[TBBT] Ch: 33 [The Billion-Dollar Distraction] (Patreon)
Content
AN: Little warning, the first half contains too much science stuff. I had to google half the shit. But it turned out good.
--
"What if I built something specifically for these calculations?"
I could purchase a more powerful workstation or even rent time on a supercomputer, but neither option particularly appealed to me. Considering what I knew about the technological advancements that would take place over the next twenty years, building something myself seemed far more interesting.
I put the marker down, returned to my desk, and opened a fresh page in my notebook.
My first idea was to build a computing cluster using PlayStation 3 consoles. Sony's PS3 used IBM's Cell processor, which was particularly good at certain parallel calculations. Researchers were already experimenting with connecting multiple consoles to create inexpensive supercomputers, and I'd recently read about Gaurav Khanna, an astrophysicist who'd connected eight PS3s to run gravitational-wave simulations involving black holes.
I also remembered something considerably more impressive from my previous life. In 2010, the United States Air Force would unveil the Condor Cluster, a supercomputer built using 1,760 PlayStation 3 processors alongside conventional computing hardware. It was an economical way to obtain substantial computing power, and the system would be used for image processing and other demanding research.
"Seventeen hundred consoles..." I murmured, reaching for my calculator. "That's certainly one way to build a supercomputer."
The hardware alone would cost hundreds of thousands of dollars, and I'd need additional money for networking equipment, cooling systems, electricity, and somewhere to house everything.
I could afford to experiment with a smaller version, and financing a larger installation was certainly possible. However, the more I considered the idea, the less attractive it became.
For one thing, Sony would remove the OtherOS feature in 2010, preventing users from installing Linux on updated consoles. The newer PS3 Slim models wouldn't support it either. I could avoid updating existing machines, but finding suitable replacements would eventually become troublesome.
The consoles also had limited memory, and their architecture wasn't suitable for every type of scientific calculation. I'd be building an enormous collection of gaming hardware that would become increasingly outdated as newer processors entered the market.
Besides, even if I managed to build something comparable to the Condor Cluster three years ahead of schedule, what would that actually accomplish?
I'd have a powerful computer, certainly, but I'd still be dependent on hardware designed and manufactured by someone else.
I leaned back in my chair and considered the problem from another angle. What I really needed was a computing platform that I could improve continuously, adapting its architecture to whatever research I happened to be pursuing.
And preferably, something I could eventually manufacture and sell.
That brought me to the idea of developing a dedicated mathematical accelerator.
The concept wasn't entirely new. CPUs were excellent general-purpose processors, while GPUs could perform enormous numbers of calculations simultaneously. NVIDIA had introduced CUDA the previous year, and GPU computing was already attracting considerable interest among researchers.
However, I knew where the industry was heading.
Google would eventually develop Tensor Processing Units, while NVIDIA would introduce Tensor Cores specifically designed to accelerate mathematical operations used in machine learning and artificial intelligence.
Those technologies would become incredibly important over the next two decades.
And there was absolutely no reason I had to wait for those companies to invent them.
I stood up, walked over to the whiteboard, and erased the equations I'd been working on.
"All right. Let's see what we can do."
I wrote a heading across the top.
Peterson Matrix Accelerator (PMA)
The basic idea was straightforward. I'd develop a specialized processor capable of performing matrix multiplication, vector operations, and other mathematical calculations far more efficiently than a conventional CPU.
For my first prototype, I could use a field-programmable gate array, or FPGA. It would allow me to test different processor architectures without spending millions of dollars manufacturing custom silicon.
Once I had a working design, I could move toward developing my own chips.
Of course, manufacturing technology in 2007 was far behind what I'd known in 2026. Modern AI processors relied on fabrication processes, memory systems, and specialized components that wouldn't exist for years.
That was an engineering problem worth solving.
If the components I needed weren't available, I'd develop them. If existing manufacturing processes couldn't produce the designs I wanted, I'd investigate how to improve those processes or establish new ones.
After all, every technology had to be invented by someone.
I wasn't expecting to manufacture a twenty-first-century supercomputer in my apartment tomorrow, but I had the knowledge to understand where the technology needed to go. With enough resources and proper research facilities, I could begin closing that gap years ahead of everyone else.
I turned my attention back to the whiteboard and started sketching the first prototype.
