Pages:
Author

Topic: [ANN] ccminer 2.3 - opensource - GPL (tpruvot) - page 76. (Read 500193 times)

legendary
Activity: 2716
Merit: 1094
Black Belt Developer
GDDR5X GPUs still have issues with compatibility. A 1080 for instance will sometimes get the full 30% speed increase over a 1070, but often times it wont. About half the algos worth mining right now have problems with GDDR5X. You'll get anywhere from 10%-30% (usually around 15%). A 1080ti is no different. It's a shitshow, skein doesn't have problems.

why "half the algo", i only have bad result on ETH and feathercoin, all other algo i'm doing 60-70% more versus a 1070, with my 1080ti and 80-90% on skein/jha

You should be getting +30% compared to a 1070 with a 1080 and +85% compared to a 1070 with a 1080ti according to cuda counts and pricing. If you're getting 60-70% more then a 1070 with a 1080 you have something misconfigured with your 1070 as it's underperforming. A 1080 and 1080ti will underperform compared to a 1070, but not the opposite direction unless you have something setup wrong. GDDR5 is quite mature, algos have problems with GDDR5X.

As far as what algos are having problems, you've quoted someone of them. I'm not going to tell you about them though.

i said 70% more than a 1070 with a 1080ti not a 1080, 90% only on jha and skein, i'm not aware of any algo as bad as ETH and feathercoin, for a 1080ti, which i don't care anyway they are far from being the best algo to mine

They aren't 'as bad', Dagger is crippled. There are a couple others, but you're still going to be getting a speed boost, but instead of +85%, you're getting 40-50%.

what are those other bad algo with 50% only, i'm curious, i see that bitcore algo timetravel 10 is another one with 60%, which is not good

Timetravel (8 or 10) is just like x11, almost no ram usage.
If it looks slower on gddr5x, is just because of too low intensity.

Even with higher intensity (which I've already tried) it doesn't run at full speed. Protip, 1070 is also running at too low of intensity, so it doesn't change anything.

Yes it does. The 1070 and 1080 have the same ram size (the 1080ti has a little more, but not enough to increase intensity sensibly), but the 1080 has more cuda cores and slower ram, hence why it needs higher intensity. That doesn't necessarily mean you can take advantage of it, but I know it's true because I tried it myself.
sr. member
Activity: 472
Merit: 250
lux8.net
tried alexis miner on skein and it's perferom really better on my 1070 rig, but usually start only 4gpu's out of 6.
tried add --cuda-schedule 0, but it freeze my system for some time.
legendary
Activity: 1764
Merit: 1024
GDDR5X GPUs still have issues with compatibility. A 1080 for instance will sometimes get the full 30% speed increase over a 1070, but often times it wont. About half the algos worth mining right now have problems with GDDR5X. You'll get anywhere from 10%-30% (usually around 15%). A 1080ti is no different. It's a shitshow, skein doesn't have problems.

why "half the algo", i only have bad result on ETH and feathercoin, all other algo i'm doing 60-70% more versus a 1070, with my 1080ti and 80-90% on skein/jha

You should be getting +30% compared to a 1070 with a 1080 and +85% compared to a 1070 with a 1080ti according to cuda counts and pricing. If you're getting 60-70% more then a 1070 with a 1080 you have something misconfigured with your 1070 as it's underperforming. A 1080 and 1080ti will underperform compared to a 1070, but not the opposite direction unless you have something setup wrong. GDDR5 is quite mature, algos have problems with GDDR5X.

As far as what algos are having problems, you've quoted someone of them. I'm not going to tell you about them though.

i said 70% more than a 1070 with a 1080ti not a 1080, 90% only on jha and skein, i'm not aware of any algo as bad as ETH and feathercoin, for a 1080ti, which i don't care anyway they are far from being the best algo to mine

They aren't 'as bad', Dagger is crippled. There are a couple others, but you're still going to be getting a speed boost, but instead of +85%, you're getting 40-50%.

what are those other bad algo with 50% only, i'm curious, i see that bitcore algo timetravel 10 is another one with 60%, which is not good

Timetravel (8 or 10) is just like x11, almost no ram usage.
If it looks slower on gddr5x, is just because of too low intensity.

Even with higher intensity (which I've already tried) it doesn't run at full speed. Protip, 1070 is also running at too low of intensity, so it doesn't change anything.

