Error CUDA mining: an illegal memory access was encountered

[rizwansarwar]

Compiled from master, after a few minutes I get this. Mining on CUDA using GTX 1070's. Not sure what is this, the error is not very descriptive and I am not code wizz.

CUDA error in func 'search' at line 365 : unspecified launch failure. ✘ 15:26:10|cudaminer0 Error CUDA mining: unspecified launch failure CUDA error in func 'search' at line 365 : unspecified launch failure. ✘ 15:26:10|cudaminer4 Error CUDA mining: unspecified launch failure CUDA error in func 'search' at line 365 : unspecified launch failure. ✘ 15:26:10|cudaminer1 Error CUDA mining: unspecified launch failure CUDA error in func 'search' at line 365 : unspecified launch failure. ✘ 15:26:10|cudaminer3 Error CUDA mining: unspecified launch failure CUDA error in func 'search' at line 365 : unspecified launch failure. ✘ 15:26:10|cudaminer2 Error CUDA mining: unspecified launch failure

[Angel996]

xstead, who's using ethminer these days? Are you from 2015, dude? )))

[xstead]

@Angel996 someone who knows what they are doing...

[evilny0]

Don't feed the troll.

[xstead]

@evilny0 ;)

[Angel996]

evilny0, what do you mean, exactly? ethminer is the slowest ethash miner nowadays, what's the point of using it?

[AndreaLanfranchi]

@Angel996 if you're so convinced about this simply do not use it and do not denigrate the hard work it's ongoing to improve it.

[Angel996]

AndreaLanfranchi, are you convinced otherwise? Are you saying ethminer is faster and/or more power efficient than Claymore's?

[AndreaLanfranchi]

Simple answer yes.

[Angel996]

Ok, so I believed you, I spent some time, dl'ded latest ethminer release.

Tried it on an Ubuntu 16.04 LTS rig with 5x Palit 1060 Stormx 3GB Samsung. Core -200, mem +1200.

And here is the result:

Claymore's: ~23.5 Mhs per card, ~116 Mhs per rig. Consumption ~ 90 Wt/card. Ethminer: hardly 19.5 Mhs per card, ~96 Mhs per rig. Consumption ~ 90 Wt/card.

VERDICT: It's a good thing to be "convinced" about something, but reality is a bit different.

[MariusVanDerWijden]

@Angel996 We're an open source project. We're not getting payed to do any of this. I think it is really cool, that so many developers contributed to this project and did this in their spare time. Without developers like this, there wouldn't be any cryptocurrencies, because no one would (and should) trust closed source code! In my own test, i find about ~6 percent difference to claymore, which is a price I am willing to pay for knowing what code is executed on my machine

[tonyaik]

Not to feed the troll but here it goes: @Angel996 Whats the real hashrate at the pool? Claymores shows aprox 10% more than actual hashrate at the pool. Ethminer shows the exact same at the pool as in the miner. My 6x 580 8GB rig showed ~180 in claymores with ~168 at the pool. With a correct setup ethminer i get ~175MH/s at both the pool and in the miner. I love ethminer, keep up the great work!

[Angel996]

MariusVanDerWijden -- benefits of open source software? Sure. But when it comes to mining, it all about making money. That kind of renders slower miners totally useless. And please don't tell me you mine for educational purposes.

tonyaik, don't look at hashrate reported, did you count the actual number of shares submitted? I got two identical rigs, I can run a test, say, for an hour and see. What exactly is "correct setup ethminer"? I looked thru --help output, I don't see a lot of options there. OC? I use same settings as with Claymore's.

[tonyaik]

@Angel996 Yes, I looked at the actual number of shares. Better hashrate at the pool and no dev-fee.

--cl-local-work and global-work is what I mean. For me, the defaults weren't optimal.

[Angel996]

cl options are for AMD cards which are not in question (alghough I have those too). As for dev fee, here:

https://github.com/JuicyPasta/Claymore-No-Fee-Proxy

Since stratum protocol is a merely plain text TCP session, works like a charm if SSL is not used by miner.

[tonyaik]

Saw the exact same differance with my 7x 1060 rig.

Using a no-fee-proxy is a shitty thing to do. Don't want to play by the developers rules, don't use it imho. Anyhow. I'm out of this thread.

[AndreaLanfranchi]

VERDICT: It's a good thing to be "convinced" about something, but reality is a bit different.

