Originally Posted by malventano
...Power users wouldn't normally have enough 'real work' for the SSD to cause it to actually heat up enough to throttle. I had to write over 100GB sequentially to a 950 Pro 512GB to get it to throttle with 0 airflow across it. No power user is going to have 100GB worth of data to be written at >1GB/s constantly to even put on this SSD in the first place, and even if they did, it was still writing at >1GB/s even when throttled.
Heh. Read up on finite element modeling - especially in the electromagnetic realm with products like Ansys HFSS. I regularly do need to write that size of data not just in generating final output files but in interim operations for temporary scratch files as well. Maximizing RAM is obviously the biggest knob in solution speed relative to # of unknowns (so is # of processing paths - core count), but after that disk I/O is the next significant bottleneck. (And beyond that, networking, if you're trying to run a distributed solution on 8 workstations which each have 24 cores and 256GB RAM...Ethernet bites your solution speed in the tail hard. Need infiniband.)
Granted I'm building for home and gaming - not for work - so your point is valid for me; I just feel a bit more comfortable with the larger form factor. Same reason I'm overprovisioning on my CPU cooler relative to my actual desire to overclock.
I do intend to see if I can benchmark the speed difference between having design files and temp scratch space defined on a PCIEx4 vs. SAS drive...hope to use the data to convince work to make some more optimal purchase decisions for the workstation configurations we use there.
The downside is writing 2-300 GB of temporary files, several times, followed by a similar amount of final field and mesh data per simulation may eat up the wear leveling allocation in a hurry. For work I'm actually recommending these not as the OS/program drive for that reason, but as replaceable computation space. Of course by the time I convince our IT department of the benefits Intel's new thingie will be out (XPOINT) which, assuming it has speed approaching that of RAM, is nonvolatile, and doesn't have a wear date on it like NAND, will be a godsend.Edited by rtrski - 10/23/15 at 10:46am