Could the EC2 infrastructure be used to efficiently launch a 51% attack?

Question

This excellent answer from Serith states that one would need 31150 ATI 5870 graphics cards to launch a 51% attack. Amazon EC2 offers computers for rent for 2.10$ per hour. These instances have 2 x NVIDIA Tesla “Fermi” M2050 GPUs. Given that these are as fast as the ATI 5870 (I have absolutely no idea whether this is the case), and that Amazon has enough of these instances available, one should be able to launch a 51% attack for one hour for a mere 65415$. If the hash rate of the Bitcoin network varies over the time of the day or under such events as a power cut, then it could be cheaper.

Is this a potential threat to the Bitcoin network?

Answer 1

An NVIDIA Tesla “Fermi” M2050 is not as fast as an ATI 5870. According to the wiki, the NVIDIA card makes around 70Mh/s and the 5870 makes around 400Mh/s.

If they were similar and if Amazon had that amount of cards available for rent, then it would be possible. Just don't forget that you would be paying $65415 per hour to sustain an attack that will not give you that much power over the network. See this question for an explanation of what the attacker could do.

Answer 2

EC2 would be prohibitively expensive to attack bitcoin but would be well suited against so called "GPU unfriendly" block chains. Against bitcoin EC most likely usage would be to "top off" an attackers hashing power. If a hypothetical attacker built a massive hashing farm but was slightly short of 51% hashing power the attacker could use EC2 instances to push the attack "over top".

EC2 is ill suited as primary source of an attacker's hashing power because EC2 GPU prices are expensive, the GPU Amazon chose is poorly suited for hashing, and the number of instances is lmited (Amazon only has so many GPU instances). CPU instances are significantly cheaper but they have on average 1/50th the hashing power of average GPU which significantly increases the number of instances necessary to achieve 51% hashing power.

For so called "GPU-unfriendly" block chains (litecoin, tenebrix, solidcoin, etc) use of EC2 instances is far more efficient and economical. Since these chains chose to exclude GPUs (a dubious decision) they have in effect removed the performance advantage between the hardware used by honest nodes and the most cost effective hardware available to attackers. Each honest node is now significantly less powerful (being limited to on average a single CPU) and CPU power (in a variety of forms) is much cheaper and easily accessed by an attacker. In essence these chains have commoditized the hardware used for hashing and as such left themselves vulnerable because large amounts of CPU hardware can be obtained for very cheap if only needed for a short period of time (like in a 51% attack).

Although the problem exists in all "GPU unfriendly" chains lets take a closer look at one chian, Litecoin. Litecoin's hashing power is roughly 30MH/s. To defeat that would require only 6000 Amazon Compute Units (virtual CPU roughly equivelent to a modern 1GHz Xeon CPU). The largest Amazon instances have 30 compute units. Thus it would only require 200 of the largest instances or roughly $300 per hour at current prices to achieve 51% of Litecoin's network hashing power.

Every other "GPU unfriendly" coin faces the same problem. By excluding high performance GPU they have made it easier for an attacker to leverage large sources of easily obtained low cost CPU power putting the network at risk