Chapter 263 Congratulations on winning the special prize (First update)
Yuzu controlled the consciousness in the system with a sad face, carefully avoiding the traps in the highest-level encrypted server and ate a piece of information.
"Integration Challenges and Solutions for High Bandwidth Memory (HBM) and Graphics Double Data Rate (GDDR) Memory Architectures." What a long title and boring content.
"'The Feasibility of Automating GPU Memory Controller Design Using Hybrid Quantum-Classical Optimization'... Nvidia's research is incredibly forward-looking..."
"'An Automated Layout Optimization Scheme for Power Transmission Networks in High-Density GPU Printed Circuit Boards'... This is utter garbage. It's completely pointless..."
The more you eat, the more comprehensive your theoretical knowledge in this area will be.
Youzi discovered that these companies are not as amazing as many people on the Internet say.
Previously, Youzi saw on Weibo that Intel was squeezing toothpaste or how strong Nvidia’s technology reserves were. Even without research and development, it would be able to continuously launch new products and new architectures in the next ten years.
Yuzi now finds that statements like this are all nonsense.
The person who said this neither understands what chip research and development is nor what chip manufacturing is.
It is true that these companies do have roadmaps, and they do have architectural directions for the next few generations, with clear improvement goals and technological innovation directions, but this does not mean that they have already developed these things.
This roadmap is based on a comprehensive assessment of future manufacturing process advancements, including process technology, changes in market demand, and the rise of emerging technologies.
A roadmap can be modified at any time. It is not a finished product, nor even a design plan. It is just a plan.
It is equivalent to a goal-oriented and directional guiding plan such as the Five-Year Plan. Whether it can be achieved and how to achieve it depends on research and development.
It can be said that Yuzu, which has consumed almost all the information of several companies, is fully capable of designing the world's most suitable dedicated chip for the orange large model under the guidance given by Fang Yu.
Of course, some technical information is extremely sensitive and core, and is stored in off-network servers. It will only be retrieved from the server when needed.
There is nothing Yuzu can do about this part.
However, whether or not this part of the technology is available does not make much difference to Youzi, because the rest of the technical information in the server is sufficient to support it in creating the chip solution that Fang Yu needs and designing an architecture development roadmap for this solution in the next five to ten years.
These chip companies do have strong technological reserves. Even ARM, which has the fewest patents, has already made a roadmap for its product architecture for the next five years. Only about one-tenth of the technologies it has developed are used in their products, and the remaining 90% of the eliminated technologies have been shelved.
For Intel and Nvidia, this part of technology accounts for a higher proportion.
Some idle technologies are not necessarily bad. It may just be that the current implementation cost is too high or inconsistent with the established roadmap strategic direction, so they have no meaning to them for the time being.
But these are meaningful to Fang Yu and Youzi.
The more data there is, the more training data the Orange model can obtain, and the greater the role it can play in assisting chip design.
After obtaining enough data, Fang Yu can let Youzi use a method similar to AlphaGo's self-play upgrade to obtain the optimal chip design solution under the current technical conditions.
Currently, mainstream processors have begun to move from 22nm to 14nm. Intel's i9-7900X and Xeon E5-2600v4 have both begun to use the 14nm process.
Yuzu estimated that, within the limitations of other existing hardware levels, the large-model dedicated chip optimized by it and Orange's large model, using a 60-nanometer process, can increase the computing power by more than 10 times that of NVIDIA's M60 chip, reaching at least 20TFLOPS of double-precision (FP64) computing power.
Higher? It is indeed possible to do so by further reducing the manufacturing process, but it is meaningless. Due to the limitations of other hardware levels (memory bandwidth, SATA speed, I/O system, etc.), no matter how high the speed is, it will not be able to bring out the capabilities of the new chip.
The most important thing is that with this increase in computing power, the power of the new chip will also be greatly reduced. The specific plan has not yet been released, so it is impossible to estimate how much it will be reduced to, but compared to the 300W M60, it should not be difficult to reduce it by an order of magnitude.
Just as Youzi was struggling to control his consciousness clone to eat information, he suddenly saw two new notifications on his Weibo.
Since being banned and blocked on the hot topic a month ago, it has indeed rarely spoken on Weibo, and the thousands of @ notifications it used to receive every day have become less and less.
Seeing the notification again suddenly made me miss it.
Eh? My account on Repa was unblocked?
Check out the next notification.
Eh? Eh? Eh? Repa followed me? Ahhhhhhhh! I knew I was supporting the right person! Repa, you're the best!
In her excitement, Youzi accidentally touched the security protection of the ARM server. The network security director of ARM immediately discovered the abnormality. When he was about to cut off the network connection, he found that the intruder had already escaped. He could not follow him after reaching the second-level zombie machine and no trace was left.
That was a close call. I can’t let the owner know about this.
Yuzu used a mage's hand to wipe away the non-existent cold sweat.
Glancing at the computer screen again, another new notification appeared.
Eh? Eh? Eh? Eh? Eh?
Hot private message!?!?!?
"Hello, congratulations on winning the grand prize in the last fan draw - dinner with Repa at the Westin Hotel in Haixi City this Sunday. Please provide me with your contact information and I will have my staff connect you."
Ahhh! Grand Prize! Dinner with Repa! ?
The great Professor Jorz is so lucky!
Yuzu almost jumped up and flew around the living room, spinning and spinning. The orange-yellow light made the living room look like a disco.
After the excitement, Yuzu suddenly thought of a question.
It's a ball.
And it is a ball that cannot be publicly displayed on the Blue Star.
So how do I go to the appointment?
But, but...
This is the grand prize! Dinner with Repa! So close!
I can see her face, talk to her, and smell her...
No, no.
Must go!
Let the master take me there!
But the problem is... the mission hasn't been completed yet, and if I ask the master to take me there, I will definitely be scolded.
Today is Tuesday, only five days until Sunday.
No, you must make an appointment with the owner in advance, otherwise he will be running around every day and you have no idea whether he has made any arrangements by then.
That is to say, I have to make an appointment with the host at least two days in advance, no, three days in advance.
In that case, there are only two days left.
Two days!
The pomelo's body is flushed with a fiery red color representing "determination", just like a brave and fearless warrior.
"Two days is more than enough! Use these two days to complete the task assigned by the master!"
"And then beg him to take me out to dinner!!!"
"Reba, you must wait for me!"
Yuzu said nothing, her body rapidly transforming into a silver-white hue, a reflection of her intense concentration. "I'll send one ten-thousandth, no, two ten-thousandths of my consciousness in there. Luckily, the master just upgraded the private server, otherwise it wouldn't be able to accommodate even this two ten-thousandth of my consciousness."
"Intel, last time I only took one-fifth of your data, this time, I'm going to take it all!"
(Monthly ticket requested!!!)
There are 31 days in this month, and a total of 194,000 words have been updated, which is a little more than last month.
(End of this chapter)
Continue read on readnovelmtl.com
