Ethereum GPU mining?

[DrMrLordX]

I'm not an electrician but I believe the units are grounded via the molex to sata plugs provided by your power supply.

Sorry was thinking of the jumper wire thing:

http://www.gobitgo.com/articles/1001/How-To-Correctly-Use-and-Install-PCI-E-Riser-Cables/

which is something you need to do when using motherboards that will have problems determining that a PCIe slot is in use by a video card otherwise. Anyway, I'm thinking those risers might not require that. Maybe?

Just a harsh reminder of the difficulty curve. I expected this to happen but not so quickly.

https://etherscan.io/charts/difficulty

Sorry to be the bringer of bad news.

Eh it ain't so bad. But I can see that as a deterrent from anyone wanting to join the fray now, or expand operations.

btw, my MSI 390 Gaming pulls maybe 290W from the wall on an EVGA P2 750 power supply. That's with a -100mv undervolt. My generic R9 270 pulls around 155W from the wall on the same setup. That's with VDDC set to 1.1v (small undervolt).

 

[Magic Carpet]

Would of joined with my single R9 270, but can't be bothered. Too much hassle. The other night spent $200 on drinks alone. This won't make it back, lmao.

Hope it all goes well for you, though! Good luck

 

[RussianSensation]

Does 400w for the total system sound right for a 390x?

Ya that sounds right without a platinum PSU and an older CPU platform but a tad on the high side though. XFX 390X and MSI Gaming 390X draw a lot more power unless undervolted.

At stock voltage R9 295X2 @ 1125mhz + 390 @ 1165mhz + stock i7 6700K (iGPU disabled) = 930-934W for me with a 1000W SeaSonic Platinum. I read that 1000W SeaSonic Platinum and 1250W SeaSonic Gold are very similar. Contemplating adding a 4th card into this rig but not sure it's worth the risk. Will do more reading on the PSU.

It's funny how so many PC builders skim on the PSU and get some 350/430/500/550W 80 Plus/Bronze/Silver junk for $40. This PSU paid for itself so many times over with Bitcoins, if the market had 1200-1600W Platinum units back then, would have stepped up . The EVGA Platinum 1200W P2 is $210-220. If I expand to more rigs, will probably get one of those.

Output bounces between 87-95MH or 30-31.5 per each GPU. I'll run some flag commands and post what gave me the highest average output.

The average electricity rate (averaged every hour in the week from off-peak/mid-peak/high-peak) is $0.1075 CAD per kWh or $0.0839 USD per kWh. On weekends, it's $0.083 CAD per kWh 24/7 ($0.0648 USD). Electricity is cheap in Ontario.

 

[Madpacket]

Sorry was thinking of the jumper wire thing:

http://www.gobitgo.com/articles/1001/How-To-Correctly-Use-and-Install-PCI-E-Riser-Cables/

which is something you need to do when using motherboards that will have problems determining that a PCIe slot is in use by a video card otherwise. Anyway, I'm thinking those risers might not require that. Maybe?



Eh it ain't so bad. But I can see that as a deterrent from anyone wanting to join the fray now, or expand operations.

btw, my MSI 390 Gaming pulls maybe 290W from the wall on an EVGA P2 750 power supply. That's with a -100mv undervolt. My generic R9 270 pulls around 155W from the wall on the same setup. That's with VDDC set to 1.1v (small undervolt).

I'm not sure if it's because of the risers but I don't need any dummy plugs or do any trickery to the PCIe ports. They just work. For mining motherboards I stick to Gigabyte GA-990FXA-UD3 Rev 4 or 5. Throw in a Sempron 140 + 4GB of RAM and you have 6 PCIe ports to work with.
These boards are very well built (compared to the AsRock BTC boards) and you can disable all on-board power sucking devices to bring down the energy consumption.

That 290W you're measuring, is that just the card or the entire PC?

My 4xMSI 390's running at 1111Mhz core, 1275 Mem, all undervolted to -100 running in the above mentioned board pulls 950-1025W from the wall. That's on a EVGA 1300W G2 I bought on sale so factor in approx 10% losses and I'm not doing too bad.

Mining speeds bounce between 115-125Mh on this setup.

 

[Madpacket]

Ya that sounds right without a platinum PSU and an older CPU platform but a tad on the high side though. XFX 390X and MSI Gaming 390X draw a lot more power unless undervolted.

At stock voltage R9 295X2 @ 1125mhz + 390 @ 1165mhz + stock i7 6700K (iGPU disabled) = 930-934W for me with a 1000W SeaSonic Platinum. I read that 1000W SeaSonic Platinum and 1250W SeaSonic Gold are very similar. Contemplating adding a 4th card into this rig but not sure it's worth the risk. Will do more reading on the PSU.

