Re: [JCE] Ultrafast CN-Heavy/Tube/XHV miner, low power, Vega56 1750+h/s
Hi all,
looks like most of you appreciate the 0.33-sync, fine
I'll turn it into the official 0.33b8 soon
@Unclwish: did at least the -sync solved your problem of stuck thread?
Strange question from myself: i just plugged back my HD6950 that hasn't mined (except for quick tests) since the golden age of Claymore 9.7 and so far i cannot use multi_hash over 256 (=512Mb), while i got report from GitHub support tickets the value 864 was good. Did somebody managed to allocate >512M on those cards? I'm still using the drivers 15.11 on win8.1 i used with the card it replaced (a 7900), not tried yet to use the recommended 14.4
looks like most of you appreciate the 0.33-sync, fine
I'll turn it into the official 0.33b8 soon
@Unclwish: did at least the -sync solved your problem of stuck thread?
Quote
I don't understand where is this "33-sync" version
That's the lone .exe in the github page. Ensure it doesn't get blocked by an antivirus. It's a 64-bits app as before.Quote
I suggest you to contact cryptunit team to get listed in their website
good idea, i'll send them an email. They should also add Cast (good miner so far) and remove Claymore (obsolete).Quote
JCE, b5+ to sync versions requires more power than b4 version
yes, measured +20% speed and +6% power on RX. Still a good deal Quote
"GPU", "worksize" : 8, "alpha" : 64, "beta" : 16, "gamma" : 4, "delta" : 4, "epsilon" : 4, "zeta":4, "index" : 3, "multi_hash":944 },
That's very close to the max memory, and OpenCL may not let you alloc all in one thread. I strongly advise you use two threads with multi_hash = 464, and on the -sync version the double threads give really more speed. Only very old cards like HD7000 may mine better with one thread. Also try with beta = 8 for speed, it may be a bit better or worse, depends on the cards.Strange question from myself: i just plugged back my HD6950 that hasn't mined (except for quick tests) since the golden age of Claymore 9.7 and so far i cannot use multi_hash over 256 (=512Mb), while i got report from GitHub support tickets the value 864 was good. Did somebody managed to allocate >512M on those cards? I'm still using the drivers 15.11 on win8.1 i used with the card it replaced (a 7900), not tried yet to use the recommended 14.4
Catalyst 14.4 is a must to allocate full memory of the card. But beware they are very tricky. Some users raport no hashing at all.
Just to be sure that all my cards are ok I was hashing on B4 for 30h with average hashrate 2082 ( 3xHD 6990+CPU). No problem.
Now it is time to test 0.33-sync. Tomorrow first results.
One think. I noticed that Win10 gave me the notification: " Access to the graphic procesor was blocked"
Edit:
No go with 33-sync on HD6990. It initialize all threads. Set difficulty from the pool and soon after that miner crashes without any error. Windows is just closed. This the first time when your miner does not start at all.