Reality is subjective. I had the same tests over and over again.

To compare with you I have a 6x EVGA 1060 3Gb on Micron (not Samsung) and I can squeeze from each one roughly 18.67 Mh/s. using -200/+850 at 72.50 Watts using Claymore (pushing harder makes the whole system unstable and/or unresponsive), while on ethminer I get 18.52 Mh/s using -200/+750 at 72.50 Watts (system quite stable running for batches of 12 hours each). And yes ... I am limiting power as much as i can as I do my maths.

The drop in reported hashrate is nothing in respect of the 1% fee. In addition must say claymore's uses, on average, 26% more of CPU which causes my rigs with not very powerful celerons quite laggy. With ethminer my machines work smoothly.

Plus ... my measurements using ethermine.org depicts that the reported hashrate is quite overlapping the effective and average hashrate. Those overlapping lines where never seen with claymore's. (disregard the last 3 hours where I had connectivity problems.) I always had suspicion that claymore's reports higher hashrate than effective ... but as we cannot read the code ...

Last but not least: ethminer is free to use without your "cheats" (which may well be worked around by claymore's in near future). And is open-source: the value of being able to read the code and being sure that nothing unwanted or unexpected happens behind the curtains is much appreciated.

VERDICT : to express absolute verdicts is always not a good idea.

And I stop here.

[Angel996]

tonyak>> Claymores shows aprox 10% more than actual hashrate at the pool.

That's absolutely NOT true! The reported hashrate is sent to the pool for people to actually be able to make a comparison of real hashrate vs reported. Here, I've run a 4x 1060 1x 1050ti rig for 20 hours in a row. Results:

Average Hashrate for last 6 hours: 105.3 Mh/s Last Reported Hashrate: 106.2 Mh/s

Pool is eth.nanopool.org. The slight difference is actually 1% devfee of Claymore's. This is also a good way to check if pool is stealing shares. Probably your experience is due to the fact of share theft, not Claymore misreporting hashrate.

[Angel996]

AndreaLanfranchi>> To compare with you I have a 6x EVGA 1060 4Gb on Micron (not Samsung) and I can squeeze from each one roughly 18.67 Mh/s. using -200/+850 at 72.50 Watts using Claymore (pushing harder makes the whole system unstable and/or unresponsive)

I can express verdicts because I have built many rigs. I do it for myself and also for money for other people.

Micron memory is pretty fast too, it should give you about 21 mh/s (at least!). If your system gets unstable/unresponsive with further OC, it's not GPU problem, it's your power supply or wiring. Because OC over limit should just throw "GPU is off the bus" error, the rig should not hang or get unresponsive. By lowering clocks and playing with powerlimit you just get system stability as a tradeoff instead of providing steady power to your equipment.

If you use SATA -> 6PIN power converters, ditch them. Solder quality thick wires directly to your PSU and you'll be surpirsed how much better your rig would perform. I've been thru that.

Alternatively, you might want to try removing 5x GPUs and running your system on 1 GPU. See if you can OC it better. My bet is, you can (as the GPU gets more power). Also, that you mention system becoming unresponsive, suggests your CPU/motherboard/memory is underpowered because of too many GPUs in your rig, or, again, bad wiring.

[Angel996]

UPDATE:

I tried it on a rig with 2x 1060 Hynix and 2x GTX 970 Hynix.

Claymore's: GPU0 18.473 Mh/s, GPU1 18.465 Mh/s, GPU2 10.292 Mh/s, GPU3 10.536 Mh/s

Ethminer: gpu/0 18.60 gpu/1 18.60 gpu/2 10.49 gpu/3 10.49

It's a tad faster even. )) That's very interesting. Also, might be a clue for ethminer developers. How does hashrate relate to memory speed?

p.s. GTX 970 used to be much faster (up to 21 mhs), but they got drastically slower after certain ethash epoch :((

[miningpronto]

4x1080ti evga Windows 16gb ram

This code appeared twice today. Both times I was logging in via vnc. It appears itnis what triggered it. 70% power 149 overclock 499 memory overclock

[rosefun]

I meet same problems when I try to use GPU in keras.

InternalError: CUDA runtime implicit initialization on GPU:0 failed. Status: an illegal memory access was encountered

[shaunstoltz]

For me this was a straight up overheating problem. Once I sorted out the card that was throwing this error with better air flow, the error went away.