Topic: delete - page 2. (Read 6143 times)

Who cares.  Bunch of retards in that camp anyway.   LTC is doing well and it sounds like they're feeling threatened to me.  So they should be


Yup.

Please provide a cite.

Information about the new instruction set actually leaked ages ago, and it looks promising on paper.

Not really... I was just implying that deepceleron was being rational and that his reasoning wasn't what BCX intended when he posted this thread with the title he used and incomplete logs that could be wrongly interpreted by people who don't pay much attention to details.
I guess I could've told you this in my previous post... sorry.

Far as I can tell you're implying that only BEX would rent CPU power to attack a CPU coin when there is actually an exchange, and it would be far smarter to mine once the exchanges are all closed?

Then you should put your brain to work.

Not sure what your point is Psy.

How dare you to be rational when it doesn't fit douchebagexpress agenda?

The next gen is going to be different.

AMD is keeping the 78xx series card the "same".  It is simply a die shrink.  Same GPU, more shaders, smaller, cooler, faster, cheaper.  We can make pretty good gueses as to performance per watt and performance per $.

The 79xx however is a completely new architecture (code named "graphics core next", with new instruction set, and shader design.  There are also signficant hardware changes both inside and outside the GPU.  The cards will move to higher cost but lower latency XDR2 ram for example.

How well it stacks up on both MH/w and MH/$ remains to be seen.  

The point they seem to be discussing is, with the availability of cloud computing rental services, if you want to generate a few blocks in a row on any of these ponzicoins for nefarious purposes, all you have to do is rent the horsepower to do it. However, you can just wait for the creators of any of these faulternate blockchains to let them fizzle after they pump-and-dump, and then you can be the only one mining anyway (that is, if it's not completely broken by poorly informed implementation.)

From the currently available information, it's looking like 79xx should be at least as efficient at it as previous generations of AMD cards; all the parts still seem to be there. No idea about NVidia though. FPGA and structured ASIC is more dependent on what kind of pricing you can get than anything else.

If it is just an issue of implementation then one could simply make a better CUDA miner.  The reality is that Nvidia has critical "flaws" (or maybe design decisions is better way to say it) that limit throughput on their existing hardware.

Bitcoin doesn't require any specific implementation.  Any method that produces valid SHA-256 hash is fine.

sisoft throughput is horrible which would indicate a non-optimized solution (likely masking inferiority of Nvidia GPU).

For example: HD 6970 3649 MB/s
3649 MB/s *8 = 29192 Mbps
29192 / 512 = 51.6 MH/s (512 bits in SHA-256 hash)

Bitcoin MH is actually SHA-256(SHA-256(block header))
51.6MH/s /2 = 25.8 Bitcoin MH/s

There has been a lot of work in optimizing bitcoin hashing performance.  Things like vectors, workgroup sizes, BFI, etc.  Sisoft may have simply decided on an implementation and stuck with it to provide comparable performance but their results don't indicate the peak of what AMD cards can do.

True, I didn't mean completely different hashing algorithms but different SHA-2 implementations. Sisoft Sandra benchmark also uses SHA256 and the performance is completely different:
http://www.sisoftware.net/?d=qa&f=cpu_vs_gpu_crypto

Even old SHA-1 can be extensively optimized:
http://software.intel.com/en-us/articles/improving-the-performance-of-the-secure-hash-algorithm-1/

True but as you state it wasn't intentional.  There is no guarantee SHA-256 will be efficient on the new 79xx series card which use a new architecutre.  Nvidia has indicated they intend to improve GPGPU integer performance so future cards may close the performance gap.  There is no way to know how well algorithms will work on future FPGA or Structured ASIC designs.

IMHO one shouldn't try to optimize the algorithm.  I doubt Satoshi did.  SHA-256 is an industry standard hashing algorithm with public and documented crypto-analysis.  It is currently secure and has been extensively tested.  That is more important than trying to find a "fair" algorithm.  NVidia GPU tend to perform poorly on most block ciphers not just SHA-256.  I imagine market forces will demand NVidia improve integer performance in future generations.

We miss you in the solidcoin channel BitcoinEXpress.

Well, I agree but it isn't that simple. Bitcoin's mining algorithm cripples any hardware other than Radeons HD5000 & 6000. Of course not intentionally but still. There are efficent hashing algorithms out there that runs better on Nvidia hardware or runs similarly od HD4000 and HD5000 hardware (with the same number of SPUs). Bitcoin's doesn't.

