Question Looking for guidance on multi core FEA/CFD solves: new AMD vs. Team blue.

[billmce]

I did a search and not much anything came up that is anywhere near current on engineering applications and CPU's and frankly not all that hopeful given the emphasis on gaming which apparently is where the money is at. Does anybody know a site more appropriate for this kind of application? I have read everything that Tom's has written in say the last 12 months on processors and IO. They all quote some benchmark for multi threaded performance which doesn't mean much to me like cinebench or what not. Any help on what benchamrk would be indicative of what i am seeking would be helpful as long as it is a benchmark that is actually run in reviews. So, just to set some context: My current box which is more than a bit dated and needs a refresh and since I tend to keep them for a while I want to make a good call as the seconds add up for me when solves can take hours to run. Current system is a 3.3GHz X5680 dual socket (2x6 core) with 192 GB of registered ECC RAM (at one point it was 48GB or non registered ECC RAM)and a Quadro P4k ( which I modernized from the original Quadro 6k). I have a few terabytes of SSD SATA 2 storage, some PCIe SSD storage (1TB) and some HDD archive drives. OS is Win 10 - hyper threading=off (yeah I know some version of Linux would be faster but we all know the reasons I am not likely to go there). It started its life as a DELL T7500 circa 2011 or near enough. It is still a pretty solid box in terms or reliability. Practically an antique in this space. I want to cut my solves time by some multiple without spending another fortune. I run ABAQUS (in simple terms an FEA code) which can use either SMP pr MPI for a parallel computation, I also run SolidWorks Flow Sim (aka Mentor Graphics FloEFD) and on occasion I run SWX Sim (don't really care about how this works though). So, to the question: This new upcoming AMD Ryzen series 3 - looks like a deal for 12 or 16 cores which is what I need (the software tokens to support more CPU's cost as much as what I expect to pay for my new system which i hope will double or more my through put). Does anybody know:

1. how much RAM can they support? I figure 256GB would be good but I could live with less but 128Gb is not enough. How come Tom's reviews or this stuff never talk about what the addressable RAM is but it is always mentioned with respect to the i9 or i7's what have you? Gamers maybe are not so interested in a lot of RAM I suppose.
2. how do they stack up against Intel in real world multi threaded matrix solves (sparse, symmetric, unsymmetric and dense). Rumor has it that Puget systems ran tests and the AMD's just didn't stack up but that was likely for stuff that was a few years old at least but it was just hear say from a software vendor but I am sure it was well intentioned. These applications tend write big blocks of data to disc/storage on a regular basis as well which slows things down while it happens so recommendations on fast storage tech also welcomed.

Not all that interested in GPU calcs at this point as they are not supported for all problem types.
Any help or insights would be welcome.
Thanks, in advance, for your time,
Bill

 

[InvalidError]

Maximum memory is determined by the platform, not by the CPU. DDR4 goes up to 32GB per slot, mainstream platforms have two channels with up to two DIMMs each for a total of 128GB which is far more than any normal person will need within the next 10+ years. ThreadRipper doubles the channel count, which doubles the max RAM capacity to 256GB. Same goes with Intel HEDT except Intel isn't actively pushing HEDT anymore, just like it isn't pushing much in terms of meaningfully new desktop parts. If you want more RAM than that, you will have to step up to EPYC which has eight channel memory controller with optional buffered DIMM support which can take max memory beyond 1TB per socket.

For FEA/CFD, ThreadRipper and EPYC will match the nearest thing Intel has to offer at half the price if not less.