Claymore's CryptoNote AMD GPU Miner v11.3

[ol92]

Claymore are you using VMPROTECT on the EXE files?

If so that could/would explain a lot of the problems people are having.  It has some problems when dealing directly with hardware.

iam pretty sure it does..... use vmprotect

If so it would explain a lot of the memory errors and loosing touch with the hardware.

Yes I use vmprotect. Do you know a lot about it to state that it's the reason of problems with GPUs? If so, let me know more details, I will be glad to check it.
I always test both original and vmprotect versions on my systems and I never saw any differences for several months in different miners and other software I created.
For CryptoNote miner, for example, both versions don't work until I set pagefile 16GB.
If you have problems with vmprotect - miner will not start on your system at all because it does its job at starting only. If you see at least one line in console after starting - everything works properly.
The problem is in AMD drivers, they don't want to work with large memory sizes. Drivers just don't allow OpenCL app to allocate a lot of GPU memory at once. For example, 290X card has 4GB but OpenCL gives me less than 1GB at once by default. GPU_MAX_ALLOC_PERCENT variable is a solution (I can assign up to 3GB), but buggy solution as you can see. And AMD found a solution - today I tried Catalyst 14.6 and even with GPU_MAX_ALLOC_PERCENT I cannot allocate more than 1GB at once now. So they solved the problem I have a couple of ideas about possible workaround, I hope it will help, new version will be available soon. Anyway, working with AMD OpenCL is some kind of magic which is much more unstable than vmprotect.

Maybe this can help :
http://devgurus.amd.com/thread/159516 :
"
Hi vanja_z, welcome to the forum,

 

  Currently OpenCL users are limited to 25% of device memory,

 

I don't know where you get this from, perhaps it's a rumor, but it's certainly not correct.

(there is a 512MB limit per allocation call but you can allocate as much as you like)

 

I do predominately scientific computing and often need very large and fast memory so I am mostly using the 7970. On the 7970, I often allocate a single contiguous buffer that uses just shy of 3GB, the device limit. It's very simple, all you do is allocate in chunks of 512MB or less and make sure the chunks are rounded to about 0x4000 bytes, then they will be placed contiguously. Example, allocating 2GB you might have kernel buffers like

 

__kernel(global float *A, global float *B, global float *C, global float *D){}

 

Since this is C language and A,B,C,D are memory pointers, you can use A to reference all of memory.

Here is a printout from a typical program start:

 

open:devices 3 gpus, 1 cpu, device(0) = Tahiti

start(cl):ndevs=3 gpus=1 time=57.136

<readback of actual allocation map>

buffer 0 start 01D1E000 to 21D1E000 size=20000000  Gap = 00000

buffer 1 start 21D1E000 to 41D1E000 size=20000000  Gap = 00000

buffer 2 start 41D1E000 to 61D1E000 size=20000000  Gap = 00000

buffer 3 start 61D1E000 to 81D1E000 size=20000000  Gap = 00000

buffer 4 start 81D1E000 to A1D1E000 size=20000000  Gap = 00000

buffer 5 start A1D1E000 to B0E1E000 size=0F100000  Gap = 00000

buffer 6 start B0E1E000 to BF21E000 size=0E400000  Gap = 00000

buffer 7 start BF21E000 to BFE1E000 size=00C00000  Gap = ----  (last address on GPU is BFFFFFFC)

 

The last couple of buffers are different size for an unrelated reason. Note, I have not used GPU_MAX_ALLOC

type parameters and have never seen a need to. This also works on Cayman, and Barts devices but I prefer

Tahiti because the memory is so large and fast. Sorry, I don't know much about Nvida devices because I

usually choose hardware based on specifications.

 

Hope it helps."

[1714926826]

1714926826

[Claymore]

The last couple of buffers are different size for an unrelated reason. Note, I have not used GPU_MAX_ALLOC
type parameters and have never seen a need to. This also works on Cayman, and Barts devices but I prefer
Tahiti because the memory is so large and fast. Sorry, I don't know much about Nvida devices because I
usually choose hardware based on specifications.

Hope it helps."[/i]

I already use this trick, starting from first version And it works, but not on all systems as you can see, especially on multi-GPU systems. So I will try to improve this trick a bit.

[pinoyminer1977]

Yes I use vmprotect. Do you know a lot about it to state that it's the reason of problems with GPUs? If so, let me know more details, I will be glad to check it.
I always test both original and vmprotect versions on my systems and I never saw any differences for several months in different miners and other software I created.
For CryptoNote miner, for example, both versions don't work until I set pagefile 16GB.
If you have problems with vmprotect - miner will not start on your system at all because it does its job at starting only. If you see at least one line in console after starting - everything works properly.
The problem is in AMD drivers, they don't want to work with large memory sizes. Drivers just don't allow OpenCL app to allocate a lot of GPU memory at once. For example, 290X card has 4GB but OpenCL gives me less than 1GB at once by default. GPU_MAX_ALLOC_PERCENT variable is a solution (I can assign up to 3GB), but buggy solution as you can see. And AMD found a solution - today I tried Catalyst 14.6 and even with GPU_MAX_ALLOC_PERCENT I cannot allocate more than 1GB at once now. So they solved the problem I have a couple of ideas about possible workaround, I hope it will help, new version will be available soon. Anyway, working with AMD OpenCL is some kind of magic which is much more unstable than vmprotect.