Yeah I was loose and fast with the numbers.  It would make sense to let each algorithm float on its own (independent difficulty) as a result they both would achieve similar economic benefits (i.e. value of blocks over cost of hardware & electrical costs).  Still I don't think that system would be wise because while it would be better than a "GPU-hostile" chain it is still inferior to an "open chain" in terms of making the network as difficult.  It is really just a half step to be more "fair" (a dubious reason IMHO).  

Why are GPU used over CPU?  Simple they are more efficient.  If a person is only concerned about the security of the network (not their personal gain) then you want the most efficient hardware possible protecting the network.  Sadly I think in 5 or so years the rise of APUs mean that botnets will represent a greater danger to Bitcoin because it closes the multiplier (difference in hashing power between average attacker node and defender node).  All of AMD APU have modest hashing performance (~60MH/s) and AMD default drivers enable OpenCL now so as time increases we should expect greater performance from zombie computers.

So what is the solution? Not reggressive nonsense like "GPU hostile" alt-chains.  The solution is EVEN GREATER efficiency.  It is an arms race.  Granted that likely will result in a second wave of "not fair" and probably a spwan of "GPU friendly alt chains" because now FPGA and other exotic hardware will be considered the "not fair enemy".  Hopefully by the time APU make botnets more dangerous a significant fraction of Bitcoin hashing power will be from devices with even greater efficiency keeping that vital multiplier between defender hashing power and attacker hashing power high.

Trying to criple efficient hardware is a deathblow to security of any blockchain.  You are simply bringing defenders down to the same level as the botnets.  The bad news is botnets win in that crippled race by pure numerical superiority.



Yeah it is pretty sweet.  I am glad they built it.  EFF said that DES was "broken" but most people dismissed it as "theoretical attack".  So they collected some donations (include some from me) and built one.  It only took 18 months and about $250K.  It could crack any DES password in 4-5 days (max time 10 days).  That lead to acceleration in adopting stronger hashes (albeit with the interim step of the fugly Triple-DES).

Well, in this case okay, but what you describe here isn't a "comparable benefit" between GPU and CPU (top of the range GPU=somewhere around 800MH, not 100).

Nice DES cracking board, btw

Well no because most people would use massive GPU farms.  For example if Bitcoin had an "alternate" hash algorithm that got say 100MH from a top of the line CPU.  There would be a lot more CPU mining BUT GPU mining would still have an advantage.  You can fairly easily add 8 GPU to a rig.  8 socket servers aren't economical. Now I don't think CPU mining is needed AT ALL.  I think it is a fad.  A case of "not fair he has all those GPU toys, well I am going to come over here and play w/ my CPU toys - no GPU allowed".

Still a hyrbrid chain would at least gain the hashing power necessary to present at least a challenge to CPU based attackers.  Granted it wouldn't be as strong as an "open chain" which doesn't attempt to exclude CPU/GPU/FPGA/ASICS.  The funny thing is restricting hardware is simply reactionary.   Mining is getting to be a lower profit enterprise.  Soon economics will force the move to FPGA.  Eventually it will move from FPGA to Structured ASICS (like Altera Hardcopy) and finally to fully custom ASICS (if crypto currency lives long enough).  Even the move to custom chips isn't the end of the road.  Those designs will increase in complexity and density.  They will go from single chip module to multichip boards to finally massive parallel clusters. 

Like this:
http://upload.wikimedia.org/wikipedia/commons/thumb/b/bd/Board300.jpg/572px-Board300.jpg

This is a board from EFF "Deep Crack" a custom built DES cracking array (to prove DES is insecure).  Each board is doublesided and has 64 custom built ASICS optimized to do nothing but crack DES.  5 boards of 64 ASICS ea were installed into each rack chassis.  The entire system was 6 chassis connected together.  A single off the shelf PC controlled the array of 1856 ASICS.

You can't stop technology.  It will always be pushing forward.  This is a good thing.  Custom built hashing arrays could achieve higher throughput, higher density, and lower cost per hash than less customized systems.  This means the attacker has a reverse multiplier.  The attacker is using less efficiency equipment which means it takes more attackers to equal the power of one "good guy".

Doesn't work: CPU remains efficient and cheap, only we can expect to have maybe twice the hashing power because people will mine both with CPU and GPU. That is, supposing they're willing to drop BTC mining in favor of this new chain where GPU doesn't have an advantage over CPU.