Look here or here or here or here.
Beware: The first two links are in German language.
ctrl-f for "SSE2" ;)
"Eine mögliche Lösung findet sich in den Dokumenten von AMD. Dort heißt es, dass bei bestimmten Konstellationen hinsichtlich des Codes, zusätzliche Wartezyklen entstehen."
"A possible solution could be found in the documentation from AMD. There it is stated that specific constellations regarding the (integer)code is leading to additional wait-states."
Seems to be common knowledge that - at least on the first generation - SSE2 was very slow when using Integer-SSE2
Look here or here or here or here.
Beware: The first two links are in German language.
ctrl-f for "SSE2" ;)
"Eine mögliche Lösung findet sich in den Dokumenten von AMD. Dort heißt es, dass bei bestimmten Konstellationen hinsichtlich des Codes, zusätzliche Wartezyklen entstehen."
"A possible solution could be found in the documentation from AMD. There it is stated that specific constellations regarding the (integer)code is leading to additional wait-states."
Seems to be common knowledge that - at least on the first generation - SSE2 was very slow when using Integer-SSE2
Thanks for the links!! Now this finally makes some sense, I guess, and the CPU detection code was after all well intentioned, but not so well written.
Again, for those who join the discussion later, what we discuss here is not a bug in the code written by science guys here, but something in the math library that comes with the compiler.
The Clawhammer has in fact been crunching a WU with the "patched" app all day. It looks as if it's at least stable, not a hint at problems so far. I can't say much about performance yet, but I asked my friend to remove the patch and do another WU with the vanilla app, so we can compare.
The Clawhammer has in fact been crunching a WU with the "patched" app all day. It looks as if it's at least stable, not a hint at problems so far. I can't say much about performance yet, but I asked my friend to remove the patch and do another WU with the vanilla app, so we can compare.
That will be interesting to see. At worst it will not improve, I guess.
Roadrunner, I don't really get your last comment ... .you are pregnant or what ?? :-)
Obviously you did not attended school in the GDR (but so did i for a mere four years ;) ).
There every pupil owned a booklet in which notes for the parents were written down.
When you achieved something outstanding - such as beeing ever punctual or knew the answer to difficult questions - you would have gotten a bee stamped into this booklet.
This booklet was called "Muttiheft". ;)
Obviously you did not attended school in the GDR (but so did i for a mere four years ;) ).
There every pupil owned a booklet in which notes for the parents where written down.
When you achieved something outstanding - such as beeing ever punctual or knew the answer to difficult questions - you would have gotten a bee stamped into this booklet.
This booklet was called "Muttiheft". ;)
LOL!!
Ahhhhh, I see.... . In some englisch speaking countries, I think "(New) Mothers Booklet" refers to a booklet where you document examinations of the pregnant mother and the baby later on :-) . And that didn't make much sense to me in the context :-).
And yes, you deserve a bee. There was a Einstein@Home get together/ (Tag der offenen Tür) last year in Germany, wasn't it? If it's repeated this year, we should all bring our booklets for stamping :-)
I'm hearing of this get-together for the first time... when and where would that be? I'd be really interested in meeting some fellow crunchers and staff members RL.
I know someone who has a
)
I know someone who has a Clawhammer in his notebook, any tests we should do?
RE: I know someone who has
)
Yes indeed, it would be instructive to see
a) The performance of the unpatched app (
b) if it runs ok under the patched version
c) whether it runs any faster with the patched version.
That would be cool
CU
BRM
Look here or here or here or
)
Look here or here or here or here.
Beware: The first two links are in German language.
ctrl-f for "SSE2" ;)
"Eine mögliche Lösung findet sich in den Dokumenten von AMD. Dort heißt es, dass bei bestimmten Konstellationen hinsichtlich des Codes, zusätzliche Wartezyklen entstehen."
"A possible solution could be found in the documentation from AMD. There it is stated that specific constellations regarding the (integer)code is leading to additional wait-states."
Seems to be common knowledge that - at least on the first generation - SSE2 was very slow when using Integer-SSE2
RE: Look here or here or
)
Thanks for the links!! Now this finally makes some sense, I guess, and the CPU detection code was after all well intentioned, but not so well written.
Again, for those who join the discussion later, what we discuss here is not a bug in the code written by science guys here, but something in the math library that comes with the compiler.
CU
BRM
There you are! ;) I only
)
There you are! ;)
I only needed a little tweaking with search-parameters on google. :D
Besides: Do i now get a bee in my mothers-booklet? :D
The Clawhammer has in fact
)
The Clawhammer has in fact been crunching a WU with the "patched" app all day. It looks as if it's at least stable, not a hint at problems so far. I can't say much about performance yet, but I asked my friend to remove the patch and do another WU with the vanilla app, so we can compare.
RE: The Clawhammer has in
)
That will be interesting to see. At worst it will not improve, I guess.
Roadrunner, I don't really get your last comment ... .you are pregnant or what ?? :-)
CU
BRM
Obviously you did not
)
Obviously you did not attended school in the GDR (but so did i for a mere four years ;) ).
There every pupil owned a booklet in which notes for the parents were written down.
When you achieved something outstanding - such as beeing ever punctual or knew the answer to difficult questions - you would have gotten a bee stamped into this booklet.
This booklet was called "Muttiheft". ;)
RE: Obviously you did not
)
LOL!!
Ahhhhh, I see.... . In some englisch speaking countries, I think "(New) Mothers Booklet" refers to a booklet where you document examinations of the pregnant mother and the baby later on :-) . And that didn't make much sense to me in the context :-).
And yes, you deserve a bee. There was a Einstein@Home get together/ (Tag der offenen Tür) last year in Germany, wasn't it? If it's repeated this year, we should all bring our booklets for stamping :-)
CU
BRM
I'm hearing of this
)
I'm hearing of this get-together for the first time... when and where would that be? I'd be really interested in meeting some fellow crunchers and staff members RL.