• Guest, The rules for the P & N subforum have been updated to prohibit "ad hominem" or personal attacks against other posters. See the full details in the post "Politics and News Rules & Guidelines."

Ethereum GPU mining?

Page 40 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

DrMrLordX

Lifer
Apr 27, 2000
18,473
7,342
136
For those of you who have used the Claymore miner: is there any way to make the thing completely ignore/not use an entire GPU? I can't get it to stop mining on my iGPU.
 

Feld

Senior member
Aug 6, 2015
287
95
101
For those of you who have used the Claymore miner: is there any way to make the thing completely ignore/not use an entire GPU? I can't get it to stop mining on my iGPU.
From the claymore documentation:

-di GPU indexes, default is all available GPUs. For example, if you have four GPUs "-di 02" will enable only first and third GPUs (#0 and #2).
 

sm625

Diamond Member
May 6, 2011
8,172
136
106
Jeez they should put a proper post on the back of the supply, so you can run a real cable:



I'd like to see that melt.
 

MrTeal

Diamond Member
Dec 7, 2003
3,131
800
136
I don't think it's quite as simple as bus width. I have a single machine running the following cards under 15.12 and Claymore
7990 (GCN 1.0) - 1000MHz/2048 SP (4.10TFLOPS), 384-bit/1500MHz (288 GB/s), ~19.5MH/s (x2)
7950 (GCN 1.0) - 1000MHz/1792 SP (3.58TFLOPS), 384-bit/1250MHz (240 GB/s), ~17.5MH/s
380 (GCN 1.2) - 1100MHz/ 1792 SP (3.94TFLOPS), 256-bit/1375MHz (176 GB/s), ~19.5MH/s
290 (GCN 1.1) - 1100MHz/ 2560SP (5.63TFLOPS), 512-bit/1300MHz (333GB/s), ~31MH/s

The Tahiti based cards have a massive bandwidth and bus width advantage over Tonga, but there really isn't a big difference in hashrate per flop within the margin of error for those two. The only real outlier is the 290, which is about 15% higher. The 7990 gets 4.75MH/s per TFLOP, the 7950 gets 4.88, the 380 gets 4.95 and the 290 gets 5.5.

My laptop has the especially gimped Oland GPU with 2GB of GDDR5, I wonder if I could get it mining and see what it does with a 128-bit bus.
Be prepared to be blown away.
8790M (GCN 1.0) - 900MHz/ 384SP (.69TFLOPS), 128-bit/1000MHz (64GB/s), ~2.9MH/s
4.2MH/s per TFLOP.
 

DrMrLordX

Lifer
Apr 27, 2000
18,473
7,342
136
Having run Claymore for 24h, I have a few items to report.

1). Bandwidth usage by Claymore isn't that high. Apparently it's ability to automatically use proxy mode (I'm using eth-proxy mode, to nanopool) keeps bandwidth usage pretty low. A 7700k + 290 + 390 had 6.88 MB of data usage total over a 24-hour period. That's 79.7 Bps (or 637 bps for those who prefer their bits).

2). Performance seems . . . maybe a wash? Using standard ethminer, my 7700k should do around 2.0-2.5 MH/s, and my 290 + 390 should do around 25-2.51 MH/s @ 900 MHz (currently heavily underclocked/undervolted). After mining for 24h, the three GPUs together managed 51.8 MH/s over the last 6 hours according to nanopool. Mind you, that's counting only the shares sent to nanopool . . . 1% of all shares were mined elsewhere thanks to the Claymore fee, so the actual MH/s should have been around 52.1 . That was with intensity 8 on the Hawaii GPUs and intensity 0 on the Spectre, and Spectre using algorithm setting '1' for older cards. No decred was mined during this process.

After 24+ hours of mining with ethminer + eth-proxy, nanopool showed only ~22.5 MH/s with the 290 and 390, with reported speeds of around 23.6 MH/s. That was with a farm recheck setting of 3000 which seems to impose a penalty for jobs that should see completion in less than 3 seconds, though I may be off-base there. I might rerun ethminer with farm recheck 1000 through eth-proxy to see if I get any improvements. It does look like mining through eth-proxy uses about the same amount of bandwidth (I showed 6.7 MB over a 24-hour period of mining with 290 + 390).

Though I can conclude that mining through Claymore is at least a wash. It has enough nice software features that I'm considering replacing ethminer and Genoil ethminer with it on all my rigs. I'll have to be careful though since at least my 390 required more volts to be stable using Claymore. It required a jump of maybe +30 mV?
 

