F@H PPD Database

[DemonfangArun]

as we all know the original ocn cpu ppd database has been unmaintained for quite a while now, and is getting filled up with empty lines.

ocn one: https://spreadsheets.google.com/spreads ... zQmc&gid=0

i propose making a new one, with more details added, and listing by project instead of points for project (which is, imo, a bad way to do it).

replacement one is now available for public editing: https://docs.google.com/spreadsheet/ccc ... ZMXc#gid=0

[GreyWhiskers]

Couple of quick comments.

1. On new INTEL CPU tab, the header CPU is repeated, and there is no entry for Mem Amount, as there is in the AMD tab
2. The GPU tab could benefit from a column for Driver Version.
3. We're likely to start seeing some INTEL GPUs to show up with Haswell and other uses of integrated graphics. Should there be a tab for that?

Thanks.

[DemonfangArun]

Couple of quick comments.

1. On new INTEL CPU tab, the header CPU is repeated, and there is no entry for Mem Amount, as there is in the AMD tab
2. The GPU tab could benefit from a column for Driver Version.
3. We're likely to start seeing some INTEL GPUs to show up with Haswell and other uses of integrated graphics. Should there be a tab for that?

Thanks.

i'll add one for intel
woohps, forgot about driver version

and dammit, i thought i got rid of all the duplicate columns -.- <- edit: figured out what happened there

[JTG]

looking for data on a 4p 6348's PPD

[Nathan_P]

tpf would be roughly 13:15 on 8101, so about 350k PPD, note I have extrapolated this from some 6274 data, I cannot find any tpfs for 63xx cpu's yet

[JTG]

TY Nathan_P ...
however ...
i dont believe an extrapolation applies
to quote this site ....
http://www.theregister.co.uk/2012/11/05 ... rver_chip/

"the Piledriver cores include four new instructions.
FMA3 is a floating point fused multiply add instruction that is used for vector and matrix math and polynomial calculations commonly used in physical, chemical, and quantum simulations as well as in digital signal processing"

and this site ....
http://www.cpubenchmark.net/multi_cpu.html

saying 2x 6348 out preform 4x6278s

AND ...
with a full load Turbo frequency of 3.1 GHz (6348s)

this leades me to believe ... 6300s are more efficient at folding than 6200s ...
and with the 6348s being low/mid end of the food chain of the 6300 series ... one might be able to find some good deals
on some opterons that got a good "Humm" to them

keep me posted ... if you find any hard data

thanks again

[7im]

The four new instructions do not apply to folding performance. None of the current fahcores use FMA. Theoretical benchmarks need to use the same methods as fah for those benchmarks to be useful. Those do not.

FAH will support FMA in some form or another at some future time, but not right now.

[JTG]

Hay 7im

i believe that is a significant point ... not that FAH doesnt support FMA
but rather that ... it will ... "in some form"
i makes me think that 6300s are "future proofing" my folding rigs

very important ....
that along with being able to "productively" put a GPU in 6200 & 6300 4p boxen

hummmm ....

[Nathan_P]

JTG, without FMA 63xx are reportedly only ~2% faster than 62xx cpu's. I'll update once I find some numbers but for now its a pretty reasonable set of info. Turbo was taken into account.

[JTG]

Nathan_P
do you have some links that explain this in greater detail ..?
as more understanding facilitates better choices ...
TY

[Nathan_P]

Nathan_P
do you have some links that explain this in greater detail ..?
as more understanding facilitates better choices ...
TY

JTG, performance improvements in 63xx you can get off the web, linky below is to a bigadv spreadsheet with lots of tpf/point data on it.

https://docs.google.com/spreadsheet/ccc ... DcHc#gid=0