -i 18
==3632== Profiling result:

Time(%)      Time     Calls       Avg       Min       Max  Name
 46.87%  2.75600s       143  19.273ms  19.181ms  21.679ms  lyra2v2_gpu_hash_32_2(unsigned int, unsigned int, __int64*)
 35.42%  2.08255s       286  7.2817ms  7.2457ms  8.2117ms  cubehash256_gpu_hash_32(unsigned int, unsigned int, uint2*)
  6.10%  358.73ms       143  2.5086ms  2.4916ms  2.8360ms  blakeKeccak256_gpu_hash_80(unsigned int, unsigned int, unsigned int*)
  4.33%  254.58ms       143  1.7803ms  1.7564ms  2.0057ms  skein256_gpu_hash_32(unsigned int, unsigned int, __int64*)
  3.66%  215.44ms       143  1.5066ms  1.4988ms  1.6974ms  lyra2v2_gpu_hash_32_1(unsigned int, unsigned int, uint2*)
  1.89%  111.38ms       143  778.90us  773.77us  877.74us  lyra2v2_gpu_hash_32_3(unsigned int, unsigned int, uint2*)
  1.71%  100.41ms       143  702.18us  698.09us  792.84us  bmw256_gpu_hash_32(unsigned int, unsigned int, uint2*, unsigned int*, unsigned int)
  0.01%  703.33us       143  4.9180us  4.5440us  7.1050us  [CUDA memcpy DtoH]
  0.01%  470.55us       143  3.2900us  3.1040us  7.6490us  [CUDA memset]
  0.00%  11.744us         7  1.6770us  1.3120us  3.7440us  [CUDA memcpy HtoD]

sp-mod private #6:

==820== Profiling application: ccminer.exe -a lyra2v2 --benchmark -i 18
==820== Profiling result:
Time(%)      Time     Calls       Avg       Min       Max  Name
 68.70%  5.65170s       147  38.447ms  38.297ms  43.242ms  lyra2v2_gpu_hash_32(unsigned int, unsigned int, uint2*)
 22.94%  1.88714s       294  6.4188ms  6.4047ms  7.2165ms  cubehash256_gpu_hash_32(unsigned int, unsigned int, uint2*)
  3.96%  325.75ms       147  2.2160ms  2.2060ms  2.4968ms  blakeKeccak256_gpu_hash_80(unsigned int, unsigned int, unsigned int*)
  3.21%  263.95ms       147  1.7956ms  1.7774ms  2.0206ms  skein256_gpu_hash_32(unsigned int, unsigned int, __int64*)
  1.18%  97.455ms       147  662.96us  659.58us  745.59us  bmw256_gpu_hash_32(unsigned int, unsigned int, uint2*, unsigned int*, unsigned int)
  0.00%  385.64us       147  2.6230us  2.5270us  3.7110us  [CUDA memset]
  0.00%  260.56us       147  1.7720us  1.5680us  3.1360us  [CUDA memcpy DtoH]
  0.00%  6.2390us         7     891ns     704ns  1.3440us  [CUDA memcpy HtoD]

LYRA2V2 HASH RATE--

I am getting close to 14MH/s on my GTX 960 for the revised lyra2v2 kernel.  It used to be about 6.5 MH/s on the same card.  Now, it is faster than Quark by 3MH/s.

NiceHashMiner reports that I am earning at a rate of $1.96 a day on this single card.  Cool!       --scryptr

[crysx]

We're giving away 100% speedup on Lyra2REv2: https://www.nicehash.com/?p=news&id=80

Best regards,
NiceHash team.

Where did the lyra2v2 kernel come from? Is it open source?

djm34 created lyra2rev2 ...

he has always had the leading hand with this and the neoscrypt algo ...

cuda and these algos have been his forte for a long time now ...

#crysx

[Amph]

so we can except 20MH on a 970 with this new optimized version?

still something is tellimg me that it is not profitable

edit yes 20MH for each 970, so now mona and vertcoin are again profitable, good enough for me some backup when etheruem will fade away due to pos..

[giagge]

We're giving away 100% speedup on Lyra2REv2: https://www.nicehash.com/?p=news&id=80

Best regards,
NiceHash team.

Thanks .

[Velgelm]

so we can except 20MH on a 970 with this new optimized version?

still something is tellimg me that it is not profitable

edit yes 20MH for each 970, so now mona and vertcoin are again profitable, good enough for me some backup when etheruem will fade away due to pos..

agreed to :
so close :