All these people having PSU issues in this thread, it's like they've never built a system using 300 watt stock tbp gpus before.

How to size PSUs:

Step 1 - Add up the stock power of all of your gpus.

Step 2 - Divide by 0.8 so that if your OCs get reset and you don't notice you've sized to pull no more than 80% of the psu's rated power.

Step 3 - Take the combined power of your power supplies and make sure your wall socket can provide 15-25% more power than that without tripping the breaker.

Example:

You want a system with a cheap mobo + 4 Radeon VII gpus.

Step 1 - Power Draw = 60 watts (mobo) + 300 watts (gpu) x 4 = 1260 watts

Step 2 - 1260 / 0.8 = 1575 watts, or we need a PSU rated for 1600 watts or higher.

Step 3 - 1600 watts power draw * 1.25 = 2000 watts.  2000 watts / 110vac (rms) = 18 amps, or you need a 20 amp breaker dedicated to just this 1600 watt psu, powering just four Radeon VII gpus.


You really don't want to be maxing your psus out anyway for efficiency sake, you're burning more power by not running closer to the peak of the psu's efficiency curve.




On another topic, I've found that for ethash the point at which the gpu core no longer bottlenecks the memory on this gpu to be when the gpu clock is about 1.8-1.83x the memory clock.  Or 1800mhz core for 1000mhz mem, or 1620mhz for 900mhz mem.  Is everyone seeing similar in their fiddling?

Radeon VII has production issues with making the die package level. This made the company have to resort to using higher end thermal pads between the die/hbm and the heatsink.  This is a good bit inferior than a properly thin layer of thermal paste, however since not all die packages were level this level of thinness couldn't be garunteed. So, in order to ensure all cards perform at least within spec, AMD used a thicker solution, a thermal pad. 

Many have modded their cards to replace the pad with paste, hoping their dies are level enough. Others are tightening the mounting screws a bit more, and of source there are water cooling plates out there now, as well as people tossing on aftermarket coolers like the morpheus II.

Add this to the stock voltages being way to high for every clockspeed, and needing a bit of external airflow over the backplate (as would be had in a well ventilated case) and it results in many ways to address cooling. Don't just have this card sitting on a testbench....it needs a bit of airflow over it.

Finally don't forget that throttling temp is based of the new tjunct value, which is the highest temp in a much dense array of sensors as compared to hot spot temp value of vega 56/64.  This difference is more about a change in measurement resolution rather than the chip actually running any hotter or cooler than last gen. As long as your tjunct isn't throttling, pay more attention to your general core temps.

How goes progress on the new XMR algo for the upcoming fork?

Love the miner, would also love to see P2Pool support for vertcoin.

Did you try the different memory timing options?  Did they increase your hashrate or lower your stable overclock?

My order of acorns is also in fedex's hands and on the way. Scheduled to reach me this Wednesday.

I've not used any no, non-reference should run cooler, so many the gpu die would overclock better.  But generally your hashrate will be identical to mine if you run your gpu core at the same speed.  The only question would be if you would be stable at that speed at the same voltage level.

They're all vega64, all reference blower cards, and all from the same manufacturer.  Different gpu cores just overclock or undervolt better than others, it's the silicone lottery.

Since then I've turned down all cards to 1300mhz for long term better stability, but still running at those voltages.  They're getting about 72mh/s each now with the newest version of the Lyclminer.

I've got 2 cards at 1360mhz / 875mv gpu core getting 75mh/s
My other 4 cards at 1300mhz / 890mv gpu core getting 70mh/s
All vega 64s

The system with all the cards is pulling 1500watts from the wall.

Also you have to use 18.5.1 adrenalin drivers or newer, the blockchain drivers won't work for me.

There is none. It was written to claim a bounty and code was made public. No dev fee was part of the bounty requirement.

I see that there is a new open-source miner for Lyra2REv2 on AMD cards now: https://github.com/CryptoGraphics/lyclMiner

Performance seems better for me too 10%-15% faster at same settings on my Vega 64 cards, is mkxminer gonna be updated with code from this?

Any chance this miner will add stratum server support?

Has anyone gotten this miner to work mining on P2Pool?

Thanks for the heads up I'll check it out, would very much like to be mining lyra2rev2 right now instead of other algos with my cards, especially with networks needing all the hashrate/security they can get with the spate of 51% attacks coming from nicehash.

EDIT: 5/30/18 Seems the newest version (3.10) is now working normally again for me. I've not yet run it overnight, will report back with longer term results.

Edit: 6/1/18 The behavior of the v3.1 mkxminer is back to what it was before May 21rst.

Something was changed about this miner's devfee server on the night of May 21rst and it has made it completely unusable.  I'm fine with a fee existing, I knew about it when I chose to use the program, but this miner is borked right now.  Undo whatever was done to it on that night and the following days.

EDIT: 5/30/18 Seems the newest version (3.10) is now working normally again for me. I've not yet run it overnight, will report back with longer term results.

Edit: 6/1/18 The behavior of the v3.1 mkxminer is back to what it was before May 21rst.

Mining is a hobby not an investment.  So I knowingly leave some $$$ on the table, I don't want to increase the number of trades I have to report by auto-swapping, so I mine the coin I want to hodl.  On top of that my returns have been higher than most online calculators have predict, so on average my mining returns are fairly close to any other algorithm.

Periods like the past week or so with XMR and vega are a bit of an exception, but I have other reasons that I want to push and tune my vegas for lyra2re2 as well.

Speaking of which, I have all 6 of my vega 64s running at 60MH/s (barely) while at -25% power limit. (1460watts at the wall for the whole rig) 

I think I've hit the limit without further increasing voltage/power draw, as stability (and hashrate) goes down if all cards aren't running at the same speed, and it went unstable when I tried ~5mhz higher.  So I think I'm pretty much done.  I might try to see how far I can lower memory voltage....I probably have a bit more room there.  Helps with heat control.

Something else that gave me a little bit more OC headroom, is to find out the average voltage that your core oscillates around at your desired power limit, and edit your power play tables to have more options around that average.

In my particular case, my cards average at ~900mV at their power limit, so I have p-states at 800,850,875,900,925,950,1025,1100.  So a higher number of voltage options with smaller voltage jumps right around that average for the card to jump around to.  That is useful if like me you can't get your cards stable at a constant voltage.

Have 6xvega 64s, running at ~58-59MH/s each while pulling ~1470w from the wall while overclocked/undervolted.  Still dialing in optimal settings.  Using the newest version of the miner with intensity 23 and including the --asm command in my batch.



The miner will occasionally stall when trying to adjust difficulty, and just go idle......so I use HWinfo Alerts to run a script that restarts the miner if it detects my gpu temps fall for a minute.



Anyone also running these cards?  What settings do you use?

Thank you very much.  I was just coming to this thread to ask about stability for vega gpu mining.  I'll load this up tonight and give it a try.

That being said, a question for anyone, what drivers for windows 10 64bit do people recommend for Vega with this?  So far I've been using the blockchain drivers and been following the advice of this guide to disable/enable vegas in the device manager after each reboot and to turn off "EnableUlps" and "EnableCrossFireAutoLink" registry options. http://vega.miningguides.com/#DriverInstall

I've heard the latest video game drivers (Adrenalin Edition 18.3.4) may have included some mining improvements, but I don't know if that most applies to memory limited algorithms rather than GPU core based ones like Lyra2rev2.

My current rig has 3 vega 64 reference cards running at 1408mhz / 925mv core, each outputting ~55mh/s.