Message boards :
Number crunching :
NPU inclusion?
Message board moderation
| Author | Message |
|---|---|
|
Send message Joined: 2 Nov 17 Posts: 1 Credit: 35,521 RAC: 0 |
Few words about future of tech & NPUs? What do you know? What do you think? Is NPU going to be included into LHC@h? What about some other project? Will it be limited to CPU-NPU (like intel or AMD) or will it include some other NPU devices? Such as Googles Tensor & Coral TPU/NPU? Falcon cards? Lets start the discussion here...   non-profit org. Play4Life in Zagreb, Croatia, EU |
 Send message Joined: 9 Feb 09 Posts: 25 Credit: 2,785,379 RAC: 7,052  
|
In the near future it will not be and if it is it will not be with NPU. I think it could be with a dedicated graphics card exclusively for AI within desktop computers, having 2 graphics cards, one for games and one for AI or using the integrated graphics in the cpu and using the card for AI. An RTX 2060 is 5 times more powerful than the best NPU that will hit the market in a few months and is the most basic card for at least 3 years. The first problem is that in the world the large Asian chip foundry companies (without Western competition in the sector) are all immersed and collecting resources in the manufacture of AI chips mainly for the big brands Amd/nvidia/ampere/qualcon/ exynos... These Asian companies agree on prices among themselves and these annual increases are transferred to the final consumer. If we add the factory increases of cpu/gpu/ddr/hbm3 memories, it is a joint increase in their value of 1040%, which creates enormous inflation in the West since none of the cpu/gpu brands can avoid that increase for the consumer. Really, the most advanced in AI is Nvidia, which in 2005 made a very big bet on R&D in both hardware and software that paid off with Cuda and tensors. Tesla and other companies already having the previous generation of Nvidia will buy the GB200 CPU+integrated GPU. The market will suffer a temporary shortage of GPUs due to the high demand for AI chips, including the memories that are needed. USA sanctions China but those sanctions do not really work, thousands of new companies are created that buy enormous quantities of AI gpu....orders of 105 million dollars where that order is delivered, that order goes to China and the company is already hill to avoid responsibilities. You can exercise real control in your country, not in that of others. An AI in England has managed to discover how to make magnets without using rare earths after 6 months of work, achieving magnets that are 20% cheaper and 70% more ecological, it is the first evidence of the result of AI as we know it in scientific development . |
|
Send message Joined: 9 Feb 09 Posts: 25 Credit: 2,785,379 RAC: 7,052
|
Mistral IA https://mistral.ai/ |
©2025 CERN