Hi, everyone! I'm a newcomer.
Recently I want to select a new laptop for my daily drive.
I'm both a gamer and a programmer.
I noticed the AMD's new products are very competitive in consumer platform.
In fact, 6000 series high-end laptop platform really touched me.
But unfortunately, i find out newer ROCm will only officially support professional and datacenter cards.
ROCm 5.x
ROCm 4.x
My development sometimes includes AI, Computer Vision and some GPU-based acceleration.
Most time competiter's solution "CUDA" is the major tool to develop.
But since ROCm started in 2016, I really hope it can become popular.
I have a RX580 8G card, I tried a lot development on that card with early version ROCm (2.x before).
But AMD developers found a serious bug since 3.x version and didn't continue this type cards support.
As you guessed, it's really make me upset.
I have to choose update Ubuntu instead of use the older version ROCm because of other requirements.
Since then, I bought a new laptop with Intel CPU & NVIDIA's GPU.
Due to the requirements, i have to choose such combinations.
I really like 3000/5000 Series AMD laptop platform and some AMD's mobile graphics solution at that time.
But AMD's scientific calculation and AI supports are really not good enough in both CPU & GPU.
Software Ecosystems with AMD's two competitors are convenient and really easy to apply on consumer platform.
For example, I have an older card than RX580, GTX960 2G from NVIDIA.
It barely can run recent games due to the limit of performance.
But I can use CUDA Toolkit to run some light weight task to achieve my develop requirements.
For my RX580, it only can run some games. I can not use it with other purpose.
So I just want to confirm whether will AMD improve the scientific computing & AI on the consumer platforms or not.
I'm not sure AMD's bussiness strategies can provide this solution.
I spent a lot of time in the computer science labs.
Maybe most researchers know that all cost-limited AI training server are Intel + NVIDIA's consumer platforms.
I don't think all the labs and researchers can afford high-end server.
For cloud, that's not very comfortable for offline application & pre-research because of some security reason.
Besides, AMD has bought Xilinx. I'm also a Xilinx FPGA developer for acceleration.
Vitis & Vivado Products are not very compatible with its own order version in most versions.
I need to prepare about 5 to 6 version to handle the project version problems.
That's very big for storage. I think it takes my 160GiB storage maybe.
They are also very unstable and would like to eat up all the system memories. 16GB is always not enough.
It seems that Quartus Prime will perform better when develop FPGA application.
At last, I hope AMD could really pay attention to the Software Toolchains.
Both hardware and software optimization will perform maximum potential.
Thanks for AMD's engineers hardwork on the excellent hardware and best wishes to AMD's software developers!