For the initial design, I settled on an eight-by-eight systolic array containing sixty-four processing elements. Each element would perform multiply-accumulate operations while passing data to its neighboring units, allowing the processor to reuse information without constantly retrieving it from memory.
The architecture had existed for decades, although its potential was far from exhausted.
At a clock speed of 100 megahertz, the array could theoretically perform 6.4 billion multiply-accumulate operations every second.
That was a reasonable starting point for an FPGA prototype, but I wasn't particularly impressed by the number. NVIDIA's contemporary graphics hardware could already deliver considerably higher peak performance in certain workloads.
My goal was to develop an architecture that could perform specialized calculations more efficiently while consuming less power and moving less unnecessary data.
And sixty-four processing elements were merely the beginning.
I could eventually expand the architecture to contain thousands of processing units, develop custom memory controllers, and connect multiple accelerator cards to work together. More importantly, I could design the hardware and software as a single integrated system rather than trying to adapt existing equipment to my requirements.
I spent the next hour working through the architecture, occasionally returning to my computer to check calculations.
The biggest concern was keeping the processing elements supplied with data. There was little point in building a powerful processor if it spent half its time waiting for information to arrive.
After considering several configurations, I added local memory buffers and a controller to manage data transfers. I'd use a PCI Express interface to connect the accelerator to its host computer and concentrate on fixed-point calculations for the earliest prototype.
Higher-precision floating-point operations would come later, particularly for scientific simulations that required more accurate results.
I made several changes to the memory layout, discarded two inefficient configurations, and eventually settled on a preliminary design that looked promising.
"Now we're getting somewhere."
I examined the drawing before turning to the software side of the project.
The accelerator wouldn't be particularly useful without proper drivers and mathematical libraries. Researchers needed to be able to send calculations to the hardware without worrying about how every processing element worked.
Fortunately, I already knew the advantages and disadvantages of several programming approaches that would become popular in the future. I could take the best ideas, improve the parts that needed improvement, and build a development environment specifically for my hardware.
The more I worked on the design, the more possibilities began presenting themselves.
My biological life-support project was only one potential application. A powerful mathematical accelerator could be useful for medical imaging, aerospace engineering, astronomy, computational chemistry, and countless other scientific fields.
Artificial intelligence was another enormous opportunity.
I knew how important machine learning would eventually become, and I knew that demand for specialized AI hardware would explode once the technology began attracting serious commercial investment.
If I could establish myself in that industry early enough, I wouldn't have to compete with NVIDIA or Intel on their terms. I could develop an entirely new product line and potentially become one of the companies shaping the future of computing.
And there was no reason to stop at accelerator cards.
Once I understood the manufacturing process and established the necessary infrastructure, I could expand into custom processors, high-performance computing systems, and eventually entire computing architectures designed around technologies that the rest of the industry hadn't even considered yet.
I could already imagine establishing my own semiconductor company, assembling a team of engineers, developing new fabrication techniques, and competing with companies like Intel and NVIDIA. With my knowledge of future technologies, I could potentially get years ahead of the competition.
"This could turn into a pretty serious business," I muttered, looking at the PMA heading on the whiteboard.
I returned to my desk and started preparing a list of equipment and components for the first development stage. I'd need FPGA development boards, testing instruments, high-quality power supplies, specialized software, and a dedicated workstation for hardware design.
As the list grew, however, I began to realize something.
My apartment wasn't going to work.
Between my computers, music equipment, books, and everything else I'd accumulated, I was already running short on space. Building and testing specialized hardware required proper workbenches, electrical equipment, cooling, ventilation, and enough room to expand.
And if I intended to pursue semiconductor research seriously, I'd eventually need access to fabrication facilities and much more sophisticated equipment.
I could purchase or lease a suitable property somewhere in Pasadena, establish a private research laboratory, and gradually expand it as my projects became more ambitious. I certainly had enough money to get started, and finding qualified engineers wouldn't be particularly difficult with Caltech nearby.
Actually, why stop at a small laboratory? I could establish a proper research facility with dedicated departments for electronics, computing, materials science, and prototype development. Eventually, I could even build my own semiconductor fabrication plant.
I was already considering possible locations when I stopped writing and stared at the increasingly ridiculous list in my notebook.
"Wait a minute. What the hell am I doing?"
I leaned back in my chair and looked between the notebook and the whiteboard.
