end of ETH mining on 2G cards

[nerdralph]

Even though the DAG is still < 1.8GB, the lite DAG for on-GPU DAG generation takes up some extra space.  Genoil and Claymore are both failing to allocate enough memory on a 2GB card I tested.  I've thought about trying to tweak the kernel to free up a bit more memory, but I don't think it is worth it for a few more epochs.  For the really desperate, you can probably still mine a bit longer with the official ethminer which uses DAG files.

edit: Swapped in a 4G card for the 2G, and although it has no problems allocating the DAG, it's hashing at half speed.  Maybe I have an issue with the rig...

edit2: rebooted, adjusted OC settings and now have a couple 4G R9 380 cards running smooth at 21Mh/s on 840/1500 clocks.

Even if the issue is not the DAG size, we're not far from it.  While the actual DAG size won't reach 2G until the new year, with the extra memory for kernel code, lite DAG, etc., I'd say 2016 is the last year for 2GB cards mining eth.

[1715329945]

1715329945

[1715329945]

1715329945

[1715329945]

1715329945

[adaseb]

Try

setx GPU_FORCE_64BIT_PTR 0
setx GPU_MAX_HEAP_SIZE 100
setx GPU_USE_SYNC_OBJECTS 1
setx GPU_MAX_ALLOC_PERCENT 100
setx GPU_SINGLE_ALLOC_PERCENT 100



Or put those 2GB gpus in a rig with at least a 3GB or higher GPU.


Currently works for me.

[nerdralph]

Try

setx GPU_FORCE_64BIT_PTR 0
setx GPU_MAX_HEAP_SIZE 100
setx GPU_USE_SYNC_OBJECTS 1
setx GPU_MAX_ALLOC_PERCENT 100
setx GPU_SINGLE_ALLOC_PERCENT 100



Or put those 2GB gpus in a rig with at least a 3GB or higher GPU.

Currently works for me.

Looks like you just cut/paste from somewhere, since the only GPU_SINGLE_ALLOC_PERCENT needs to be set for 2GB cards.  Using another card with more memory is an interesting comment.  I have 2 other 4GB cards in the same rig, and it doesn't make a difference.  I could see how it might though, if the driver's crossfire support allows allocating RAM from another GPU, with DMA over the PCI-e bus.  That would only be OK for the lite DAG due to the much lower bandwidth of the PCI-e bus.

[maxmad_x]

Currently solution for all those who are having 2G cards will be to move to Zcash which has almost exactly same profitability as ETH. Sell ZEC and buy ETH for the same price
I am sorry this is not a technical advice from any point of view but a crappy solution to unfortunate miner buddies.

[nerdralph]

Currently solution for all those who are having 2G cards will be to move to Zcash which has almost exactly same profitability as ETH. Sell ZEC and buy ETH for the same price
I am sorry this is not a technical advice from any point of view but a crappy solution to unfortunate miner buddies.

For R7 370 cards (2 or 4G), ZEC mining is better.  They get only 13-14Mh mining eth, but 80+ sols mining ZEC.  The R9 380 does better mining eth, and I don't expect that to change.  The cards that are slow on eth mining like the 370 and the R9 280x will be used for ZEC.  Cards like the Rx 480 will be used mostly for ETH, except for the occasional price pumps on ZEC that briefly make it more profitable to mine.

[Camnates]

Currently solution for all those who are having 2G cards will be to move to Zcash which has almost exactly same profitability as ETH. Sell ZEC and buy ETH for the same price
I am sorry this is not a technical advice from any point of view but a crappy solution to unfortunate miner buddies.

Those cards will be more profitable to mine the ZCash anyway. I have all my 7970, 7950 running the ZCash.

[greaterninja]

I ran into this issue a 1-2 months ago.

The primary display video card needs to be greater than 2gb.  I used 1 x 380 4gb
The others can be 2gb.  I used 5x 370 2gb.

You also need to set the environment variables as posted above.


I restarted that rig today to see if it would run into issues, its still hashing.  The lifetime though...who knows.  Zec is probably a better alternative for 370 as nerdalpha says.


They are slightly different commands in linux and windows.

[eckmar]

Umm i'm still mining with r9 380 2Gb each and they are doing just fine. There is no 4Gb card in the system but as i said no problem here. Looked at used memory, its 1870 on primary card

[greaterninja]

Umm i'm still mining with r9 380 2Gb each and they are doing just fine. There is no 4Gb card in the system but as i said no problem here. Looked at used memory, its 1870 on primary card

I think it depends on the OS, driver, and how its configured.
On my Ubuntu 14.04 LTS rig it allocates 200-300 MB for the primary video adapter unless I set it to headless mode.

