Huang Renxun put the red marker back into the whiteboard slot, turned around and looked out the window.
"Do you have any plans tonight?"
Zhao Chen Close the Henglong Research Institute manual and put it back into Kraft Paper File Bag.
"No."
"Okay. I know a Japanese food store, not far from here, on an inconspicuous commercial street Santa Clara."
Huang Renxun took off his leather jacket and put it on his arm, revealing the black T-shirt underneath--
The Nvidia logo is printed on the chest, and the collar has been washed a little loosely.
"It's not a Michelin-starred one that you need to book weeks in advance, but it's just a place I go to regularly.
The boss knows me and will reserve the corner seat where I am used to sitting."
He picked up the landline on the table and called his assistant, and simply said, "All trips tonight are cancelled."
Then he hung up the phone and put his leather jacket back on.
"We won't talk about Financial Report tonight, we won't talk about the stock price, and we won't talk about the growth rate of Data Center."
"What are we talking about?"
Huang Renxun walked to the door, put his hand on the door handle, and paused for a moment.
"Talk about something no one usually listens to."
The Japanese food store is hidden at the end of a commercial street Santa Clara.
Next door is a print shop and a closed dry cleaner.
The boss is a middle-aged Japan man wearing a dark blue kimono.
When he saw Huang Renxun pushing the door in, he just bowed slightly.
Without any extra greetings, they were led directly to the innermost corner of the bar against the wall.
"I have been eating at this restaurant for many years."
Huang Renxun took off his leather jacket and hung it on the back of the chair. He took the hot towel handed by the boss and wiped his hands.
"The first time I came here was when Nvidia first moved to Santa Clara.
At that time, the company had less than one quarter of cash flow left on its books.
I was talking about financing with an investor at Starbucks across the street, and the other person said that GPUs have no future.
After the conversation, I sat here alone and ate eel rice.
After eating, I went back to the office and continued to modify the circuit diagram of NV1.
Later, every time I encountered difficulties, I would come here to sit down."
The boss brought two cups of warm brown rice tea and a plate of edamame, and Huang Renxun pushed one cup in front of Zhao Chen.
The last ray of dusk outside the window was fading from the signboards of the commercial street, and the low hum of traffic could be heard faintly on Highway 101 in the distance.
Zhao Chen took the teacup and gently rubbed the rough ceramic cup wall with his fingertips.
"You just said that you did chip design at AMD and LSI Logic when you were young."
Zhao Chen picked up the tea cup and took a sip. The burnt aroma of brown rice tea melted on the tip of the tongue.
"AMD was in 1984, LSI Logic was in 1985.
At that time, AMD was still fighting against the x86 compatible chip Intel.
The team I was in was responsible for designing a math coprocessor, but the project was later canceled.
LSI Logic learned a lot during those years -
The entire process of ASIC design, the connection with the wafer factory, and the cost control of chip mass production were all accumulated during that time.
But what really made me decide to start my own business was my last year at LSI Logic.
At that time, I discovered a trend - CPU performance growth could not keep up with the demand for graphics computing.
A general-purpose processor can never render pixels as efficiently as a specialized chip.
This judgment was very niche at the time, because the entire Semiconductors industry was pursuing CPUs.
But I have done too many ASIC projects at LSI Logic, and I know that the performance ceiling of dedicated hardware is much higher than that of general-purpose processors."
He put down his tea cup and drew a simple chip architecture diagram on the bar table with his finger.
When your fingertips scratch the wood grain, they make a very subtle friction sound.
"The essence of graphics rendering is massive parallel computing, and the CPU is designed for serial tasks.
This structural mismatch is the opportunity for Nvidia."
"So the founding gene of Nvidia is not 'to be a chip company', but to 'solve parallel computing problems'.
Graphics rendering is the earliest parallel computing application scenario, but it is not the only one.
Later CUDA, Data Center, and AI training—all grew out of this gene."
Huang Renxun picked up the tea cup and looked at Zhao Chen through the hazy water vapor on the bar, with a hint of surprise in his eyes.
"You are the first person to catch the underlying logic of Nvidia at a glance without any technical background.
Most investors had to wait until CUDA came out before they could barely understand what we were doing.
Some people didn't realize it until Data Center's revenue exceeded that of the game.
Before you buy Nvidia, you have stretched this line from 1993 to 2018."
"I read your technical review in "Electronic Engineering Times" in 1993.
At that time, you had just founded Nvidia less than half a year ago, and the NV1 architecture was still in the design stage.
You said in your article that the computational complexity of graphics rendering will grow faster than the Moore's Law of general-purpose CPUs in the next ten years.
Dedicated graphics processors are not optional peripherals for PCs, they are the infrastructure for the next era of computing.
This article is the earliest and most accurate expression of the core strategy of Nvidia among all the information I have read about Nvidia.
I saw the comparison between 'optional peripherals' and 'infrastructure'.
Just know this guy isn't selling graphics cards - he's defining a computing paradigm that hasn't been named yet."
Huang Renxun put down the tea cup, and the bottom of the cup hit the bar with a soft sound.
He put his elbows on the bar, leaned forward, and lowered his voice.
The passion I felt on the GTC stage in front of the whiteboard just now has faded.
It was replaced by a tired and frank tone that could only be expressed at the Canteen bar late at night.
"After I finished writing that article, the editor called me and said that the content was too technical and could not be understood by ordinary readers.
I said that I did not write for ordinary readers, I wrote for myself.
In 1993, the three of us signed the company registration documents at Denny's restaurant with only two hundred US Dollar in our pockets.
I don't have the money to hire a consulting company to help us analyze the market prospects, I don't have the money to do road shows, and I don't have the money to convince investors.
My only capital is my own judgment on the trend of parallel computing.
I wrote this judgment into an article and submitted it to industry publications, not for publicity--
It's to force me to think clearly.
If I can't use the logic of an article to clearly demonstrate the proposition that 'graphics processing will become computing infrastructure',
Then I am not qualified to meet with Sequoia people with a business plan."
Zhao Chen picked up the brown rice tea and took a slow sip.
"In that article you demonstrated the parallel computing nature of graphics rendering.
But then you did something crazier than that article - CUDA.
You turned a company that sells graphics cards into a general-purpose parallel computing platform.
When CUDA first came out in 2006, the market was full of doubts--
The power consumption and cost of GPU are not suitable for general computing,
Developers will not learn new programming for a non-mainstream platform Model. The market for parallel computing is too small.
Wall Street said you wasted billions of US Dollar R&D expenses on a project with no short-term commercial return.
What makes you, when no one is optimistic about you,
Betting billions US Dollar on a technological paradigm that will take more than ten years to materialize?"
Huang Renxun placed the tea cup on the bar and slowly turned his fingers around the rim of the cup.
The boss served two plates of sashimi - fish and sweet shrimp. The presentation was simple and neat, without any unnecessary decoration.
He picked up a piece of fish with chopsticks and put it into his mouth, chewed it a few times, then placed the chopsticks across the edge of the plate.
His eyes fell on the dish of sashimi, but the focus was obviously not on the fish, but on some place further away.