NVIDIA 465.19.01 Linux Beta Driver Released

[phoronix]

Phoronix: NVIDIA 465.19.01 Linux Beta Driver Released

While looking forward to the NVIDIA 470 series Linux driver for Wayland support improvements, before getting there NVIDIA is first introducing the 465 driver series. Today marks the first publicly available NVIDIA 465 Linux driver beta...

https://www.phoronix.com/scan.php?page=news_item&px=NVIDIA-465-Linux-Beta

[zexelon]

Hey Michael, when your done with those can I buy them off you??

[skeevy420]

hey michael, when your done with those can i buy them off you??

rofl :d

[ThoreauHD]

Kudos on yelling "I got water!" in a desert. You got cajones. Let's hope the next testsuite addition isn't Darwin's SOTF Benchmark.

[zexelon]

skeevy420 Hey nothing ventured nothing gained right?

The key here is to get the first post in when begging!

[Danniello]

Hey Michael, when your done with those can I buy them off you??

Be good Linux user and buy Linux friendly GPU, so definitely not proprietary lock-in products like nVidia GeForce.

I know that there is problem with buying AMD GPU on the market, but please be patient

PS. Michael, I also like Jack Danniell's

[zexelon]

Be good Linux user and buy Linux friendly GPU, so definitely not proprietary lock-in products like nVidia GeForce.

Alas, the reality is that while AMD/ATI have made significant improvements over the last couple of years with their hardware and software, they have nothing comparable to a 3090/CUDA setup. The vast majority of my day is spent in Tensorflow programming and ROCm is really about 10 years behind and even still looks like its just a pet project not a focused investment.

So yah, while I would really like to explore an AMD/ATI linux desktop... I really have no use for it outside of rendering my desktop background and some windows.

My Threadripper, Gigabyte 3090 combo took my ai training from 8.5 hrs to 1hr when it was built... and that was not even on an optimized code path. AMD/ATI just cant touch that.

[Danniello]

Alas, the reality is that while AMD/ATI have made significant improvements over the last couple of years with their hardware and software, they have nothing comparable to a 3090/CUDA setup. The vast majority of my day is spent in Tensorflow programming and ROCm is really about 10 years behind and even still looks like its just a pet project not a focused investment.

Your setup require CUDA®, so I understand that you have no other option than lock-in with nVidia, because AMD/ROCm is not good enough...

My knowledge with that topic is zero, but I hope that in the future ROCm or some Intel or Khronos equivalent will be "good enough" to replace or maybe even emulate CUDA® like Vulkan is quite good with emulating DirectX...

[zexelon]

Your setup require CUDA®, so I understand that you have no other option than lock-in with nVidia, because AMD/ROCm is not good enough...

My knowledge with that topic is zero, but I hope that in the future ROCm or some Intel or Khronos equivalent will be "good enough" to replace or maybe even emulate CUDA® like Vulkan is quite good with emulating DirectX...

Honestly, I hope that AMD/ROCm commit to their game! They are on the right path, if they stick to it and invest appropriately in it, it wont take them long to catchup to CUDA. It would be awesome to have a legit second in the A.I. space!

[mppix]

Alas, the reality is that while AMD/ATI have made significant improvements over the last couple of years with their hardware and software, they have nothing comparable to a 3090/CUDA setup. The vast majority of my day is spent in Tensorflow programming and ROCm is really about 10 years behind and even still looks like its just a pet project not a focused investment.

So yah, while I would really like to explore an AMD/ATI linux desktop... I really have no use for it outside of rendering my desktop background and some windows.

My Threadripper, Gigabyte 3090 combo took my ai training from 8.5 hrs to 1hr when it was built... and that was not even on an optimized code path. AMD/ATI just cant touch that.

Why not?
I'm running a stack of Titans myself in tensorflow and pytorch. Both fully support ROCm now. You just need a good CDNA GPU..