stevascha
Member
Offline
Activity: 312
Merit: 10
|
|
January 09, 2018, 10:23:48 AM Last edit: January 09, 2018, 11:48:59 AM by stevascha |
|
First use hex because it's easier to map the bits to cores. 255 = 0x00ff = 0b0000000011111111 Only 8 bits/cores for 16 threads? My previous post showed how to run 14 miner threads leaving one core (2 threads) free. Some users when mining 8 threads have reported better performance with affinity 0x5555.
Second, priority isn't very useful unless you're also running background tasks and want to prioritize between them and mining.
i didn't see what a different using hexa or dec im try 255, its running well 1700x yescryptr16 avx-sha around 1300khs (many reject) 4way-sha around 1200khs (many reject) avx2 around 1100khs (many reject) the reject appears after difficulty pool has changed im not tested long day because have ram problem stability (3200mhz) I didn't say there was a difference with hex, I said it's easier to understand. As an example which cores are used with affinity= 43690? If you think it's running well with 16 threads and affinity 255 then you don't understand, That's only 8 cores for 16 threads meaning you have 8 cores idle and 8 cores with 2 threads each. The rejects and the share flood are pool issues. yeah i dont understand difference core and thread more confused you say affinity 43690 and binary i try 0xfffd, 0xfefe, 0xffee, it decrease the gpu hash best so far 0x5555 because less decrease than others is it necessary use -t command for mapping?
|
|
|
|
moreAndMoreAndMore
Newbie
Offline
Activity: 3
Merit: 0
|
|
January 09, 2018, 11:04:16 AM |
|
Hi. I tried multiple times to use the API feature but I think I didn't get how it works. I tried to open in Chrome http://127.0.0.1:4048 but I received "ERR_EMPTY_RESPONSE", same if I try http://127.0.0.1:4048/apiI also tried with telnet 127.0.0.1 4048 but I don't receive any response there as well. I am using the release version for Windows x64 3.7.8 You need to enter a command in telnet to get anything, start with help. I tried to write something in the telnet but as soon as I write a character it closes the session. When I open the cmd I write "telnet 127.0.0.1 4048", then a black page with a cursor blinking appears and stays there. As soon as I write any character in the that page the telnet session closes and I am back to the cmd prompt.
|
|
|
|
4ward
Member
Offline
Activity: 473
Merit: 18
|
|
January 09, 2018, 11:21:34 AM |
|
Hi. I tried multiple times to use the API feature but I think I didn't get how it works. I tried to open in Chrome http://127.0.0.1:4048 but I received "ERR_EMPTY_RESPONSE", same if I try http://127.0.0.1:4048/apiI also tried with telnet 127.0.0.1 4048 but I don't receive any response there as well. I am using the release version for Windows x64 3.7.8 You need to enter a command in telnet to get anything, start with help. I tried to write something in the telnet but as soon as I write a character it closes the session. When I open the cmd I write "telnet 127.0.0.1 4048", then a black page with a cursor blinking appears and stays there. As soon as I write any character in the that page the telnet session closes and I am back to the cmd prompt. try this: https://github.com/JayDDee/cpuminer-opt/blob/master/api/websocket.htm
|
|
|
|
MagicSmoker
|
|
January 09, 2018, 02:15:58 PM |
|
I hope I'm doing something wrong because this is a real disappointment compared to Claymore CPU Miner 3.9 - I get around 400 H/s with my Ryzen 5 1600 on Cryptonight with Claymore but just under 200 H/s with this. I'm using the recommended exe for Ryzen - cpuminer-avx-sha.exe - but without any command line optimizations enabled. Desktop interactivity of this computer is also reduced running this miner.
|
|
|
|
4ward
Member
Offline
Activity: 473
Merit: 18
|
|
January 09, 2018, 02:22:23 PM |
|
I hope I'm doing something wrong because this is a real disappointment compared to Claymore CPU Miner 3.9 - I get around 400 H/s with my Ryzen 5 1600 on Cryptonight with Claymore but just under 200 H/s with this. I'm using the recommended exe for Ryzen - cpuminer-avx-sha.exe - but without any command line optimizations enabled. Desktop interactivity of this computer is also reduced running this miner.