There is two other ones, everyone has already mined them. I'd rather not deplete my earnings by talking about them here though. They are there and it definitely makes a GDDR5X cars much less lucrative.
legendary
Activity: 1510
Merit: 1003
i've managed to install 9 cards into one mb (asus strix z270f, windows 10 1703) but unfortunately none of ccminer forks support such a config
it just freezes the system when i run ccminer (tried tpruvot's 2.0, alexis's etc)
if i remove any card to get 8, ccminer works as usual

ccminer limit is 16, so should be another issue, try two instances to check

no, it will not work.
There is no miner on windows to work with > 8 cards.
Linux only
legendary
Activity: 1484
Merit: 1082
ccminer/cpuminer developer
i've managed to install 9 cards into one mb (asus strix z270f, windows 10 1703) but unfortunately none of ccminer forks support such a config
it just freezes the system when i run ccminer (tried tpruvot's 2.0, alexis's etc)
if i remove any card to get 8, ccminer works as usual

ccminer limit is 16, so should be another issue, try two instances to check
sp_
legendary
Activity: 2926
Merit: 1087
Team Black developer
Timetravel (8 or 10) is just like x11, almost no ram usage.
If it looks slower on gddr5x, is just because of too low intensity.

There is mem usage but you have linear memory reads so the cache will work at optimal conditions.
gddr5x is slower on algos that use random memory access in big buffers.  (ethereum, monero etc..)
legendary
Activity: 2716
Merit: 1094
Black Belt Developer
GDDR5X GPUs still have issues with compatibility. A 1080 for instance will sometimes get the full 30% speed increase over a 1070, but often times it wont. About half the algos worth mining right now have problems with GDDR5X. You'll get anywhere from 10%-30% (usually around 15%). A 1080ti is no different. It's a shitshow, skein doesn't have problems.

why "half the algo", i only have bad result on ETH and feathercoin, all other algo i'm doing 60-70% more versus a 1070, with my 1080ti and 80-90% on skein/jha

You should be getting +30% compared to a 1070 with a 1080 and +85% compared to a 1070 with a 1080ti according to cuda counts and pricing. If you're getting 60-70% more then a 1070 with a 1080 you have something misconfigured with your 1070 as it's underperforming. A 1080 and 1080ti will underperform compared to a 1070, but not the opposite direction unless you have something setup wrong. GDDR5 is quite mature, algos have problems with GDDR5X.

As far as what algos are having problems, you've quoted someone of them. I'm not going to tell you about them though.

i said 70% more than a 1070 with a 1080ti not a 1080, 90% only on jha and skein, i'm not aware of any algo as bad as ETH and feathercoin, for a 1080ti, which i don't care anyway they are far from being the best algo to mine

They aren't 'as bad', Dagger is crippled. There are a couple others, but you're still going to be getting a speed boost, but instead of +85%, you're getting 40-50%.

what are those other bad algo with 50% only, i'm curious, i see that bitcore algo timetravel 10 is another one with 60%, which is not good

Timetravel (8 or 10) is just like x11, almost no ram usage.
If it looks slower on gddr5x, is just because of too low intensity.
legendary
Activity: 3248
Merit: 1070
GDDR5X GPUs still have issues with compatibility. A 1080 for instance will sometimes get the full 30% speed increase over a 1070, but often times it wont. About half the algos worth mining right now have problems with GDDR5X. You'll get anywhere from 10%-30% (usually around 15%). A 1080ti is no different. It's a shitshow, skein doesn't have problems.

why "half the algo", i only have bad result on ETH and feathercoin, all other algo i'm doing 60-70% more versus a 1070, with my 1080ti and 80-90% on skein/jha

You should be getting +30% compared to a 1070 with a 1080 and +85% compared to a 1070 with a 1080ti according to cuda counts and pricing. If you're getting 60-70% more then a 1070 with a 1080 you have something misconfigured with your 1070 as it's underperforming. A 1080 and 1080ti will underperform compared to a 1070, but not the opposite direction unless you have something setup wrong. GDDR5 is quite mature, algos have problems with GDDR5X.

