Ethereum mining still profitable?

[fricircled]

Do you use Windows MSI Afterburner to undervolate the 390? What is the undervolate level? What is the maximum you can do? Is that over -100 mV?

Yes, afterburner. Maximum is -100mV on both cards.

How much is your power consumption for the XFX 390 cards? What is your clock frequency and the power consumption read from the GPUZ?

I did a test, took one old s775 mobo with E8400 + 2 x 2GB DDR2, slammed in two 390's (8GB each), laptop hdd. PSU is not rated corsair CX750M, voltage 228V. Windows 7 x64. Asus has hynix memory and xfx elpida.

Clocks 1125/1500, limiting factor is core. Memory controller load is max 80%.

From the wall:

idle ~100W
mining 680W
mining -100mV 580W

Hashrate 30MH+ 30MH- around 30MH/card. Block  .

The Power in is 136W, 2x = 272W, 580-272 = 310W, the idle = 100w, so there is 210W difference. Are they from the fans or you have a very inefficient power supply?

I don't trust software when calculating power consumption, they always show too low numbers, both nvidia&amd. Took another card out and power consumption went down 280W.

At those loads cx750m efficiency is ~86%.

-edit- Did some cleaning. Idle 90W, mining 1125/1200, core & aux -100mV, 525W from the wall.

What was the aux voltage before the cleaning? Do you believe it is due to the memory frequency going down from 1500 to 1200? Did you see the difference in power in/out?

[antantti]

Do you use Windows MSI Afterburner to undervolate the 390? What is the undervolate level? What is the maximum you can do? Is that over -100 mV?

Yes, afterburner. Maximum is -100mV on both cards.

How much is your power consumption for the XFX 390 cards? What is your clock frequency and the power consumption read from the GPUZ?

I did a test, took one old s775 mobo with E8400 + 2 x 2GB DDR2, slammed in two 390's (8GB each), laptop hdd. PSU is not rated corsair CX750M, voltage 228V. Windows 7 x64. Asus has hynix memory and xfx elpida.

Clocks 1125/1500, limiting factor is core. Memory controller load is max 80%.

From the wall:

idle ~100W
mining 680W
mining -100mV 580W

Hashrate 30MH+ 30MH- around 30MH/card. Block  .

The Power in is 136W, 2x = 272W, 580-272 = 310W, the idle = 100w, so there is 210W difference. Are they from the fans or you have a very inefficient power supply?

I don't trust software when calculating power consumption, they always show too low numbers, both nvidia&amd. Took another card out and power consumption went down 280W.

At those loads cx750m efficiency is ~86%.

-edit- Did some cleaning. Idle 90W, mining 1125/1200, core & aux -100mV, 525W from the wall.

What was the aux voltage before the cleaning? Do you believe it is due to the memory frequency going down from 1500 to 1200? Did you see the difference in power in/out?

aux -100mV adds another 15W of powersaving, from 540 to 525W. Memory downclock savings are minimal, hard to see at the wall.

No difference in power in/out. 1125/1200 is the ratio where memory controller load is 96-97%.

[RustyNoman]

What was the aux voltage before the cleaning? Do you believe it is due to the memory frequency going down from 1500 to 1200? Did you see the difference in power in/out?

aux -100mV adds another 15W of powersaving, from 540 to 525W. Memory downclock savings are minimal, hard to see at the wall.

No difference in power in/out. 1125/1200 is the ratio where memory controller load is 96-97%.

I thought the IC power consumption is linearly proportional to the circuit frequency. Why the drop from 1500 to 1200 MHz has not effect?

[antantti]

What was the aux voltage before the cleaning? Do you believe it is due to the memory frequency going down from 1500 to 1200? Did you see the difference in power in/out?

aux -100mV adds another 15W of powersaving, from 540 to 525W. Memory downclock savings are minimal, hard to see at the wall.

No difference in power in/out. 1125/1200 is the ratio where memory controller load is 96-97%.

I thought the IC power consumption is linearly proportional to the circuit frequency. Why the drop from 1500 to 1200 MHz has not effect?