Accord99

Platinum Member
Jul 2, 2001
2,241
137
106
The main benefit from Claymore over Ethminer I'm seeing is that it's much more stabler, I no longer have half my systems crashing every time the DAG changes size. It's also nice to easily and quickly see the impact of core/mem clock changes.
 

Madpacket

Platinum Member
Nov 15, 2005
2,068
326
126
Having run Claymore for 24h, I have a few items to report.

1). Bandwidth usage by Claymore isn't that high. Apparently it's ability to automatically use proxy mode (I'm using eth-proxy mode, to nanopool) keeps bandwidth usage pretty low. A 7700k + 290 + 390 had 6.88 MB of data usage total over a 24-hour period. That's 79.7 Bps (or 637 bps for those who prefer their bits).

2). Performance seems . . . maybe a wash? Using standard ethminer, my 7700k should do around 2.0-2.5 MH/s, and my 290 + 390 should do around 25-2.51 MH/s @ 900 MHz (currently heavily underclocked/undervolted). After mining for 24h, the three GPUs together managed 51.8 MH/s over the last 6 hours according to nanopool. Mind you, that's counting only the shares sent to nanopool . . . 1% of all shares were mined elsewhere thanks to the Claymore fee, so the actual MH/s should have been around 52.1 . That was with intensity 8 on the Hawaii GPUs and intensity 0 on the Spectre, and Spectre using algorithm setting '1' for older cards. No decred was mined during this process.

After 24+ hours of mining with ethminer + eth-proxy, nanopool showed only ~22.5 MH/s with the 290 and 390, with reported speeds of around 23.6 MH/s. That was with a farm recheck setting of 3000 which seems to impose a penalty for jobs that should see completion in less than 3 seconds, though I may be off-base there. I might rerun ethminer with farm recheck 1000 through eth-proxy to see if I get any improvements. It does look like mining through eth-proxy uses about the same amount of bandwidth (I showed 6.7 MB over a 24-hour period of mining with 290 + 390).

Though I can conclude that mining through Claymore is at least a wash. It has enough nice software features that I'm considering replacing ethminer and Genoil ethminer with it on all my rigs. I'll have to be careful though since at least my 390 required more volts to be stable using Claymore. It required a jump of maybe +30 mV?
I slowly switched from ethminer + eth-proxy on ethermine (moved away from Nanopool a while ago due to fees) over to Claymore a few weeks ago. My observations are that Claymore requires either more voltage or less aggressive clocking (or both). However once you have the settings dialed in the stability of Claymore seems to make it a worthwhile switch. I also believe even with slightly lower clocks (albeit using a bit more wattage) I'm being paid out a little more on average than with ethminer + eth-proxy.

So overall I seem to have less downtime and more stable hashing results using Claymore and to me that makes it worthwhile. I was skeptical at first but after some consideration I came to the conclusion that uptime is the most important metric when you're talking small percentage differences (both in power and performance) so I'm sticking with Claymore for now. I don't dual mine as I find the extra stress on the cards adds unnecessarily risk.

I also really like that Claymore loads the DAG in the memory on the cards themselves rather than periodically downloading a large DAG file. The startup time is also much faster than ethminer. This all saves time and therefore money.

Claymore also has cool things like remote management capabilities but I'm very careful about turning these features on for security reasons. So far I haven't used them but may have soon for management purposes.

I haven't tried Genoil yet but it seems promising as well. If you do any comparisons using that please report back.
 
Last edited:

Madpacket

Platinum Member
Nov 15, 2005
2,068
326
126
It seems as soon as you overload the 8-pin connector, I think there is risk involved if the cable itself has a fault tolerance. Even if it's at spec, it seems long-term there is still some risk involved. I am not an electrician or a PSU engineer so please feel free to correct me. In my case because the R9 295X2 is using 500W of power, each of the 8-pin PCIe cables is outside of spec. The EVGA 1200W P2 has integral capacitor at end of PCIe cables.
I read somewhere that the 295X2 does something silly like instead of pulling load evenly across the the 8 pin connectors, it pulls as much as it can from one port and the remaining current from the other one when needed. This ends up with a serious imbalance like 75% load on one cable and 25% on the remaining one. Could this maybe be one of the reasons why the cable burned up?
 