It would be great to have your new version soon, and hope that we will be seeing some improvements again in the hashrate...say 50%

[pepi]

Starting both Claymor miners is tricky if you have Kaspersky antivirus.  In both cases if Kaspersky is active aplication starts and computer hangs or even print BSOD.
Work around is to disable Kaspersky, start app then reenable protection. But in version 2.1 of GPU ( and sometimes CPU miner) after re-enabling protection process crash and stops. And sometimes only restart help!
So if you use some kind of software protection: change it... it looks like it is more damage then good thing from that.

[HardwarePal]

Is anyone getting this error message at the start of the miner before it hashes :

Warning : Linking two modules of different data layouts ... "amdil64-pc-amdopenc1" ?

Edit Found it on a previous post https://bitcointalk.org/index.php?topic=638915.msg7227539#msg7227539

[Xccel]

hello dev plz add claymore to minergate 

[mig6r]

hello dev plz add claymore to minergate 

+1 It would be really great !!

[koyking]

i can mine for only 10-20 minutes, keep saying gpu memory allocation error.

plz help.

Tried, set Virtual memory to 16 GB+.

miner  claymore v 2.2
5 * r9 280x
windows 8.1
13.12 amd catalyst driver

plz fix
Posted from Bitcointa.lk - #JfPiKev7HoIzzspR

[pepi]

And if I may to say: there is nothing so special in Catalyst 13.12. I install latest drivers and didnot loose even one hash/sec... So if someone has problems with 13.12 then tray later version...

[ivanlabrie]

And if I may to say: there is nothing so special in Catalyst 13.12. I install latest drivers and didnot loose even one hash/sec... So if someone has problems with 13.12 then tray later version...

Not so accurate...with my 290 I see a big big difference, because of the memory allocation bug of opencl.

[zetent]

Had the miner on last 2 days with 3x 290 no problems at all.

I have 8gb ram and I've not specified any GPU allocation nor page file size.

[nrg_wolf]

Had the miner on last 2 days with 3x 290 no problems at all.

I have 8gb ram and I've not specified any GPU allocation nor page file size.

what driver version and what brand of card and brand of memory on the gpu's

[restless]

The 2.2 version works on 6970 with 2GB ram, returning 120h/s
But it seems GPU_MAX_ALLOC_PERCENT  doesn't work  - the program always allocates the max 2048MB of memory. It has same behaviour on 290 - no matter the number i enter for alloc_percent it always allocates 2560MB

[wanvo]

dev plz add solo mining.Thanx.

[drr0ss]

i can mine for only 10-20 minutes, keep saying gpu memory allocation error.

plz help.

Tried, set Virtual memory to 16 GB+.

miner  claymore v 2.2
5 * r9 280x
windows 8.1
13.12 amd catalyst driver

plz fix
Posted from Bitcointa.lk - #JfPiKev7HoIzzspR

You don't need exactly a pagefile 16 g +.
Opencl allows use to only 2 g of the video card.
Windows occupies the pagefile as a mirror of RAM.
I mean if you are running only Windows occupied RAM is example from 500g to 1g , miner for 2 cards gets 1 g for Windows + 2x2g for each of cards, or a total of 5 g.
Leave a spare 2 g.
In other words, 8 g pagefile are sufficient for a machine with 2 video cards and 4g RAM.
This amount will increase proportional if you increase the number of cards as forecast a 2 g for each card.
Try and see
Too big amount of pagefile may have negative effect, decrease performance of system.

[nrg_wolf]

driver 14.6 beta seems to yield more stable results so far, useage does not seem to bounce around anymore, how ever it yields lower hash rates.... about 50H/S lower then 13.12 drivers on dual card setup so 25H/S loss per card at stock clocks. gpu max allocate seems to have no affect on newest version of drivers weather its set at 80 or 100 and even set to run via cmd prompt, seems to not register on the miner and it enters bad performance mode.

[kylam]

Hello,
After the update will not support 14.6 BETE drive?

[nrg_wolf]

Hello,
After the update will not support 14.6 BETE drive?

what are you talking about? iam useing 14.6 beta drivers on my gb 280x rig, given the elpida memory so far it seems more stable then 13.12..... its just alittle slower H/S wise but the miner still works.

[zetent]

Had the miner on last 2 days with 3x 290 no problems at all.

I have 8gb ram and I've not specified any GPU allocation nor page file size.

what driver version and what brand of card and brand of memory on the gpu's

13.12, Gigabyte reference with elpida on all of them

[AceCobra1]

So I left my 2 rigs running overnight last night but I when I woke up, I got this error and the miner stopped:

SOMETHING WRONG WITH THE DEVFEE CONNECTION. MINING STOPPED.

Any ideas how this error can be resolved? I have already checked and my WiFi connection is perfectly fine on both rigs.