HackerTrans
TopNewTrendsCommentsPastAskShowJobs

zepmck

no profile record

Submissions

[untitled]

1 points·by zepmck·há 3 anos·0 comments

Operation of IBM Summit system gets extended to 2024

olcf.ornl.gov
1 points·by zepmck·há 3 anos·1 comments

[untitled]

1 points·by zepmck·há 4 anos·0 comments

ARM Neoverse V2 Nvidia Grace CPU

4 points·by zepmck·há 4 anos·0 comments

comments

zepmck
·há 3 anos·discuss
NVIDIA is showing real performance through using independent MLPerf benchmark.
zepmck
·há 3 anos·discuss
Does this hide any problem with AMD Frontier?
zepmck
·há 3 anos·discuss
Software developers, especially for certain type of components, e.g. compilers, are NOT easy to find out.
zepmck
·há 3 anos·discuss
The main reasons of "monopoly" are long-term vision and enormous effort spent in software development. AMD is lacking behind because they have not yet seriously invested on their own software stack.
zepmck
·há 3 anos·discuss
The next challenge is software stack.
zepmck
·há 4 anos·discuss
People pretending to do professional work with low budget/quality workstation. If you cannot afford to operate a 4090 according to specs, you cannot blame NVIDIA if things go wrong.
zepmck
·há 4 anos·discuss
You blame CUDA because you have never played AMD + ROCm. Furthermore OpenCL is on dead-end street. With standard programming languages natively supporting CUDA, it has no sense to add extra layers.
zepmck
·há 4 anos·discuss
The most powerful and unfortunately unusable supercomputer of the world. AMD's approach to GPUs is on a failing track since its inception. The only software stack available is super fragile, buggy and barely supported. Rather than building a HPL machine I would have preferred see public money spent in a different way.
zepmck
·há 5 anos·discuss
If you have GPUs, NVTabular outperforms most of the frameworks out there: https://github.com/NVIDIA/NVTabular
zepmck
·há 5 anos·discuss
This not completely correct because while a carrier transport anything indescrminately, social networks manipulate and control information exchange, for instance giving more priority to a content wrt to another.
zepmck
·há 5 anos·discuss
I think this work is quite ridiculous. As far as I could see, they trained the net with a batch size equal to 1.
zepmck
·há 5 anos·discuss
In terms of PCI lanes efficiency, there is no competition between Intel and AMD. Intel is much ahead wrt AMD. Don't not be impressed about number of lanes available on the board.
zepmck
·há 5 anos·discuss
When? Where? How can it be miles ahead if the hardware has not been released yet?
zepmck
·há 5 anos·discuss
One-API is not completely open source. Support for Ponte Vecchio will be not released open source for many reasons.