I thought so too but it looks like we were wrong. It does have an effect as long as core can utilize memory controller 100%. Lower mem - lower hash and power consumption, higher mem - nothing happens, mc just sits and waits for work.

I went so far that I tested 720/750, again same ratio, mc load close to 100%.

I didn't have time to verify this but it looks like 390 could be the first no Nano 20MH/100W card. 310W from the wall and ethminer hashrate somewhere at 40MH area.

[rossi.joel]

I think I would recommend Etherium mining profitable time to invest now

[tolazy]

did the difficulty go up in the last few days?

mining @ http://dwarfpool.com/ and uses there proxy software.
i did go from 0.9 ETH/day to 0.7 ETH/day

anyone with a 390x ? i get about 28MH on my card stock clock
any tips on settings for a 390x i use
ethminer --opencl-platform 1 --opencl-device 0 -G -F http://URL/gpu1

[Kalder]

did the difficulty go up in the last few days?

mining @ http://dwarfpool.com/ and uses there proxy software.
i did go from 0.9 ETH/day to 0.7 ETH/day

anyone with a 390x ? i get about 28MH on my card stock clock
any tips on settings for a 390x i use
ethminer --opencl-platform 1 --opencl-device 0 -G -F http://URL/gpu1

You can find the difficulty there. http://etherscan.io/charts/difficulty
It rose a lot recently since the price of Ethereum rose.

[Kwiksave]

What was the aux voltage before the cleaning? Do you believe it is due to the memory frequency going down from 1500 to 1200? Did you see the difference in power in/out?

aux -100mV adds another 15W of powersaving, from 540 to 525W. Memory downclock savings are minimal, hard to see at the wall.

No difference in power in/out. 1125/1200 is the ratio where memory controller load is 96-97%.

I thought the IC power consumption is linearly proportional to the circuit frequency. Why the drop from 1500 to 1200 MHz has not effect?

I thought so too but it looks like we were wrong. It does have an effect as long as core can utilize memory controller 100%. Lower mem - lower hash and power consumption, higher mem - nothing happens, mc just sits and waits for work.

I went so far that I tested 720/750, again same ratio, mc load close to 100%.

I didn't have time to verify this but it looks like 390 could be the first no Nano 20MH/100W card. 310W from the wall and ethminer hashrate somewhere at 40MH area.

From your observation, the hash rate is limited by the Core frequency. The more core, the better. So the 390x should have higher hash rate, but it is at best same as 390. Quite strange.

[antantti]

What was the aux voltage before the cleaning? Do you believe it is due to the memory frequency going down from 1500 to 1200? Did you see the difference in power in/out?

aux -100mV adds another 15W of powersaving, from 540 to 525W. Memory downclock savings are minimal, hard to see at the wall.

No difference in power in/out. 1125/1200 is the ratio where memory controller load is 96-97%.

I thought the IC power consumption is linearly proportional to the circuit frequency. Why the drop from 1500 to 1200 MHz has not effect?

I thought so too but it looks like we were wrong. It does have an effect as long as core can utilize memory controller 100%. Lower mem - lower hash and power consumption, higher mem - nothing happens, mc just sits and waits for work.

I went so far that I tested 720/750, again same ratio, mc load close to 100%.

I didn't have time to verify this but it looks like 390 could be the first no Nano 20MH/100W card. 310W from the wall and ethminer hashrate somewhere at 40MH area.

From your observation, the hash rate is limited by the Core frequency. The more core, the better. So the 390x should have higher hash rate, but it is at best same as 390. Quite strange.

It is. Based on my tests to get mc 100% loaded at 1500MHz you would need 1400 core which is obviously impossible.

[zalucia]

Can you speculate why the 390x is not faster than the 390 at the same frequency. It has more core, 2816 vs. 2560.

[antantti]

Can you speculate why the 390x is not faster than the 390 at the same frequency. It has more core, 2816 vs. 2560.

I have no idea. 10% difference should be easy to see, anyone with both 390 and 390x?

