Haha - "sol/s"...  someone's been mining ZEC?

I have mostly 64s, but my one 56 is running around 1950 H/s at sub-160W ATW.  It doesn't seem to be the best asic quality, and is a bit old, so I would expect better from a larger sample set.  I also flashed to 64 rom a while ago, which is useless for CNv2 (unless you don't care about power use,) so even lower power use should be expected on a stock rom given that 56 mem runs at a lower voltage, while it's tighter mem timings allow similar results at lower clocks.

For 2K H/s (and best efficiency,) I would suggest a voltage setting of around 840mv (or less if you can get away w/ it,) while targeting a 1400 effective core clock.  I say effective, because the low voltage will almost certainly cause significant core clock throttling, so you will likely have to set your P7 clock much higher than 1400 to get to that level while under a mining load.  For example, I run a 1450 @ 835mv P7 setting, and only get a ~1360 effective clock out of it.  On some of my 64s, my clock setting ranges from 1430-1500 (@ 815-850mv) to get to ~1400 effective.  As for mem clock, I believe you're looking for a ~950Mhz setting for a 56 - higher is obviously better, if possible.

Regarding the PPT, it really is only necessary for two reasons.  The first is to lower your P0-5 power state voltages to allow undervolting P6/7 appropriately.  This is necessary via PPT, because P0-5 can't be set w/ ODNT/Wattman/etc, but power state voltage settings must be in ascending order.  In other words, w/o PPT, the floor for P6/7 voltage is whatever the stock P5 voltage is set to - around 1.1v iirc.  With that in mind, I set my P0-P5 @ 800-805mv via PPT, then tune P7 (while miner is running) as appropriate via ODNT. 1450 @ 840mv is prob a good P7 starting point while tuning.

The second purpose of PPT is to lock in your final settings once you're done tuning, as it seems the drivers don't seem to always honor the ODNT voltage settings on load, but if those same settings are written to PPT, they 'stick'.  This also technically removes the need for loading ODNT profiles at startup, though it doesn't hurt anything to continue to do so.

So in short, your steps are:

  1. Create a first PPT simply to drop your voltage floor
  2. Tune while miner is running to get to an appropriate/stable P7
  3. Recreate PPT using final P6/7 values

A couple other notes:

 - a cn_config setting (for trm) of 15+15 seems to be better than 16+14, at least for these lower clocks

 - CNv2, esp w/ the TRM rocketship, is an unstable beast power wise.  Settings which may seem stable at first (even for hours,) may lead to a crash at some point due to power starvation.  Even once you think you've found the correct settings, expect to have to dial voltages up (or clocks down) by increments of 5-10 over time to get to a truly stable setup.

 - You can create PPTs via ODNT, by running in admin mode, then clicking on the GPU icon (top left) and selecting 'PPTable Editor'.

 - As a starting point, here is my ODNT profile for my 56, though every card, properly tuned, will have different P7 values (also note my mem clock setting is for a 64 bios):



EDIT: I always run latest drivers - never had a problem w/ them other than when using stak/derivatives (TRM is an entirely separate codebase, so doesn't suffer from the same issues.)  It should be noted that the latest Adrenaline drivers allow setting P0-5 now, which should remove the need for PPT.  Sadly, based on my testing it would seem they forgot to extend this capability to the mem power states, which also act as voltage floors to the core.

Push up your core frequency, you will get 2150h/s at 1448/1100. If you have a higher core frequency, you will get more

Only 6 for "... Successfully initialized GPU X".

I can’t get more than 5 of my VEGA 56 to hash at the same time, currently running rigs with 8 x Vega 56

Celeron processor
Ubuntu mate 18.04
Amdgpupro 18.30

The miner initializes all 8 cards, all 8 cards show full activity on LEDs, but always 0 hashrate reported for 3 of them, only 5 ever have a hash rate above 0

Any ideas?

Win10 LTSB. 18.6.1, 8x Vega56.
It's not random, it's GPU 6 and GPU7 every time, and it doesn't show GPU6 and GPU7 at startup, it can be seen in the thread. Using -d 0,1,2,3,4,5,6,7, the problem remains

Yes, me too.

Will be back at my desk shortly and will provide better feedback. Meanwhile, can you describe the behavior those times you’re lucky and all cards initialize ok, does mining work fine after that in those cases?

I think we need to add an option of doing a staggered initialization and a delay, that no card starts mining before the full init process has completed.

The miner is actually missing on-cpu verification, we didn’t have time to add it, so that’s not the issue at least .

Guys,

I need help.


My mining rigs are riserless motherboards with 7x vega.

These boards have Celeron 3885 CPU that appears to be very slow.

When the miner starts, it in parallel initializes each card and connects to the pool. But once the shares start coming, the CPU is busy with something (probably verifying the found shares?) and never completes initialization of some cards. Sometimes 3 cards start, sometimes 6 cards start, sometimes all 7. But I have to manually stop and start the miner many times until I get lucky and initialize all 7 cards.

Is it possible to introduce some delay between card initialization and mining start, or disable CPU share verification?


Thanks!

Have your software work in ryzen apu?

Why not try the JCE CPU miner?