Claymore would probably do much better for 2 reasons: a. it autodetects the amount of cache and amount of necessary threads try allocating 1 thread per each 2mb of L2 cache you have b. it supports memory locking mechanism that speeds up memory transfers cpuminer is good in many things, but cryptonight is not one of them )
|
|
|
|
StelioKontos
Sr. Member
Offline
Activity: 840
Merit: 267
Chad Hodler since 2013
|
|
January 09, 2018, 02:24:32 PM |
|
New in v3.7.9
Partial 4way optimizations for veltor, skunk, polytimos, lyra2z. Thank you for the update and for you great work. Is lyra2z same as lyra2z330 and lyra2zoin?
|
|
|
|
MagicSmoker
|
|
January 09, 2018, 02:42:58 PM |
|
Claymore would probably do much better for 2 reasons: a. it autodetects the amount of cache and amount of necessary threads try allocating 1 thread per each 2mb of L2 cache you have b. it supports memory locking mechanism that speeds up memory transfers
cpuminer is good in many things, but cryptonight is not one of them )
Aye, after writing this I tried setting "-t 8" as my CPU has 16MB of L3 cache, but there was no real change (if anything, it might have been a little slower!?). Also should have mentioned I always run miners as administrator just in case large memory pages can be used, but, apparently this is not the case here. Ah well, I'll keep my eye on this project anyway because I like rooting for the underdogs.
|
|
|
|
joblo (OP)
Legendary
Offline
Activity: 1470
Merit: 1114
|
|
January 09, 2018, 03:24:43 PM |
|
Claymore would probably do much better for 2 reasons: a. it autodetects the amount of cache and amount of necessary threads try allocating 1 thread per each 2mb of L2 cache you have b. it supports memory locking mechanism that speeds up memory transfers
cpuminer is good in many things, but cryptonight is not one of them )
Aye, after writing this I tried setting "-t 8" as my CPU has 16MB of L3 cache, but there was no real change (if anything, it might have been a little slower!?). Also should have mentioned I always run miners as administrator just in case large memory pages can be used, but, apparently this is not the case here. Ah well, I'll keep my eye on this project anyway because I like rooting for the underdogs. There's definitely a problem at your end. I get 280 H/s on a i7-6700K with 8 MB L3 cache and 4 threads. Check your CPU load to make sure all cores are working. You may have to set a custom cpu-affinity. I don't have a Ryzen so I don't have any personal experience but you could browse back in this thread. There have been discussions about this gfrom real Ryzen owners. But as someone else pointed out I'm falling behind with cryptonight for a couple if reasons. There is a lot of competition for cryptonight and I don't feel like enterring that race. One reason I haven't implemented large pages yet is the need to be admin to use it on Windows.
|
|
|
|
joblo (OP)
Legendary
Offline
Activity: 1470
Merit: 1114
|
|
January 09, 2018, 03:26:53 PM |
|
New in v3.7.9
Partial 4way optimizations for veltor, skunk, polytimos, lyra2z. Thank you for the update and for you great work. Is lyra2z same as lyra2z330 and lyra2zoin? No, lyra2z is for zcoin.
|
|
|
|
nizzuu
Full Member
Offline
Activity: 187
Merit: 100
Cryptocurrency enthusiast
|
|
January 10, 2018, 08:14:06 AM |
|
Interesting https://github.com/skywind3000/FastMemcpySeems that one differs from cpuminer-opt's one from avxdefs.h. Don't know if it's faster or not.
|
|
|
|
joblo (OP)
Legendary
Offline
Activity: 1470
Merit: 1114
|
|
January 10, 2018, 03:12:48 PM |
|
Interesting and probably faster because it's benchmark tested. He had the benefit of seeing the results and tweaking. I gave up on super-optimizing memcpy and went with a simpler approach because all I wanted was to avoid some the overhead to detect alignment, odd sizes and vector capabilities. Most of the memcpy in cpuminer is with aligned data and integral sizes.
|
|
|
|
Salazarian2
Newbie
Offline
Activity: 7
Merit: 0
|
|
January 11, 2018, 12:05:38 AM |
|
Finally I can mine with the tool on my ubuntu server and realized that the difficulty for cryptonight is pretty high thus resulting in a low mining performance. This has already been discussed before.
