We had very high hopes for OpenCL on FPGA, as this would remove almost the entirety of the cost and effort involved. But the OpenCL bitstream compilers we've investigated are disappointingly far from usable for our purposes. 3/n
-
-
Replying to @jmgosney @jwcreation
Second, FPGA is essentially worthless for anything outside of very slow hashes with external candidate generation & comparison, or single hash brute force against fast hashes. For anything else, you need to add RAM. 4/n
1 reply 0 retweets 1 like -
Replying to @jmgosney @jwcreation
Adding a sufficient amount of RAM to an FPGA that is fast enough to facilitate anything other than single hash brute force is incredibly expensive. There are FPGA boards with DDR3 or DDR4, but these aren't fast enough. 5/n
1 reply 0 retweets 1 like -
Replying to @jmgosney @jwcreation
SRAM or FRAM would be ideal, but extremely expensive in the required quantities. GDDR5 or GDDR6 is a possibility, but no one makes such a board, and developing such a board is currently out of our financial reach. 6/n
1 reply 0 retweets 0 likes -
Replying to @jmgosney @jwcreation
And if we did have the means to create such a board, we'd probably create something far cooler than an FPGA board, such as our own custom integer coprocessor designed specifically for accelerating hash calculations.
1 reply 0 retweets 3 likes -
Replying to @jmgosney @jwcreation
However, such a coprocessor would ultimately look a hell of a lot like a GPU, minus FPU and the bits necessary to push pixels, and without the benefit of being subsidized by gamers and thus would cost far more than a GPU.
1 reply 0 retweets 3 likes -
Replying to @jmgosney @jwcreation
Therefore, there really exists no economic incentive to develop such a coprocessor, and for the foreseeable future, nothing beats GPUs in terms of price, performance, flexibility, ease of development, etc. GPU is the "sweet spot."
1 reply 0 retweets 2 likes -
Replying to @jmgosney @jwcreation
Eventually I think GPGPU will be displaced by manycore CPUs, but I think we're still a ways away from that happening.
1 reply 0 retweets 2 likes -
-
I mostly agree except on "FPGA is essentially worthless for anything outside of very slow hashes with external candidate generation & comparison, or single hash brute force against fast hashes." JtR descrypt-ztex is fast, has on-device mask mode & comparison for 2047 hashes/salt.
-
-
The on-device mask support also works in hybrid modes (the mask can be anywhere before/after/inside of candidates generated by other cracking modes on host). We also have thoughts on Bloom filters using most of the chip's BRAMs, which should allow for a few million hashes(/salt).
1 reply 0 retweets 0 likes -
not that it can’t be done, but it’s probably still just more cost effective to stick with the GPUs that already have decent amounts of ultra high speed RAM, is all
0 replies 0 retweets 0 likes
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.