As far as what algos are having problems, you've quoted someone of them. I'm not going to tell you about them though.

i said 70% more than a 1070 with a 1080ti not a 1080, 90% only on jha and skein, i'm not aware of any algo as bad as ETH and feathercoin, for a 1080ti, which i don't care anyway they are far from being the best algo to mine

They aren't 'as bad', Dagger is crippled. There are a couple others, but you're still going to be getting a speed boost, but instead of +85%, you're getting 40-50%.

what are those other bad algo with 50% only, i'm curious, i see that bitcore algo timetravel 10 is another one with 60%, which is not good
legendary
Activity: 1764
Merit: 1024
GDDR5X GPUs still have issues with compatibility. A 1080 for instance will sometimes get the full 30% speed increase over a 1070, but often times it wont. About half the algos worth mining right now have problems with GDDR5X. You'll get anywhere from 10%-30% (usually around 15%). A 1080ti is no different. It's a shitshow, skein doesn't have problems.

why "half the algo", i only have bad result on ETH and feathercoin, all other algo i'm doing 60-70% more versus a 1070, with my 1080ti and 80-90% on skein/jha

You should be getting +30% compared to a 1070 with a 1080 and +85% compared to a 1070 with a 1080ti according to cuda counts and pricing. If you're getting 60-70% more then a 1070 with a 1080 you have something misconfigured with your 1070 as it's underperforming. A 1080 and 1080ti will underperform compared to a 1070, but not the opposite direction unless you have something setup wrong. GDDR5 is quite mature, algos have problems with GDDR5X.

As far as what algos are having problems, you've quoted someone of them. I'm not going to tell you about them though.

i said 70% more than a 1070 with a 1080ti not a 1080, 90% only on jha and skein, i'm not aware of any algo as bad as ETH and feathercoin, for a 1080ti, which i don't care anyway they are far from being the best algo to mine

They aren't 'as bad', Dagger is crippled. There are a couple others, but you're still going to be getting a speed boost, but instead of +85%, you're getting 40-50%.
member
Activity: 84
Merit: 10
i've managed to install 9 cards into one mb (asus strix z270f, windows 10 1703) but unfortunately none of ccminer forks support such a config
it just freezes the system when i run ccminer (tried tpruvot's 2.0, alexis's etc)
if i remove any card to get 8, ccminer works as usual
legendary
Activity: 3248
Merit: 1070
GDDR5X GPUs still have issues with compatibility. A 1080 for instance will sometimes get the full 30% speed increase over a 1070, but often times it wont. About half the algos worth mining right now have problems with GDDR5X. You'll get anywhere from 10%-30% (usually around 15%). A 1080ti is no different. It's a shitshow, skein doesn't have problems.

why "half the algo", i only have bad result on ETH and feathercoin, all other algo i'm doing 60-70% more versus a 1070, with my 1080ti and 80-90% on skein/jha

You should be getting +30% compared to a 1070 with a 1080 and +85% compared to a 1070 with a 1080ti according to cuda counts and pricing. If you're getting 60-70% more then a 1070 with a 1080 you have something misconfigured with your 1070 as it's underperforming. A 1080 and 1080ti will underperform compared to a 1070, but not the opposite direction unless you have something setup wrong. GDDR5 is quite mature, algos have problems with GDDR5X.

As far as what algos are having problems, you've quoted someone of them. I'm not going to tell you about them though.

i said 70% more than a 1070 with a 1080ti not a 1080, 90% only on jha and skein, i'm not aware of any algo as bad as ETH and feathercoin, for a 1080ti, which i don't care anyway they are far from being the best algo to mine
legendary
Activity: 2716
Merit: 1094
Black Belt Developer

GDDR5X GPUs still have issues with compatibility. A 1080 for instance will sometimes get the full 30% speed increase over a 1070, but often times it wont. About half the algos worth mining right now have problems with GDDR5X. You'll get anywhere from 10%-30% (usually around 15%). A 1080ti is no different. It's a shitshow, skein doesn't have problems.
These are not issues, but "GDDR5X features"
bandwidth is up, but latency/access for smaller chunks is not, thus algos that do not depend on bandwidth (skein), or can use memory in fully cached chunks scale with cores.
Algos which are critical with latency or non-cached memory access doesn't scale well

What? That's common bogus developer rhetoric. GDDR5X is broken for about half the algos. It needs to be fixed. It's not a feature until it actually does something remotely beneficial. And no it's not a 'oh it's latency and it's broken 4evarrrr' thing, algos need to be tuned to it.
Go read about GDDR5x
It uses same chips as simple gddr5, but with wider 64b per access (32b for gddr5)
Memory latency, access times, its all same.
GDDR5X memory rated 10GBps uses chips which are rated 1.25GBps
GDDR5 memory rated 5GBps uses same, rated 1.25GBps chips.
Bandwidth is up, latency and time to access are same. Think sdram-ddr-ddr2-ddr3-ddr4