It's funny how so many PC builders skim on the PSU and get some 350/430/500/550W 80 Plus/Bronze/Silver junk for $40. This PSU paid for itself so many times over with Bitcoins, if the market had 1200-1600W Platinum units back then, would have stepped up . The EVGA Platinum 1200W P2 is $210-220. If I expand to more rigs, will probably get one of those.

Output bounces between 87-95MH or 30-31.5 per each GPU. I'll run some flag commands and post what gave me the highest average output.

The average electricity rate (averaged every hour in the week from off-peak/mid-peak/high-peak) is $0.1075 CAD per kWh or $0.0839 USD per kWh. On weekends, it's $0.083 CAD per kWh 24/7 ($0.0648 USD). Electricity is cheap in Ontario.

Not bad consumption for those clocks. FYI I was going to grab the 1200W EVGA Platinum but it's prohibitively expensive here so I went for the 1300W Gold and saved around 150.00. Both are SuperFlower Leadex designs. I do have the 1KW Platinum though and it's stellar. I also just picked up 1050GS for a great price (this one's SeaSonic based) and I'm tempted to run 4 cards off one of them.

I may try going with 4xPowercolor 390's instead of MSI this time as I think they might be better for power consumption (that link you provided me showed the PowerColor card leading in power consumption). Need to do some testing first though to be certain.

Although to be fair only my MSI's have constantly been capable of -100mv / 1111Mhz on all cards where one of my two powercolor cards can only run -87mv.

I really want an XFX and Sapphire as well (for comparison reasons). The only card I have no desire to pickup is the Club3D ones due to the fans used and questionable RMA process. All manufactures appear to use different VRM designs so it's very hare to compare in a mining scenario.

Also a lot depends on ASIC rating. The lower the ASIC rating, the more power consumption (on average) when comparing cards at equal clock speed.

 

[Madpacket]

How long do we have left before our earnings drop by 1/2 due to a doubling in difficulty?

If you look at that curve and assume it will continue in that direction then not very long.

During BTC days, the market value of the coin adjusted as difficulty increased and it became more scarce. Seems like that's not happening with Ether at all.

To be fair I don't think we can compare the two outside of speculation reasons. Bitcoin was first and the price was driven by a bunch of dissimilar events. Ethereum is quite different for a few reasons. The most obvious being;

- 12 second Blocktimes.
- No FPGA's or ASICS to worry about.
- Much of the regulation around crypto currencies have already been established.
- Ethereum is not a coin, it's a platform for contracts and dapps and once we hit proof of stake GPU mining may be essentially rendered useless, so the glut of new coins everyday will go away. This could drive up the price per coin considerably as scarcity will then be a factor. It all depends on how PoS is implemented though...
- Lack of awareness (two major conference are scheduled this year)
- many more experienced Bulls and Bears manipulating Ethereum via day trading.
- etc.

Of course you probably know all of this, I'm just listing it for others who may be wondering.

 

[VeryCharBroiled]

These boards are very well built (compared to the AsRock BTC boards) and you can disable all on-board power sucking devices to bring down the energy consumption.

back in the day when I BTC mined on several old PCs with VGAs disabling on board stuff like audio, com ports, unused SATAs etc there was no power savings at all. it just freed up IRQs and I/O addys.

all power savings came from undervolting and downclocking the CPU and VGAs.

these were LGA 775 mobos. as I said this was back in the day

mayhaps newer boards are different.

 

[Madpacket]

Hmm, on closer inspection comparing the powercolor and club3d cards (at least from the pictures) appear to be the same design. Perhaps they use the same ODM...

 

[Madpacket]

back in the day when I BTC mined on several old PCs with VGAs disabling on board stuff like audio, com ports, unused SATAs etc there was no power savings at all. it just freed up IRQs and I/O addys.

all power savings came from undervolting and downclocking the CPU and VGAs.

these were LGA 775 mobos. as I said this was back in the day

mayhaps newer boards are different.

The 990FX is ancient, some of the added bits (extra USB 3.0 controllers, eSATA ports, sound etc) not native to the chipset appears to save a few watts. I turn off all sata ports not necessary (hot swap ones use standby power I believe). And yes I do undervolt and underclock everything else on the board which I agree you get more gains (err or losses) from. The Sempron 140 or 145's (I have one of each) both consume very little power. I also use SSD's to save additional power. Every watt counts!

For the rig I'm building now I only have a Phenom 965 left but I was able to test it out on the bench a few weeks ago on a rev 4.0 board. You can disable all but one core and then set it to 2.0Ghz and undervolt it quite a bit, reduce NB and other voltages, lower hypertransport speeds etc. Thankfully when you do all of this the power hungry 965 acts a lot like a single core Sempron but it still uses a few watts more under light loads.

