I’m planning to create a pc for deep learning (artificial intelligence) that doesn’t become obsolete in a few years.
After searching among all available sockets, I have come to the conclusion that the 2011-v3 is the most suitable as it gives the chance to have up to 128GB of RAM (to load big datasets in RAM) DDR4 quad-channel and processors have minimum of 6 cores (12 threads).
The most important part of this computer is the GPU which will compute all the hard calculus. I’ll buy a 1070 gtx and maybe add another one in the future. My concern is how the PCI Express lane number affects the performance of the GPUs when connected in SLI. I don’t know which CPU buy: i7-5820K (28 lane) or i7-5930K (40 lane). The first one accepts the following configurations: x16/x16x8/x8x8x8/x8x8x4x4 versus the second one x16/x16x16/x16x8x8/x16x8x8x8/x8x8x8x8x8. ¿Is the difference between x16x8 and x16x16 or x8x8x8 and x16x8x8 in 1070 gtx (or similar) worth paying 300€ extra (processor and motherboard)?
After searching among all available sockets, I have come to the conclusion that the 2011-v3 is the most suitable as it gives the chance to have up to 128GB of RAM (to load big datasets in RAM) DDR4 quad-channel and processors have minimum of 6 cores (12 threads).
The most important part of this computer is the GPU which will compute all the hard calculus. I’ll buy a 1070 gtx and maybe add another one in the future. My concern is how the PCI Express lane number affects the performance of the GPUs when connected in SLI. I don’t know which CPU buy: i7-5820K (28 lane) or i7-5930K (40 lane). The first one accepts the following configurations: x16/x16x8/x8x8x8/x8x8x4x4 versus the second one x16/x16x16/x16x8x8/x16x8x8x8/x8x8x8x8x8. ¿Is the difference between x16x8 and x16x16 or x8x8x8 and x16x8x8 in 1070 gtx (or similar) worth paying 300€ extra (processor and motherboard)?
Facebook
Twitter
Instagram