[tolazy]

done a try to OC a bit now. from 1080 core til 1150 core volt +25 and power limit +50
did go from 28.4 to 30.3Mhs
temps are around 70-72c with fans on auto around 45%

[bluebox]

Can you speculate why the 390x is not faster than the 390 at the same frequency. It has more core, 2816 vs. 2560.

I have no idea. 10% difference should be easy to see, anyone with both 390 and 390x?

Memory speed may have as much or more to do with it, as well as bandwidth, since the DAG is hashed purely in gpu memory. Don't know about AMD, but GTX responds much better to maxing memclock vs any bump in core clock.

[asrilani]

The bandwidth of the 390 cards is 512 bit. So maybe the memory speed is already fast enough. It is restricted by the core.

[antantti]

Can you speculate why the 390x is not faster than the 390 at the same frequency. It has more core, 2816 vs. 2560.

I have no idea. 10% difference should be easy to see, anyone with both 390 and 390x?

Memory speed may have as much or more to do with it, as well as bandwidth, since the DAG is hashed purely in gpu memory. Don't know about AMD, but GTX responds much better to maxing memclock vs any bump in core clock.

That's what I thought too after playing with nvidias. In fact I was running these amd's long time at 1600-1700 mem thinking it automatically gives me more hash.

Nvidia with 256-bit memory lane and 3,5-4GB is memory limited, amd with 384-bit and 3GB suffers from DAG-file size, amd with 512-bit and 8GB is begging for core overclock.

[bluebox]

That's what I thought too after playing with nvidias. In fact I was running these amd's long time at 1600-1700 mem thinking it automatically gives me more hash.

Nvidia with 256-bit memory lane and 3,5-4GB is memory limited, amd with 384-bit and 3GB suffers from DAG-file size, amd with 512-bit and 8GB is begging for core overclock.

maxwell cores are typically overclocked/boosted from the factory so there may not be as much room left anyway compared to AMD's, but power to get there is still the differentiator.

I don't see 390's with less than 8GB, or less than $300+ each... but at $12+/ETH, ROI is so short for anything these days.

[tolazy]

done a try to OC a bit now. from 1080 core til 1150 core volt +25 and power limit +50
did go from 28.4 to 30.3Mhs
temps are around 70-72c with fans on auto around 45%

that didnt go so well for me computer hung after 20min
will give a bit lower oc and see if that works

[antantti]

That's what I thought too after playing with nvidias. In fact I was running these amd's long time at 1600-1700 mem thinking it automatically gives me more hash.

Nvidia with 256-bit memory lane and 3,5-4GB is memory limited, amd with 384-bit and 3GB suffers from DAG-file size, amd with 512-bit and 8GB is begging for core overclock.

maxwell cores are typically overclocked/boosted from the factory so there may not be as much room left anyway compared to AMD's, but power to get there is still the differentiator.

I don't see 390's with less than 8GB, or less than $300+ each... but at $12+/ETH, ROI is so short for anything these days.

Cheap 290 with 512-bit bus and 4GB does closer to 30MH. It just doesn't show me that memory controller load...

[antantti]

done a try to OC a bit now. from 1080 core til 1150 core volt +25 and power limit +50
did go from 28.4 to 30.3Mhs
temps are around 70-72c with fans on auto around 45%

that didnt go so well for me computer hung after 20min
will give a bit lower oc and see if that works

Try 1100, -100mV to core & aux + tap user defined fan profile. Leave power limit as it is.

[tolazy]

done a try to OC a bit now. from 1080 core til 1150 core volt +25 and power limit +50
did go from 28.4 to 30.3Mhs
temps are around 70-72c with fans on auto around 45%

that didnt go so well for me computer hung after 20min
will give a bit lower oc and see if that works

Try 1100, -100mV to core & aux + tap user defined fan profile. Leave power limit as it is.

-100mV to core & aux set.
but i see some black lines on the screen when i have 1100 like the graphics tear. get lower temps 64c now at 1080
didnt see much diffrence when i had it at 1100