 

[DrMrLordX]

I'm not sure if it's because of the risers but I don't need any dummy plugs or do any trickery to the PCIe ports. They just work.

It seems to vary from board to board/chipset to chipset.

That 290W you're measuring, is that just the card or the entire PC?

Just the card. Here are some measurements + estimates I did on my 7700k system using a kill-a-watt:

idle no cards: 53W AC
idle 270 + 390: 93W AC

estimated combined idle power: 40W AC (10W 270, 30W 390)

Mining load 270 + 390: 500W AC

estimated 270: ~156W AC/140W DC
estimated 390: ~290W AC/262W DC

Note that those are combined figures, and I also didn't take into account power draw from the CPU being more active during mining. It is overclocked/overvolted so that I can go about my normal routine (not what I did during measurements mind you; nothing but mining stuff was active during that period).

My 4xMSI 390's running at 1111Mhz core, 1275 Mem, all undervolted to -100 running in the above mentioned board pulls 950-1025W from the wall.

Wish my 390 would do 1111 MHz GPU. I see you underclocked your memory, though, which may be saving you a good bit of power. Mine's running at 1500 MHz.

 

[Madpacket]

It seems to vary from board to board/chipset to chipset.



Just the card. Here are some measurements + estimates I did on my 7700k system using a kill-a-watt:

idle no cards: 53W AC
idle 270 + 390: 93W AC

estimated combined idle power: 40W AC (10W 270, 30W 390)

Mining load 270 + 390: 500W AC

estimated 270: ~156W AC/140W DC
estimated 390: ~290W AC/262W DC

Note that those are combined figures, and I also didn't take into account power draw from the CPU being more active during mining. It is overclocked/overvolted so that I can go about my normal routine (not what I did during measurements mind you; nothing but mining stuff was active during that period).



Wish my 390 would do 1111 MHz GPU. I see you underclocked your memory, though, which may be saving you a good bit of power. Mine's running at 1500 MHz.

Hmm, you shouldn't see any real load when mining. My Sempron 140 set to 2.0 Ghz handles 4 390's at very low loads (ethminer uses less than 10%).

Your load power measurements sound about right given your not undervolting. The MSI's are very power hungry at default volts let alone overvolted. One thing you may want to do is just create a profile for mining and profile for gaming and then just switch depending on what you're doing. I suspect your computer spends more time idle / mining than gaming so it's probably worthwhile unless you're not paying for electricity

Also looking at your idle measurements, they're likely wrong. There should be less than a 5 watt difference between both cards at idle unless you've disabled some of the power management features.

Edit;;

Yes undervolting memory saves quite a bit. Remember you can't really use more than 1250Mhz of bandwidth with ethminer on a 390 so if set any higher you're just wasting electricity. Not sure about your 270 but you can easily check with GPU-Z. Just watch the memory controller load and adjust accordingly with Afterburner.

Also if you get a moment can you tell me what your ASIC % rating is for your 390?

 

[Feld]

It seems to vary from board to board/chipset to chipset.



Just the card. Here are some measurements + estimates I did on my 7700k system using a kill-a-watt:

idle no cards: 53W AC
idle 270 + 390: 93W AC

estimated combined idle power: 40W AC (10W 270, 30W 390)

Mining load 270 + 390: 500W AC

estimated 270: ~156W AC/140W DC
estimated 390: ~290W AC/262W DC

Note that those are combined figures, and I also didn't take into account power draw from the CPU being more active during mining. It is overclocked/overvolted so that I can go about my normal routine (not what I did during measurements mind you; nothing but mining stuff was active during that period).



Wish my 390 would do 1111 MHz GPU. I see you underclocked your memory, though, which may be saving you a good bit of power. Mine's running at 1500 MHz.

I'm running 2 MSI 390 Gaming cards, both undervolted -75 mv, one clocked at 1229 which draws 218 W, and the other at 1100 which draws 168 W. If you're certain your stock clocked MSI 390 is pulling 290 W, then I suspect your undervolt hasn't been applied. Perhaps you moved the voltage slider down in Afterburner but forgot to click Apply?

 

[DrMrLordX]

I clicked "apply", and I can see changes in power draw after doing so. The only thing I can think of that's throwing off my measurements is CPU power draw during mining load. I could try undervolting the CPU and redoing the measurements at lower CPU clock speeds.

edit: ASIC quality on the 390 is 77.1%.

 

[Feld]

I clicked "apply", and I can see changes in power draw after doing so. The only thing I can think of that's throwing off my measurements is CPU power draw during mining load. I could try undervolting the CPU and redoing the measurements at lower CPU clock speeds.

edit: ASIC quality on the 390 is 77.1%.

Your ASIC quality is very good. Mine are 73.2% and 70.7%.