Madpacket

Platinum Member
Nov 15, 2005
2,068
326
126
Be prepared to be blown away.
8790M (GCN 1.0) - 900MHz/ 384SP (.69TFLOPS), 128-bit/1000MHz (64GB/s), ~2.9MH/s
4.2MH/s per TFLOP.
I have a 7850 mATX and 7870K ITX system. For giggles I should test Claymore on them - 512SP @ 840Mhz / 2133 and 512SP @ 900Mhz / 2400Mhz DDR3 would probably mine at even less speeds (due to memory bandwidth).

Thanks for listing your hashing speeds for your cards. I'm a little surprised there's not much of a difference between 384 bit and 256 bit bus speeds. Does memory overclocking help your cards at all? On 290/390's setting the GDDR5 to 1225Mhz speed only saturates around 90-95% of the memory bus even at 1125Mhz GPU core. When testing out my 380 I noticed very little change in output when changing memory speeds (more with core) but that was with ethermine. I no longer have the card in my possession to test it out with Claymore.
 
Last edited:

MrTeal

Diamond Member
Dec 7, 2003
3,131
800
136
I have a 7850 mATX and 7870K ITX system. For giggles I should test Claymore on them - 512SP @ 840Mhz / 2133 and 512SP @ 900Mhz / 2400Mhz DDR3 would probably mine at even less speeds (due to memory bandwidth).

Thanks for listing your hashing speeds for your cards. I'm a little surprised there's not much of a difference between 384 bit and 256 bit bus speeds. Does memory overclocking help your cards at all? On 290/390's setting the GDDR5 to 1225Mhz speed only saturates around 90-95% of the memory bus even at 1125Mhz GPU core. When testing out my 380 I noticed very little change in output when changing memory speeds (more with core) but that was with ethermine. I no longer have the card in my possession to test it out with Claymore.
I was actually trying to see last night on my 380's and the results were negative on both core and memory. Unfortunately I can't seem to underclock them and they won't go below 980MHz core or 1375MHz memory. There's not difference in hashrate between 980/1375 and 1150/1500 though, which is a little surprising. I might need a custom bios to get them lower, not sure there.
 

VeryCharBroiled

Senior member
Oct 6, 2008
387
25
101
So overall I seem to have less downtime and more stable hashing results using Claymore and to me that makes it worthwhile. I was skeptical at first but after some consideration I came to the conclusion that uptime is the most important metric when you're talking small percentage differences (both in power and performance) so I'm sticking with Claymore for now.
agreed, up time is all. tweaking to within an inch of life for that 1% power saved, extra hashes or whatever gets wiped out fast when it crashes just once for 7 hours while youre at work or sleeping.

I switched to claymore a month or so ago. its stability vs the stock ethminer, ability to use config files, remote management, failover pools, built in stratum, and web server for stats are worth the 1% fee (I only mine eth).
 

n0x1ous

Platinum Member
Sep 9, 2010
2,524
181
106
Where is a good place to download and get instructions on claymore and get started on the switch from ethminer + ethproxy?
 

Madpacket

Platinum Member
Nov 15, 2005
2,068
326
126
OK good news all. I recieved my Sapphire 290's from NewEgg today and mange to test one out.

First, they come in sealed packages like a new video card would without the fancy AMD girl or crappy robots that cater to 12 year olds.

The cards look new. I think RS suspicion was correct and Sapphire is simply unloading what stock they have left.

Absolutely no scratches, dust, dirt etc. They look and smell brand new. At 60% fan speed (tolerable) it runs very cool at full mining speeds. No old dried out thermal paste here it looks like.

Now for the really good news (so far).

30Mh no problem. The first card does 1070Mhz undervolted -100mv core just like my better 390's do with Claymore. Also good ASIC rating and Hynix memory.

Hopefully the rest of my cards perform just as well. I haven't tested yet if these can be unlocked to full 290X and probably won't as I want to get my rig built ASAP.

Here's a quick gallery I put together, it was getting dark so not the best quality but let me know if you have any questions.

Good stuff!!!


https://imgur.com/a/32hD0
 

Madpacket

Platinum Member
Nov 15, 2005
2,068
326
126
Just go https://minergate.com/calculator/ethereum/

With ethereum, R9 290X gets 20~25MH/s , Now if you do alone , $23 per week.still low
Yeah that's not right. As Mr.Teal stated it's pretty easy to hit at least 30Mh or more with a 290/390. Add another Mh or so for a 290X/390X.