Now I have 10 ubuntu servers with ryzen 1700X CPUs.
Is there anyway to bundle the hashing power together to mine as one unit in order to get atleast the threshold needed to have a "stable" mining performance rather than just doing a stopNgo mining?
If all 10 servers mine with their standalone hashing power VS. bundling them as one hashing unit. Is there any way to do this to mine for nicehash servers?
Sincerely
|
|
|
|
cristiancs
Newbie
Offline
Activity: 8
Merit: 0
|
|
January 11, 2018, 01:56:54 AM |
|
Hi, i have a little problem, i have a dual E5649, when i set the thread count to 12 i have like 400 H/s with cryptonight, but when i set the thread count to 24 i have like 270 H/s, i was thinking that maybe if i set the --cpu-affinity to some value i can run another instance that use the other 12 cores.
|
|
|
|
joblo (OP)
Legendary
Offline
Activity: 1470
Merit: 1114
|
|
January 11, 2018, 03:19:21 AM |
|
Finally I can mine with the tool on my ubuntu server and realized that the difficulty for cryptonight is pretty high thus resulting in a low mining performance. This has already been discussed before.
Now I have 10 ubuntu servers with ryzen 1700X CPUs.
Is there anyway to bundle the hashing power together to mine as one unit in order to get atleast the threshold needed to have a "stable" mining performance rather than just doing a stopNgo mining?
If all 10 servers mine with their standalone hashing power VS. bundling them as one hashing unit. Is there any way to do this to mine for nicehash servers?
Sincerely
Stratum proxy.
|
|
|
|
joblo (OP)
Legendary
Offline
Activity: 1470
Merit: 1114
|
|
January 11, 2018, 03:28:37 AM |
|
Hi, i have a little problem, i have a dual E5649, when i set the thread count to 12 i have like 400 H/s with cryptonight, but when i set the thread count to 24 i have like 270 H/s, i was thinking that maybe if i set the --cpu-affinity to some value i can run another instance that use the other 12 cores.
How do you know it's not using both CPUs? What about the obvious, don't set the thread count and use the default? Also don't force me to make assumptions like what OS you're using, miner version, how/if you compiled etc.
|
|
|
|
Salazarian2
Newbie
Offline
Activity: 7
Merit: 0
|
|
January 11, 2018, 03:32:53 AM |
|
Finally I can mine with the tool on my ubuntu server and realized that the difficulty for cryptonight is pretty high thus resulting in a low mining performance. This has already been discussed before.
Now I have 10 ubuntu servers with ryzen 1700X CPUs.
Is there anyway to bundle the hashing power together to mine as one unit in order to get atleast the threshold needed to have a "stable" mining performance rather than just doing a stopNgo mining?
If all 10 servers mine with their standalone hashing power VS. bundling them as one hashing unit. Is there any way to do this to mine for nicehash servers?
Sincerely
Stratum proxy. Is there a tutorial how to establish it with xmr-stak or would I need a specific miner with the proxy support? Sincerely
|
|
|
|
nsummy
|
|
January 11, 2018, 03:40:52 AM |
|
Finally I can mine with the tool on my ubuntu server and realized that the difficulty for cryptonight is pretty high thus resulting in a low mining performance. This has already been discussed before.
Now I have 10 ubuntu servers with ryzen 1700X CPUs.
Is there anyway to bundle the hashing power together to mine as one unit in order to get atleast the threshold needed to have a "stable" mining performance rather than just doing a stopNgo mining?
If all 10 servers mine with their standalone hashing power VS. bundling them as one hashing unit. Is there any way to do this to mine for nicehash servers?
Sincerely
If you are pointing them at nicehash they are already being bundled. The scenario you are thinking of would only occur if you have 10 servers each individually mining to a wallet.
|
|
|
|
Salazarian2
Newbie
Offline
Activity: 7
Merit: 0
|
|
January 11, 2018, 03:43:20 AM |
|
Finally I can mine with the tool on my ubuntu server and realized that the difficulty for cryptonight is pretty high thus resulting in a low mining performance. This has already been discussed before.
Now I have 10 ubuntu servers with ryzen 1700X CPUs.
Is there anyway to bundle the hashing power together to mine as one unit in order to get atleast the threshold needed to have a "stable" mining performance rather than just doing a stopNgo mining?
If all 10 servers mine with their standalone hashing power VS. bundling them as one hashing unit. Is there any way to do this to mine for nicehash servers?
Sincerely
If you are pointing them at nicehash they are already being bundled. The scenario you are thinking of would only occur if you have 10 servers each individually mining to a wallet. All my 10 servers mine to the same wallet and same pool. Does this mean that their hash performance is bundled? So basicly the more servers I would have (due to current cryptonight situation) the better it is to achieve a more stable performance ratio? BTW my current performance: https://i.imgur.com/IaMrWvv.pngI want to avoid these downtimes...
|
|
|
|
cristiancs
Newbie
Offline
Activity: 8
Merit: 0
|
|
January 11, 2018, 04:17:38 AM |
|
Hi, i have a little problem, i have a dual E5649, when i set the thread count to 12 i have like 400 H/s with cryptonight, but when i set the thread count to 24 i have like 270 H/s, i was thinking that maybe if i set the --cpu-affinity to some value i can run another instance that use the other 12 cores.
How do you know it's not using both CPUs? What about the obvious, don't set the thread count and use the default? Also don't force me to make assumptions like what OS you're using, miner version, how/if you compiled etc. Sorry, I know im not using both looking at "top" https://i.imgur.com/zdz7I0B.pngInteresting i have a little better performance using taskset -c 0,1,2,3,4,5,6,7,8,9,10,11 before the ./cpuminer (i expected to get better performance with even or odd cores considering that when i run numactl -H i get the even cores on one node and odd on the other) If i don't set the thread count it uses all the cores but the performance drops to ~300 H/s. Miner version is 3.7.9 and was compiled following the instructions on https://github.com/JayDDee/cpuminer-opt/blob/master/RELEASE_NOTES , im mining XMR and the OS is Centos 7.
|
|
|
|
joblo (OP)
Legendary
Offline
Activity: 1470
Merit: 1114
|
|
January 11, 2018, 06:01:56 AM |
|
Hi, i have a little problem, i have a dual E5649, when i set the thread count to 12 i have like 400 H/s with cryptonight, but when i set the thread count to 24 i have like 270 H/s, i was thinking that maybe if i set the --cpu-affinity to some value i can run another instance that use the other 12 cores.
How do you know it's not using both CPUs? What about the obvious, don't set the thread count and use the default? Also don't force me to make assumptions like what OS you're using, miner version, how/if you compiled etc. Sorry, I know im not using both looking at "top" Interesting i have a little better performance using taskset -c 0,1,2,3,4,5,6,7,8,9,10,11 before the ./cpuminer (i expected to get better performance with even or odd cores considering that when i run numactl -H i get the even cores on one node and odd on the other) If i don't set the thread count it uses all the cores but the performance drops to ~300 H/s. Miner version is 3.7.9 and was compiled following the instructions on https://github.com/JayDDee/cpuminer-opt/blob/master/RELEASE_NOTES , im mining XMR and the OS is Centos 7. You're reading top wrong. With 12 threads it's using 12 of the 24 logical cores. You have 12 physical cores with 12 threads. You're jumping to the wrong conclusion. Now the issue is clear. Your problem is the number of miner threads, not the cpu affinity. For cryptonight the optimum thread count is: threads = L3 cache size / 2MB. Your CPUs have 12 MB each so that's 12 threads. Default affinity should spread them over all physical cores.
|
|
|
|
|