Compare 1080 with  "9000Mhz memory" with 1070 using "8000Mhz memory"
1080 uses  1.25GBps chips, 1070 uses chips rated for 2.0GBps. They have way lower latency
Think about how Ryzen fails is some tests, because it s cache latency is much higher

Yeah, still have no idea what sort of a point you're making here. I in no way, shape, or form said that GDDR5X is terrible (it works well for some things, such as gaming), but when it comes to mining it doesn't work well. You essentially made all the points for me. It's terrible in any algo that actually has to use it and while some of them can be fixed (there are algos that utilize memory that don't take a hard hit from using GDDR5X, Equihash was fixed, Lyra2v2 was fixed), there are a decent chunk of them that can't, such as Ethereum.

Calling something a 'feature' when it's broken in the application we're talking about is silly.

equihash and lyra2v2 memory usage is not "fixed" for gddr5x, has just been reduced to almost 0 :-)
hero member
Activity: 968
Merit: 624
Still a manic miner
how do you name workers with ccminer @ yiimp?? so they get listed in summary below wallet address.

thanks

USE YOUR PASSWORD--

For example, "bilbo, c=VTC" will name your worker "bilbo" and specify VertCoin (VTC) for your payout address.  Your user name, "-u ", is your wallet address.  For some algos there is more than one coin being mined, so specifying the payout coin may be helpful.  It is not always necessary.

This information is in the readme for CCminer.       --scryptr

thanks a lot Scryptr.
i checked the readme before posting, but i missed the password as an option for naming workers.
much appreciated!  have a good hashing day!
member
Activity: 121
Merit: 10
what is this nonsense, g5x and g5 are identical Wink
legendary
Activity: 1764
Merit: 1024

GDDR5X GPUs still have issues with compatibility. A 1080 for instance will sometimes get the full 30% speed increase over a 1070, but often times it wont. About half the algos worth mining right now have problems with GDDR5X. You'll get anywhere from 10%-30% (usually around 15%). A 1080ti is no different. It's a shitshow, skein doesn't have problems.
These are not issues, but "GDDR5X features"
bandwidth is up, but latency/access for smaller chunks is not, thus algos that do not depend on bandwidth (skein), or can use memory in fully cached chunks scale with cores.
Algos which are critical with latency or non-cached memory access doesn't scale well

What? That's common bogus developer rhetoric. GDDR5X is broken for about half the algos. It needs to be fixed. It's not a feature until it actually does something remotely beneficial. And no it's not a 'oh it's latency and it's broken 4evarrrr' thing, algos need to be tuned to it.
Go read about GDDR5x
It uses same chips as simple gddr5, but with wider 64b per access (32b for gddr5)
Memory latency, access times, its all same.
GDDR5X memory rated 10GBps uses chips which are rated 1.25GBps
GDDR5 memory rated 5GBps uses same, rated 1.25GBps chips.
Bandwidth is up, latency and time to access are same. Think sdram-ddr-ddr2-ddr3-ddr4

Compare 1080 with  "9000Mhz memory" with 1070 using "8000Mhz memory"
1080 uses  1.25GBps chips, 1070 uses chips rated for 2.0GBps. They have way lower latency
Think about how Ryzen fails is some tests, because it s cache latency is much higher

Yeah, still have no idea what sort of a point you're making here. I in no way, shape, or form said that GDDR5X is terrible (it works well for some things, such as gaming), but when it comes to mining it doesn't work well. You essentially made all the points for me. It's terrible in any algo that actually has to use it and while some of them can be fixed (there are algos that utilize memory that don't take a hard hit from using GDDR5X, Equihash was fixed, Lyra2v2 was fixed), there are a decent chunk of them that can't, such as Ethereum.

Calling something a 'feature' when it's broken in the application we're talking about is silly.
member
Activity: 121
Merit: 20
Please, help to solve this errors. It's --benchmark mode. ccminer crashed or print 'unknown error' at any algorithm. Seems bug related to using of Cuda8.
I try to compile from cuda-8, linux, windows branches with the same effect and errors.
It's on Ubunty 17.10
Other miners with cuda on this machine works fine.