[adaseb]

Damn,

I restarted one of my rigs which were mining ETH for like the last 250 hours and the 2GB cards wouldn't work.

Tried another rig, same thing.

Tried a 3rd rig, and for some reason the Radeon 6990 still works. Its the only 2GB card working on ETH.

So at the time of this post, it worked but since I never restarted that rig it kept hashing.


 

[paijoe]

few days ago I was frustrated
always failed mining with my vga 2GB
but I found a way, add a command --opencl-platform 1
I follow this way

Hello,

I'm getting this strange issue:

No GPU device with sufficient memory was found. Can't GPU mine. Remove the -G argument

C:\Users\venom\Desktop\ethminer-0.9.41-genoil-1.1.7>ethminer.exe -G --list-devices
Genoil's ethminer 0.9.41-genoil-1.1.7
=====================================================================
Forked from github.com/ethereum/cpp-ethereum
CUDA kernel ported from Tim Hughes' OpenCL kernel
With contributions from nicehash, nerdralph, RoBiK and sp_

Please consider a donation to:
ETH: 0xeb9310b185455f863f526dab3d245809f6854b4d


Listing OpenCL devices.
FORMAT: [deviceID] deviceName

• Ellesmere

        CL_DEVICE_TYPE: GPU
        CL_DEVICE_GLOBAL_MEM_SIZE: 8589934592
        CL_DEVICE_MAX_MEM_ALLOC_SIZE: 4244635648
        CL_DEVICE_MAX_WORK_GROUP_SIZE: 256
[1] Ellesmere
        CL_DEVICE_TYPE: GPU
        CL_DEVICE_GLOBAL_MEM_SIZE: 8589934592
        CL_DEVICE_MAX_MEM_ALLOC_SIZE: 4244635648
        CL_DEVICE_MAX_WORK_GROUP_SIZE: 256
[2] Ellesmere
        CL_DEVICE_TYPE: GPU
        CL_DEVICE_GLOBAL_MEM_SIZE: 8589934592
        CL_DEVICE_MAX_MEM_ALLOC_SIZE: 4244635648
        CL_DEVICE_MAX_WORK_GROUP_SIZE: 256


C:\Users\venom\Desktop\ethminer-0.9.41-genoil-1.1.7>Cuda-ethpool.bat

C:\Users\venom\Desktop\ethminer-0.9.41-genoil-1.1.7>setx GPU_FORCE_64BIT_PTR 0

ERFOLGREICH: Angegebener Wert wurde gespeichert.

C:\Users\venom\Desktop\ethminer-0.9.41-genoil-1.1.7>setx GPU_MAX_HEAP_SIZE 100

ERFOLGREICH: Angegebener Wert wurde gespeichert.

C:\Users\venom\Desktop\ethminer-0.9.41-genoil-1.1.7>setx GPU_USE_SYNC_OBJECTS 1

ERFOLGREICH: Angegebener Wert wurde gespeichert.

C:\Users\venom\Desktop\ethminer-0.9.41-genoil-1.1.7>setx GPU_MAX_ALLOC_PERCENT 100

ERFOLGREICH: Angegebener Wert wurde gespeichert.

C:\Users\venom\Desktop\ethminer-0.9.41-genoil-1.1.7>setx GPU_SINGLE_ALLOC_PERCENT 100

ERFOLGREICH: Angegebener Wert wurde gespeichert.

C:\Users\venom\Desktop\ethminer-0.9.41-genoil-1.1.7>ethminer.exe --farm-recheck 75 -G -S eu2.ethermine.org:4444 -FS eu1.ethermine.org:14444 -O xxx.wrk
Genoil's ethminer 0.9.41-genoil-1.1.7
=====================================================================
Forked from github.com/ethereum/cpp-ethereum
CUDA kernel ported from Tim Hughes' OpenCL kernel
With contributions from nicehash, nerdralph, RoBiK and sp_

Please consider a donation to:
ETH: 0xeb9310b185455f863f526dab3d245809f6854b4d

No GPU device with sufficient memory was found. Can't GPU mine. Remove the -G argument

C:\Users\venom\Desktop\ethminer-0.9.41-genoil-1.1.7>

You have a built-in GPU anf the miner tries to use it first.