 

[Madpacket]

I clicked "apply", and I can see changes in power draw after doing so. The only thing I can think of that's throwing off my measurements is CPU power draw during mining load. I could try undervolting the CPU and redoing the measurements at lower CPU clock speeds.

edit: ASIC quality on the 390 is 77.1%.

That's a great ASIC quality. Of the seven MSI 390's, I think 76% is the highest I have with an average closer to 70%. What does task manager report for ethminer.exe CPU usage?

 

[Feld]

Also, by dropping the VRAM clocks to 1300 it saves about 5 watts per card with no difference in mining performance.

 

[DrMrLordX]

Okay:

1). Task manager is showing 5% CPU utilization. Coretemp (finally updated after all these years; I'm using 1.0 RC8) shows 1-2 cores at 4-5% with less activity on the others.

2). Despite the high ASIC quality, the 390 is sucking up moar powar than my 290 with an ASIC quality of 70.1%. The problem is that the default VDDC for the 390 appears to be 1.191v, so -100mv only gets it to 1.094v. In contrast, the 290 I'm using in this machine right now is running at 1.063v thanks to a BIOS flash. The 390 is now flashed successfully, but it'll take me pulling the card or reinstalling the driver for the changes to take effect I think.

3). According to my Kill-a-Watt, mining with both cards pulls 550W from the wall. If I shut down the 290, it goes down to 345W. If I stop the 390, it goes down to 315W.

Bear in mind that I have both cards running their memory @ 1250 MHz now. They were at 1500 MHz.

 

[Madpacket]

Nice results. 550W after tweaks sounds about right. Yes the MSI Gamin' cards have a higher default voltage than the older 290's (probably to guarantee 1060Mhz clocks). It also has twice the VRAM which of course uses more energy.

Did you flash your 390 to a different BIOS that uses less energy?

 

[Madpacket]

Also, by dropping the VRAM clocks to 1300 it saves about 5 watts per card with no difference in mining performance.

I think you can go as low as 1200 but I set all mine to 1250 to leave a bit of breathing room.

 

[DrMrLordX]

Okay, so I messed with my 390 and the XFX 290 that's in the system now. I got total load power at the wall for the entire system down to ~525W. A simple reboot (power down + power up didn't work) finally got the BIOS changes on the 390 to take. What's interesting is that the VRMs on the 390 respond differently to static voltage inputs than the 290s. 1150 mv on the 290 produces a measured voltage ~.025-.03v lower than the same setting on the 390. So I knoced the 390 down to 1112mv and it's now measured at 1.047v

Basically:

390: 1150mv = ~1.09v
1112mv = 1.047v (huh?)

290: 1150mv = 1.055-1.063v (fluctuates)

I also dropped VDDCI on both cards to .975v. Wonder how low I could get with that before running into problems?

 

[poofyhairguy]

I dropped the memory down to 1275 and went -11mv but kept the 50%+ power level at 1165 core clocks. Same mining speed but 380w! Thanks all.

 

[VeryCharBroiled]

Your ASIC quality is very good. Mine are 73.2% and 70.7%.

mine is 67.3% sapphire nitro r9 390.

not that I really care as long as I can OC it and undervolt it fairly well. I bought it for 3 fans and quite running as its in my living room daily driver in NZXT H2 sound deadening case. we want it fairly quiet.

 

[RussianSensation]

This couldn't be better timing for AMD, they should be thanking us Ethereum miners for clearing out inventories which will allow the huge e-tailers the necessary room and capital to buy stockpiles of Polaris cards.

Someone at AMD seems to be listening. Newegg.com (US) now had 300 new XFX R9 390 cards and 257 PowerColor PCS+ R9 390. There are also at least 10 Gigabyte 390s in stock but since the limit is 10 per customer, I cannot check the actual shipment. Remember not long ago all of these R9 390 cards were completely sold out which means all of these are new shipments.The prices are holding steady as the cheapest R9 390 is the PowerColor one for $320 USD after rebates. Demand must be through the roof since that's essentially MSRP price nearly a year ago.

I cannot even imagine the flood of used R9 290/390/290X/390X cards once they no longer make sense in terms of price/MH and MH/watt. The first person on our forum to post Polaris 10 & GP104 MH/sec rates will save us A LOT of headache whether to resell Hawaii ASAP or just keep using it until Vega/Big Pascal.

 

[Accord99]

I also dropped VDDCI on both cards to .975v. Wonder how low I could get with that before running into problems?

I haven't had problems with setting a 290X to -100 mV on Afterburner but I'm only running it at 900/1250 MHz.

 

[DrMrLordX]

I haven't had problems with setting a 290X to -100 mV on Afterburner but I'm only running it at 900/1250 MHz.

You can tweak VDDCI in Afterburner? Cool. I can't do it on any of my cards. Though BIOS h4x0ring it down to .900v showed no problems. Huzzah!