Quote
[2017-06-07 18:41:11] GPU #0: unknown error
[2017-06-07 18:41:11] GPU #1: scanhash_blake2s unknown error
[2017-06-07 18:41:12] GPU #0: scanhash_blake2s:433 unknown error
[2017-06-07 18:41:12] GPU #1: ends=7fffffff range=00000000
[2017-06-07 18:41:12] GPU #1: GeForce GTX 1070, 0.00 H/s
[2017-06-07 18:41:13] GPU #0: scanhash_blake2s unknown error
[2017-06-07 18:41:14] GPU #0: ends=00000000 range=00000000
[2017-06-07 18:41:14] GPU #0: GeForce GTX 1070, 0.00 H/s
[2017-06-07 18:41:14] GPU #1: bench_algo_switch_next:120 unknown error
[2017-06-07 18:41:16] GPU #0: bench_algo_switch_next:120 unknown error
[2017-06-07 18:41:19] GPU #1: blake2s hashrate = 0.00 H/s
[2017-06-07 18:41:19] GPU #0: blake2s hashrate = 0.00 H/s
[2017-06-07 18:41:20] Benchmark algo bmw...
[2017-06-07 18:41:20] GPU #0: start=00000000 end=3fffffff range=3fffffff
[2017-06-07 18:41:20] GPU #0: unknown error
[2017-06-07 18:41:20] GPU #0: cuda_check_cpu_init unknown error
[2017-06-07 18:41:21] GPU #1: start=7fffffff end=bffffffe range=3fffffff
[2017-06-07 18:41:21] GPU #1: unknown error
[2017-06-07 18:41:22] GPU #1: cuda_check_cpu_init unknown error
Cuda error in func 'scanhash_bmw' at line 59 : unknown error.
Cuda error in func 'scanhash_bmw' at line 59 : driver shutting down.


legendary
Activity: 1151
Merit: 1001

GDDR5X GPUs still have issues with compatibility. A 1080 for instance will sometimes get the full 30% speed increase over a 1070, but often times it wont. About half the algos worth mining right now have problems with GDDR5X. You'll get anywhere from 10%-30% (usually around 15%). A 1080ti is no different. It's a shitshow, skein doesn't have problems.
These are not issues, but "GDDR5X features"
bandwidth is up, but latency/access for smaller chunks is not, thus algos that do not depend on bandwidth (skein), or can use memory in fully cached chunks scale with cores.
Algos which are critical with latency or non-cached memory access doesn't scale well

What? That's common bogus developer rhetoric. GDDR5X is broken for about half the algos. It needs to be fixed. It's not a feature until it actually does something remotely beneficial. And no it's not a 'oh it's latency and it's broken 4evarrrr' thing, algos need to be tuned to it.
Go read about GDDR5x
It uses same chips as simple gddr5, but with wider 64b per access (32b for gddr5)
Memory latency, access times, its all same.
GDDR5X memory rated 10GBps uses chips which are rated 1.25GBps
GDDR5 memory rated 5GBps uses same, rated 1.25GBps chips.
Bandwidth is up, latency and time to access are same. Think sdram-ddr-ddr2-ddr3-ddr4

Compare 1080 with  "9000Mhz memory" with 1070 using "8000Mhz memory"
1080 uses  1.25GBps chips, 1070 uses chips rated for 2.0GBps. They have way lower latency
Think about how Ryzen fails is some tests, because it s cache latency is much higher
legendary
Activity: 1510
Merit: 1003
palit gtx 1080 founder edition 700mhs skein @ 140 watt (80%tdp). 5mhs/watt, 62 degrees C with 100% fan.
palit gtx 1080 gamerock        740mhs skein @ 165 watt, 4.4 mhs/watt, 58 degrees C with 75% fan

can you share your settings and system config?
Mine does not do more than 630mh @136w (76% TDP) 65C 70% Fan
System: 6x 1080s gaming x G1820 16gb ddr3 ram

It is just an old amd socket am2+ mobo with athlon x2, win7, alexis miner. Then limiting tdp in afterburner and raise clock and mem while it is stable.
legendary
Activity: 1797
Merit: 1028
how do you name workers with ccminer @ yiimp?? so they get listed in summary below wallet address.

thanks

USE YOUR PASSWORD--

For example, "bilbo, c=VTC" will name your worker "bilbo" and specify VertCoin (VTC) for your payout address.  Your user name, "-u ", is your wallet address.  For some algos there is more than one coin being mined, so specifying the payout coin may be helpful.  It is not always necessary.

This information is in the readme for CCminer.       --scryptr
hero member
Activity: 968
Merit: 624
Still a manic miner
how do you name workers with ccminer @ yiimp?? so they get listed in summary below wallet address.

thanks
Pages:
Jump to: