Quote from: kenshirothefist on June 14, 2014, 04:05:28 PM

Quote from: Adz on June 13, 2014, 11:17:22 PM

Quote from: KiloWatts on June 13, 2014, 02:47:41 PM

Yep, it's undoubtedly broke. I'm going to stick to 1 algorithm since it seems Scrypt and N-Scrypt are neck-to-neck right now.

Yup i'm with you. Ill go back to standard mining until NiceHash Guys start posting what they are doing to address these issues

Hi. Much as we'd like to, we (NiceHash) can't do much to address various sgminer_v5 issues. NiceHash auto-switching service is running very stable, however, there are still some issues, need to be addressed regarding the sgminer_v5 itself. As I mentioned in the first post, we've sponsored the development of new sgminer, but haven't been involved in the development itself. sgminer_v5 is still in beta and mrbrdo, the lead developer of the sgminer kernel switching functionality obviously underestimated the difficulty of the kernel switching implementation. sgminer v5 with kernel switching has been tested in our labs and is working fine with basic (no overclocking or special tuning) GPU configuration (also, many users are successfully using this functionality on our service for quite some time now). However, AMD GPU world is extremely diverse and there are many combinations and configurations of various GPU+OS+drivers+overclocking+tuning that obviously raises issues and instability with kernel switching functionality as reported by the community and beta testers. I'm sure that mrbrdo and other sgminer developers will address all these issues and will have a stable working solution for various GPU+OS+drivers+overclocking+tuning combination in the near future. The community involvement is very important with reports on issues and help with testing - it has been like this from the early development of cgminer (I'm sure some of you remembers issues when fist Bitcoin ASICs arrived, etc.) and all later forks and versions. Therefore I ask you for help with reporting issues to https://github.com/sgminer-dev/sgminer/issues. The most useful reports are the one with 100% repeatable scenario and with the debug log files "sgminer ... -D 2>>sgminer.log" attached to issue reports.

One of the currently most problematic issue is the one with pool-gpu-engine and pool-gpu-memclock not being evaluated correctly when switching algorithms (https://github.com/sgminer-dev/sgminer/issues/265). This might cause freezes, lower hashrate, died GPU's, etc. Therefore I ask you to test the auto-switching functionality without setting different engine and mem clock per pool/algo - use the same engine and mem speed for all algos and use explicit low-high specification to disable engine/mem throttling. And start by using default/stock engine/settings! Also, make sure you specify low enough thread-concurrency for scrypt-N - test with various low enough settings if you're getting "-4 enqueueing error" and make sure you're using at least "worksize" : "256" for scrypt-N if you're seeing issues with scrypt-N (for now, worksize can't be configured in a per-pool fashion). Last, but not least, make sure you're using the latest version from https://nicehash.com/software/#sgminer. Here is an example of 100% working configuration for a rig with 5x Radeon R9 290 on Linux where switching between all algorithms with all combinations works 100% correct without any GPU issues and with decent hashrate

scrypt: 840 khs
scryptN: 420 khs
X11: 3.15 Mhs
X13: 2.25 Mhs
keccak: 390 Mhs

Code:

{
"pools" : [
   {
   "name" : "NiceHash_Scrypt_AS",
   "url" : "stratum+tcp://stratum.nicehash.com:4333",
   "user" : "BTC_addr",
   "pass" : "d=1024",
   "pool-algorithm" : "zuikkis",
   "pool-nfactor" : "10",
   "pool-thread-concurrency" : "20481",
   "pool-gpu-engine" : "947-947",
   "pool-gpu-memclock" : "1250-1250",
   "pool-gpu-threads" : "1"
   },
   {
   "name" : "NiceHash_Scrypt-N_AS",
   "url" : "stratum+tcp://stratum.nicehash.com:4335",
   "user" : "BTC_addr",
   "pass" : "d=512",
   "pool-algorithm" : "zuikkis",
   "pool-nfactor" : "11",
   "pool-thread-concurrency" : "20481",
   "pool-gpu-engine" : "947-947",
   "pool-gpu-memclock" : "1250-1250",
   "pool-gpu-threads" : "1"
   },
   {
   "name" : "NiceHash_X11_AS",
   "url" : "stratum+tcp://stratum.nicehash.com:4336",
   "user" : "BTC_addr",
   "pass" : "d=0.08",
   "pool-algorithm" : "darkcoin-mod",
   "pool-nfactor" : "10",
   "pool-thread-concurrency" : "20481",
   "pool-gpu-engine" : "947-947",
   "pool-gpu-memclock" : "1250-1250",
   "pool-gpu-threads" : "2"
   },
   {
   "name" : "NiceHash_X13_AS",
   "url" : "stratum+tcp://stratum.nicehash.com:4337",
   "user" : "BTC_addr",
   "pass" : "d=0.08",
   "pool-algorithm" : "marucoin-mod",
   "pool-nfactor" : "10",
   "pool-thread-concurrency" : "20481",
   "pool-gpu-engine" : "947-947",
   "pool-gpu-memclock" : "1250-1250",
   "pool-gpu-threads" : "2"
   },
   {
   "name" : "NiceHash_Keccak_AS",
   "url" : "stratum+tcp://stratum.nicehash.com:4338",
   "user" : "BTC_addr",
   "pass" : "d=1024",
   "pool-algorithm" : "maxcoin",
   "pool-nfactor" : "10",
   "pool-thread-concurrency" : "8960",
   "pool-gpu-engine" : "947-947",
   "pool-gpu-memclock" : "1250-1250",
   "pool-gpu-threads" : "1"
   }
],
"xintensity" : "300",
"vectors" : "1",
"kernel" : "ckolivas",
"lookup-gap" : "2",
"worksize" : "256",
"shaders" : "2560",
"thread-concurrency" : "20481",
"gpu-threads" : "1",
"gpu-engine" : "947-947",
"gpu-memclock" : "1250-1250",
"gpu-fan" : "0-95",
"gpu-powertune" : "20",
"temp-cutoff" : "95",
"temp-overheat" : "85",
"temp-target" : "75",
"auto-fan" : true,
"auto-gpu" : true,
"log" : "5",
"log-dateformat" : "1",
"failover-only" : true,
"failover-switch-delay" : "30",
"no-pool-disable" : true,
"scrypt" : true,
"queue" : "1",
"scan-time" : "7",
"expiry" : "28"
}

Regarding the dilemma on weather to use or not to use multi-switching functionality: well, as you've observed in the last few weeks there are truly big fluctuations in profitability for various algorithms and it looks like it's going to be this way in the near future as well. First there has only been Bitcoin - almost no work for miner - run and forget, then came Litecoin - more tuning for Scrypt needed; then came X-altcoins - lots of work for miner to keep up with different coin profitability - solved by multipools; and now there are X-algorithms - lots of work for miner to keep up with different algorithm profitability - to be solved by multi-algo-auto-switching. Said that - if you like to constantly watch algorithm profitability, then you can switch miners manually, but if you just want to run your miners and let them get best profit four you - well then you'll have to go with multi-algo-auto-switching. We're not quite there yet, but will be soon.

Thanks for this - very kind and informative.

I agree, a key way to keep multi-switching stable is by keeping the engine and memclock at the same rate. And first test with stock settings.

That being said, I'm going to test your example of a 100% stable multi-switching config. I myself have 290x's, but it should work out the same.