T+a dac 8 dsd

[Sam Lord]

Well the P100 is a totally new chip, 15 billion transistors, the biggest GPU on the block. It's battling the Xeon Phi even while designed to work with it.

 

Sent from my Nexus 6P using Computer Audiophile mobile app

[Sam Lord]

Thank you Jussi. I have no answer at all. I should have known you were following these various gpus. You've helped me out because some of the cards are just becoming available, and a project before me must make a decision. I'm neither a programmer nor physicist (yet) but am part of the team.

 

It seems to me that it's either very local architecture, such as larger registers within the cores, or the vertically attached RAM.

 

We're installing a developer system to give practice for physics students and faculty on both Nvidia CUDA coding and Intel Xeon Phi (exclusive) coding. It has a single Knights Landing cpu and will have one Nvidia card installed.

 

We won't be able to compare NVLink against Intels interconnect fabric, but will get some idea of CUDA vs Xeon Phi in programming efficiency. These new products and their libraries will inform our HPC decision in just a few months. Wish I knew more.

[Sam Lord]

On 2/26/2018 at 8:17 PM, dmance said:

So please understand that a raspberry Pi delivers the bits as reliably as a $10,000 server ...to your USB DAC ...but sounds worse because it throws huge noise into the USB. You can make the raspberry sound the same by adding an intona and $50 worth of ferrites.

 

But nobody reading this post will believe or try this. 

 

There are like 5000 posts on CA saying the same thing, essentially.  Okay not exactly with all those ferrites, better to use stardust** in a cable, it's less reactive.  You could start reading the posts, or... believe it.  If that $$$ server wins by just a hair, or even a lot more, I'll stick with a quiet, modest NAA.  Because I know it works very well for HQP.  

 

** Shunyata's FeSi material.