forums.silverfrost.com

[LitusSaxonicum]

John,

If you look at the post on "Some issues with roundoff and a whole lot else", I looked at the memory capacities for self-build computer components, and most mainboards support 32Gb, but you can, for a price, get 8-socket boards with correspondingly expensive memory. My prices are in GBP.

Just about everything I want to do can be done in 1Gb, but I recognize the need of folks like Dan and John in Oz to access all the RAM they can. It seems to me that a 1D array or REAL*8 could take up 16 Gb with 2Gb elements of 8 bytes, let alone 2 or 3D arrays.

The fact that bigger than 4-byte indexing is needed in terms of bytes doesn't seem to me to need bigger than 4-byte integers for the variables. But hey, it's all academic in the sense that I'll never need to use it.

Eddie

[DanRRight]

John, Ask your own or any 10 years old kid in the world do they want GTX 1080ti with 11GB of RAM or not. They will ask you if you can buy two. Or three. And will perfectly explain and demonstrate why even more RAM is needed.


Same but often much worse with our modeling. The modeling world is even not 3D but more multidimensional. Remember that Fortran 77 had 7-dimensional restriction ? I suffered from that because for one task i needed 8.

With plasma PIC codes for example you need for ions 3 space dimensions, 3 momentum dimensions, 1 more for ionization state and 1 for partitioning. And that for 2-3 different elements. Total 9. Total 7 dimensions needed for electrons (they all have ionization =1). Also data needed for the EM fields: 3 for electric field, 3 for magnetic field. Do the math how many particles you can include in the simulation for realistic modeling.

[JohnCampbell]

John,

[LitusSaxonicum]

Depends what you want it for. Most Fortran programs ever written will run in 1 Gb, mainly because they were intended to run on machines with a lot less.

Eddie