So that's closer to $30 USD a week at today's price. Still pretty easy to ROI on the 290's I just bought even after power costs. With projected difficulty roughly in 3 months all five cards should be paid for accounting for minimal downtime.
 
Last edited:

parkerface

Member
Aug 15, 2015
49
32
91
Yay for new toys!

God, those blower shrouds still make me cringe! haha Every time I see one I can't help but think of all those youtube vids of WHORRRRRRRRRR sounds.
 

Madpacket

Platinum Member
Nov 15, 2005
2,068
326
126
Yay for new toys!

God, those blower shrouds still make me cringe! haha Every time I see one I can't help but think of all those youtube vids of WHORRRRRRRRRR sounds.
Haha, just wait until all five cards are packed into a milk crate. This rig will be going in the furthest corner of my basement. I actually love the reference cards with the blower fans - rock solid VRM's, more power efficinet than most aftermarket cards and in my experience the blower fans run forever at high RPM's as long as you keep them clean, which is easy enough with a quick blast of air. Also no way you can pack that many non blower style cards into a milk crate. Great for mining ;)
 

parkerface

Member
Aug 15, 2015
49
32
91
Haha, just wait until all five cards are packed into a milk crate. This rig will be going in the furthest corner of my basement.
Oh, to have a basement! :wub: I'm in a 500 sq ft efficiency without carpet which is pretty much an echo chamber. Low noise levels are a must.
 

DrMrLordX

Lifer
Apr 27, 2000
18,473
7,342
136
I slowly switched from ethminer + eth-proxy on ethermine (moved away from Nanopool a while ago due to fees) over to Claymore a few weeks ago. My observations are that Claymore requires either more voltage or less aggressive clocking (or both). However once you have the settings dialed in the stability of Claymore seems to make it a worthwhile switch. I also believe even with slightly lower clocks (albeit using a bit more wattage) I'm being paid out a little more on average than with ethminer + eth-proxy.

So overall I seem to have less downtime and more stable hashing results using Claymore and to me that makes it worthwhile. I was skeptical at first but after some consideration I came to the conclusion that uptime is the most important metric when you're talking small percentage differences (both in power and performance) so I'm sticking with Claymore for now. I don't dual mine as I find the extra stress on the cards adds unnecessarily risk.

I also really like that Claymore loads the DAG in the memory on the cards themselves rather than periodically downloading a large DAG file. The startup time is also much faster than ethminer. This all saves time and therefore money.

Claymore also has cool things like remote management capabilities but I'm very careful about turning these features on for security reasons. So far I haven't used them but may have soon for management purposes.

I haven't tried Genoil yet but it seems promising as well. If you do any comparisons using that please report back.
Yeah the features of Claymore's miner put it ahead of the standard and Genoil forks in my opinion. I can try the Genoil miner this weekend on my Win10 machine and see how it stacks up vs. Claymore. I'm still finishing a 24-hour run of ethminer + eth-proxy (farm recheck 1000) get some more comparison data. farm recheck 3000 did not seem to be working out so well, at least not on this machine.

I was actually trying to see last night on my 380's and the results were negative on both core and memory. Unfortunately I can't seem to underclock them and they won't go below 980MHz core or 1375MHz memory. There's not difference in hashrate between 980/1375 and 1150/1500 though, which is a little surprising. I might need a custom bios to get them lower, not sure there.
I have problems with being unable to underclock a lot of the Hawaii cards I mess with in Win10. aticonfig under Linux seems to work no matter what the card. Regardless I flash their default clockspeed and mem down to 900/1250 so I can set them anywhere I want no matter what the OS.

OK good news all. I recieved my Sapphire 290's from NewEgg today and mange to test one out.

First, they come in sealed packages like a new video card would without the fancy AMD girl or crappy robots that cater to 12 year olds.

The cards look new. I think RS suspicion was correct and Sapphire is simply unloading what stock they have left.

Absolutely no scratches, dust, dirt etc. They look and smell brand new. At 60% fan speed (tolerable) it runs very cool at full mining speeds. No old dried out thermal paste here it looks like.

Now for the really good news (so far).

30Mh no problem. The first card does 1070Mhz undervolted -100mv core just like my better 390's do with Claymore. Also good ASIC rating and Hynix memory.
That's about what I noticed from the one I got. I returned mine anyway since I did not want to deal with the blower, at least not on that particular card. But it was good for 1070 MHz -100 mV during my testing. Power usage was acceptable.

Can you imagine a room full of these things mining away? The agony.
 

ASK THE COMMUNITY