Ok, I get it. So another question: what exactly do the flops number of these cards mean?
FLoating point Operations Per Second. Basically math operations per second. 2 Tera FLOPS (TFLOPs0 is 2 trillion math operations per second. It is quite significantly high. Take 4 trillion numbers and add them. Done, one second flat.
One (me) would think based on the published numbers (2 vs. 2.2) that the performance difference between the D300 and D500 would not be very great.
There is a gap in VRAM size. If the problems are 'too big" for the D300 than even the marginal increase of the D500 is significant. Likewise if need ECC support
if switch to double precision numbers the gap is quite large between the D300 and D500 .
----------
They don't use ECC memory, just regular GDDR5. There's no ECC tables in these vbioses.
None of the AMD powered cards use ECC memory. The ECC support is layered on top of regularly VRAM by the GPUs memory controller. They all use regular GDDR5. They would need more VRAM to store the additional checksum data.
If Apple has permanently disabled the ECC capability then are going to run into big time problems with their
"... Use OpenCL to incorporate advanced numerical and data analytics features, perform cutting-edge image and media processing, and deliver accurate physics simulations. ... "
marketing campaign. Lack of ECC is a joke for anyone with individual data points that are actually worth something.
Long term, it isn't going to help them by flushing "FirePro" features down the drain while still slapping a large price premium on the cards.