10 Things We Hate About Nvidia

[Endymio]

I guess the moral to the story is that, since this isn't Hard Drive or The Onion, most readers here, or at least the most vocal ones, are very particular about the expectations that are set when we read the headlines here.

Nonsense. Had this instead been a hit piece on the company, with a disingenuous headline that read "Ten Things we Love About NVidia", those same individuals would have posted a string of glowing comments praising it.

 

[Kn0xx]

CUDA's lock on that market is no joke. I've done some work with neural networks in the past. I never used CUDA directly, I used tensorflow and associated utilities, and those DO support using other toolkits than CUDA. In theory. In practice....

AMD ROCm as shipped by AMD is built to only support the high end/data center GPUs -- it can be built from source to support others, but they also don't support sharing the GPU between graphical use and compute.

I’m wondering, since you have more expirence in the field. Does AMD can match Nvidia on raw power/processing or does it needs a better software stack ?

Im asking, because AMD "beated" Intel not because of "features", but raw-power and efficiency. sacrificing the all known "plug-n-play" of Intel CPU's. Does Nvidia will face the same challenge ? Raw-Power vs Reliability ?

I can remember how a shock was switching from Intel to AMD, because on AMD I had to make some tuning in the bios for the CPU to get more consistent, where Intel was better in that regard.