AMI (amazon machine instance) miner?

[marcus_of_augustus]

Anyone had a go at this?

The underlying hardware available is GPU clusters (4gpu expandable to 8 per cluster):

22 GB of memory
33.5 EC2 Compute Units (2 x Intel Xeon X5570, quad-core “Nehalem” architecture)
2 x NVIDIA Tesla “Fermi” M2050 GPUs
1690 GB of instance storage
64-bit platform
I/O Performance: Very High (10 Gigabit Ethernet)

$2.10 per hour of each running "instance".

Seems like a bit of overhead in building the AMI to begin with ... and the whole thing OS, code will be running as a java virtual machine  
So don't know how much access to low-level system functions, etc. And how good are the Nvidia Tesla GPUs at hashing anyway?

Though if you did get a AMI built you could then sell it to others who want to just push the button and go.

[1713572173]

1713572173

[1713572173]

1713572173

[1713572173]

1713572173

[lachesis]

Just tested this out today. I got about 77MHPS / Tesla core using an OpenCL miner, which is far too low to justify running the box. I didn't test with Puddingpop's CUDA miner, though.

[marcus_of_augustus]

lachesis: how serendipitous.

So that's 4x77Mhash/s , i.e. 308 Mhash/s for the 4gpu AMI cluster?

[burtyb]

and the whole thing OS, code will be running as a java virtual machine  

AWS Compute and GPU instances use HVM (Hardware Virtual Machine) - the cli tools are Java (but you don't have to use those).

BB

[murasha]

oh I am trying to lauch both JAVA and Python miners for a whole day and cant bring them to work.
lachesis, can you make please a complete guide?

[lachesis]

lachesis, can you make please a complete guide?

Sorry, I don't really have time for that.

My hint would be to use a better AMI than the default, like the starcluster gpu one (ami-12b6477b), found here:
http://mailman.mit.edu/pipermail/starcluster/2010-December/000572.html.

However, I don't think it's worth it. First of all, there are only 2 GPUs as far as I can tell. Didn't see an option for a 4x GPU cluster. Furthermore, you'd need something like 1600 MHPS to justify the $2.10 cost / hour.

[jafo]

However, I don't think it's worth it. First of all, there are only 2 GPUs as far as I can tell. Didn't see an option for a 4x GPU cluster. Furthermore, you'd need something like 1600 MHPS to justify the $2.10 cost / hour.

I also came to the same conclusion, getting around 70Mhash/sec.  The route I took was rather crazy, because I had a tool-chain for Ubuntu Maverick, but the Maverick AMI didn't support the GPU machines, so I went with the CentOS AMI but I couldn't get the tools to build under CentOS.  So I built a Maverick chroot, which I copied over to the GPU machine running CentOS and ran the miner from that.

I also found there were 2 GPUs, it wasn't able to run with 4, although the specs for the cards used there seemed to imply that there were 2 GPUs per card, I didn't find that that was the case, at least from what was visible to me.