| 1 |
>Sadly, your acid_demeanor precludes you from such opportunities. So if |
| 2 |
you change your mind and want to take a professional approach to |
| 3 |
palidromes, you >might just establish some new friendships. Think things |
| 4 |
over a bit and let me know should you want to access more aggressive |
| 5 |
resources. YES, you can run >gentoo on those HPC resources. Off the |
| 6 |
record, have you tried a systolic algorithm and using rDMA via the DDR5 |
| 7 |
on a collection of GPUs to speed up >your search? James |
| 8 |
|
| 9 |
GPU is appealing, the problem is ammble to a high degree of |
| 10 |
partitioning. There is some benefit to working blocks sequentially due |
| 11 |
to arithmetic redundancy. |
| 12 |
|
| 13 |
Now the main issue is that keith's approach is to use an immense lookup |
| 14 |
table to use a radix approach. |
| 15 |
|
| 16 |
Basically if your start is "abc" then here are 10 or so things you can |
| 17 |
add to get "cba" on the tail end, this saves many millions of useless |
| 18 |
tests. The bigger the lookup table the faster the algorithm... |
| 19 |
(currently around 25gb) |
| 20 |
|
| 21 |
I am using some AMD specific features, I haven't tried to port it to intel. |
| 22 |
|
| 23 |
-- |
| 24 |
IQ is a measure of how stupid you feel. |
| 25 |
|
| 26 |
Powers are not rights. |
Archives