Hearing Gao Nian's introduction, the expression on Nvidia CEO Huang's face changed abruptly.
It was no wonder that Nvidia CEO Huang's expression would suddenly change so drastically; the fact was that Geek Technology was learning from Nvidia.
As early as 2005, Nvidia had already begun to think about using the multiple stream processors of graphics cards for parallel computing.
Finally, after more than a year, in 2006, Nvidia launched the CUDA unified computing architecture, allowing researchers, developers, and enterprises to start using CUDA for development and general-purpose parallel computing.
This book is first published on the entire network by π₯ π¨ π π π . π π π
This also marked that the role of graphics cards was no longer just a simple graphics card; it had become an invisible general-purpose computer!
Therefore, thanks to the parallel computing advantages brought by hundreds of stream processors, CUDA became an instant hit, and Nvidia's graphics card sales also began to explode.
After all, in the past, people bought discrete graphics cards for the purpose of playing games or doing graphic rendering for film and television works.
Now, graphics cards have additional uses such as scientific computing, engineering simulation, and deep learning training. How could Nvidia's graphics card sales not skyrocket?
Therefore, after Nvidia's graphics card sales ushered in a crazy surge, it naturally brought huge profits to Nvidia.
Thus, Nvidia became the richest company in the computer graphics card field.
And the most crucial thing was that Nvidia did not spend the money indiscriminately after becoming rich; instead, they used it for new technology research and development.
Thus, in the computer graphics card market, a classic plot of the strong getting stronger was staged.
Conversely, Nvidia's competitor, ATI, became increasingly lonely in the fierce competition and was eventually acquired by AMD.
It can be said that Nvidia's ability to rise to this point owes half of its success to the assistance of the CUDA unified computing architecture!
Now, Geek Technology has also entered the arena, adding another graphics card and parallel computing system that supports parallel computing.
This is truly terrible news for Nvidia!
But Gao Nian did not pay attention to Nvidia CEO Huang's ugly face. Gao Nian on the stage said with a smile:
"In a parallel computing system, the most important thing is the number of stream processors, followed by how to allocate computing tasks so that the stream processors can work together.
Instead of having some stream processors working at full load while other stream processors are just watching from the sidelines, doing nothing.
Therefore, the software level is also particularly important. So, we developed the Fuxi parallel computing platform and programming modelβ"
Gao Nian on the stage introduced the Fuxi parallel computing platform and programming model, which is basically similar to Nvidia's CUDA.
It can be said that with this Fuxi parallel computing platform and programming model, coupled with the powerful Phoenix graphics card,
This is naturally good news for countless developers.
As for the price, the Fuxi parallel computing platform and programming model developed by Geek Technology, like CUDA, is free.
After all, since others are providing it for free, Geek Technology naturally cannot charge for it.
Of course, although the Fuxi parallel computing platform and programming model are free and seem to have no benefits, instead losing research and development funding,
Gao Nian, as a transmigrator, deeply knows how much CUDA helped Nvidia in his previous life.
It can be said that more than 80% of Nvidia's trillion market value in his previous life was supported by the CUDA system!
It was because they made graphics cards no longer just graphics cards, but instead turned them into general-purpose parallel computers.
Successfully establishing a complete parallel computing hardware and software ecosystem is what led to the brilliant achievements later on.
Otherwise, just relying on the discrete graphics card business and those game players, there's no way Nvidia's market value could break through trillion US dollars.
Of course, the Fuxi parallel computing platform and programming model are ultimately professional fields and have little to do with ordinary consumers.
After all, the purpose of ordinary consumers buying computer graphics cards is actually to play games.
Therefore, Gao Nian naturally would not linger too long at the press conference.
After spending more than ten minutes introducing the Fuxi parallel computing platform and programming model, Gao Nian officially arrived at the graphics card introduction segment.
"Next is the hardware parameter introduction segment for the Phoenix graphics card."
When Gao Nian said this, the hardware parameters appeared on the projection screen behind him.
Phoenix 1080 Graphics Card Manufacturing Process: 28nm manufacturing process.
Chip Area: 594 square millimeters.
Transistor Count: billion transistors.
Core Frequency: GHz ~ GHz.
Stream Processors: 3072 stream processors.
Texture Units: 328 texture units.
Raster Units: 80 raster units.
Memory Type: GDDR5 memory.
Memory Bus Width: 256-bit.
Memory Bandwidth: 192GB/s.
Memory: 6GB.
AI Acceleration Hardware Unit: Integrated NPU neural network processor.
Maximum Resolution: (maximum support 2K resolution).
Interface: / interface.
Support: DirectX11, , Frame Generation Technology. β β .
Power Consumption: 350W .. β .
Looking at these amazing hardware parameters, the people at the scene widened their eyes and looked at the projection screen with disbelief.
Because the paper data of the Phoenix 1080 graphics card was too strong, completely exceeding people's imagination.
First of all, the number of stream processors. The current GTX580 graphics card from Nvidia only has 512 stream processors.
Even the GTX590 graphics card, which integrates two graphics card chips, only has a total of 1024 stream processors!
You've directly upgraded to 3072 stream processors, which is simply too exaggerated.
The number of texture units and raster units is equally exaggerated.
Because the GTX590 graphics card only has a mere 64 texture units and only 32 raster units.
You've directly gotten 328 texture units and 80 raster units. Aren't these hardware parameters too exaggerated?
Of course, many people who know a little bit about hardware also know what the cost is to achieve such hardware parameters through this hardware parameter list.
The first cost, and the most important cost, is that the chip area is super large, reaching an astonishing 594 square millimeters.
The larger the chip area, the lower the yield rate of the chip, resulting in higher chip prices.
At this time, the chip area of this Phoenix graphics card has reached an astonishing 594 square millimeters, which means that the chip yield rate is very low, which ultimately leads to a rather touching price for the graphics card.
Of course, a low chip yield rate does not mean that 'bad chips' cannot be sold, because there will definitely be over-design when the chip is designed.
For example, in a certain year, Intel developed the 8-core Core i7 CPU processor.
Because chip production is too complicated, there are too many procedures, and any error can cause a certain part of the chip's circuit to fail to operate.
But a certain part of the circuit failing to operate does not mean that the entire chip cannot operate.
By shielding the broken part of the circuit, you can actually take the failed chips and sell them.