Add "--opencl-platform 1" to your command line

[QuintLeo]

You also need to set the environment variables as posted above.

 I have NOT run into this issue with the environment variables set.

 Yet.

 I suspect it will become a real issue in a few months or so - and a lot of lower-end cards will move into ZEC or XMR or something where it's not an issue.


 It IS kinda funny watching one of my HD 7750 cards with 1GB mining ZEC just fine - 44/sec isn't impressively scary but not bad for a card that paid for itself via LiteCoin mining years ago.


 [edit] it's even funnier watching one of my A10-7860K mining ZEC - at a whopping 11/sec with only 512 MB allocated to it. Not really WORTH enough to leave it that way but I decided to experiment in the hopes it would be closer to the HD 7750 on performance (same # cores, newer GCN version, but much slower RAM).
 Funny part - it's probably closer to making up the electric usage on it's GPU section than the 7750 is.

[toptek]

 set the environment variables

I don't set the environment variables with my R7 370's mining eth if i mine them, mostly i mine Zec but I had set them with my 7870 when i had it, if i wanted to mine ETH with that card and both cards have 2 gb, i guess Ive been lucky with the 370 .

[nerdralph]

I suspect it will become a real issue in a few months or so - and a lot of lower-end cards will move into ZEC or XMR or something where it's not an issue.

It's weeks, not months.  It is impossible to allocate a 2GB DAG file on a 2GB card.  Kernel code, OpenCL driver code to launch & control kernels, etc takes up memory.  The light DAG (for on-GPU DAG generation) takes ~30MB as of the current epoch.
https://github.com/Genoil/cpp-ethereum/blob/master/libethash-cl/ethash_cl_miner.cpp#L442
Genoil's miner could be modified to release the light dag buffer after the DAG is generated, but that would only help for a few more epochs.  I don't think it is worth the trouble, and I'm pretty sure JW isn't going to bother either.
I'll use my 2GB R7 370's for ZEC.  I only have one 2GB R9 380, which isn't much better than a 370 for ZEC, so I'll just sell it.

[Xiestar]

I suspect it will become a real issue in a few months or so - and a lot of lower-end cards will move into ZEC or XMR or something where it's not an issue.

It's weeks, not months.  It is impossible to allocate a 2GB DAG file on a 2GB card.  Kernel code, OpenCL driver code to launch & control kernels, etc takes up memory.  The light DAG (for on-GPU DAG generation) takes ~30MB as of the current epoch.
https://github.com/Genoil/cpp-ethereum/blob/master/libethash-cl/ethash_cl_miner.cpp#L442
Genoil's miner could be modified to release the light dag buffer after the DAG is generated, but that would only help for a few more epochs.  I don't think it is worth the trouble, and I'm pretty sure JW isn't going to bother either.
I'll use my 2GB R7 370's for ZEC.  I only have one 2GB R9 380, which isn't much better than a 370 for ZEC, so I'll just sell it.

For these cards, they are more efficient to mine the ZEC any way. So let the RX 470 and 480 mine the ETH.

[adaseb]

Another 2GB stopped working today when we got Epoch #90.

It works until you re-start your program.

[QuintLeo]

I noticed a couple nights back I didn't have any of my 2GB cards in rigs I had gotten back online - and quickly fixed that.

 Both of my single-card HD 7870 rigs DID work fine prior to the most recent epoch - but they were enough more profitable on ZEC that I slid them over to that and haven't looked back yet.

 Given the 2GB issue appears to be hitting NOW (I'd neglected to consider the other stuff that has to go into the card's RAM) I might just leave them there.


 I've also got my dual R9 280x rig on ZEC, as it's profitability ratio on ZEC/ETH is pretty much identical to the HD 7870 rigs - but that rig CAN go back as those are 3GB cards.



 Been a busy week since I pulled the "in-wall electric heaters" and wired 220 outlets to their wiring to run some of my A2s on - I've actually got some POWER available again!

[adaseb]

Well I got them working again by running -eres 0.

It means it doesn't load the next 2 DAGs onto the memory, so I got 10 more days left of mining.

With the official miner probably another few DAGs along the way but I guess its bye bye 2GB.

[Redrose]

It was a sure thing to happen, though it's sad. What I hope for you guys is that you'll find those cards another use and for all people I met here and to whom I recommended to don't buy 2 Gb cards that they listened to me !

[adaseb]

I bought the 2gb cards because back in February who knew it would remain profitable this long. And the 4gb were mostly overpriced.