I'd started this afternoon looking for a way to speed up calculations for my biological life-support project. Somehow, a few hours later, I was planning an entire semiconductor company, complete with research facilities and manufacturing plants.
"Jesus Christ, Karl. Sometimes you really need to rein yourself in," I muttered, laughing at myself. "I start thinking about a better calculator, and somehow I end up planning a multibillion-dollar technology empire."
Then again, this wasn't exactly unusual for me. Whenever something caught my interest, my mind had a tendency to run wild with possibilities. One idea would lead to another, and before I knew it, I'd be planning projects that went far beyond whatever I'd originally intended.
It wasn't necessarily a bad thing. Some of my best ideas came from following those thoughts wherever they led.
Still, I had to remember why I'd started working on the accelerator in the first place.
The PMA was supposed to help me with my biological life-support research. Developing it into a commercial product could be profitable, certainly, but running a semiconductor company had never been part of my original plan.
And frankly, I had little interest in spending my days attending board meetings, negotiating with investors, dealing with corporate lawyers, or worrying about quarterly earnings.
"Yeah, that's definitely not how I want to spend my time."
I could hire people to handle those responsibilities, of course. But establishing a company of that scale would still demand a considerable amount of attention, particularly during its early years.
Then a much simpler option occurred to me.
Caltech.
I was living in Pasadena, practically next door to one of the world's leading scientific research institutions. Caltech already had laboratories, experienced researchers, specialized equipment, and connections with technology companies.
With my qualifications, securing a research appointment shouldn't be particularly difficult, especially if I could demonstrate the potential of my proposed accelerator.
I could develop the PMA using the university's research infrastructure and let its technology transfer office handle potential commercialization. Naturally, I'd need to negotiate the intellectual property arrangements before bringing my existing designs into the university.
As long as I retained the necessary rights to use the technology for my own projects and received a reasonable share of the licensing royalties, I'd be satisfied.
I'd probably earn considerably less than I would by establishing my own company, but that hardly mattered. My music career was already bringing in substantial income, and money wasn't exactly my biggest concern anymore.
Besides, if I ever decided to build a semiconductor empire, I could always revisit the idea later. The plans weren't going anywhere.
For now, I wanted to concentrate on my research.
"All right, Caltech it is," I said, crossing out the property requirements I'd written earlier. "They can worry about selling the damn things. I'll worry about making them work."
With that decided, I opened a fresh page in my notebook and started outlining a research proposal. I'd need to identify the most suitable departments, determine what laboratory equipment was available, and find the right people to approach about a research appointment.
I was still working on the proposal when someone knocked on the front door.
I barely registered the sound, continuing to write down a few ideas. A moment later, another knock interrupted my thoughts.
"Hey, Karl! Sweetie, are you in there?"
Penny's voice finally caught my attention, and I looked up from my notebook.
I glanced toward my computer monitor to check the time.
5:57 PM.
"What the hell? It's almost six already? I missed lunch."
I'd been so absorbed in designing the accelerator and considering everything I could eventually do with the technology that I'd completely lost track of the afternoon.
I quickly put down my pen and hurried toward the front door, realizing that Penny had probably been knocking for a couple of minutes already. When I opened it, I found her standing in the hallway with her purse over one shoulder and a white Cheesecake Factory takeaway box in her hand.
"Hey! Sorry I took so long," I said, stepping aside to let her in. "I got caught up working on a project and completely lost track of time. I didn't even realize it was almost six until I heard you calling."
Penny looked at me, then glanced past my shoulder toward the living room. "Yeah, I figured you were in the middle of some mad-scientist thing. I knocked like five times, and I was about two seconds away from checking to see if you'd passed out on your keyboard."
"Honestly, not far off. I sat down this afternoon just trying to speed up some math for my life-support project, and somehow I spiraled into designing a whole new kind of computer chip."
Before I could explain further, my stomach let out a rather impressive growl.
Penny looked down at my stomach, then back up at me with an amused smile. "Oh, my God! Please tell me you remembered to eat actual food today."
"Technically, I remembered about thirty seconds before I opened the door," I admitted, rubbing my stomach. "In my defense, I had a pretty massive breakfast."
"Karl!" She laughed and shook her head. "You seriously locked yourself in here doing nerdy stuff all day and forgot how to feed yourself? You're officially worse than the guys across the hall. What am I gonna do with you?"
"Well, feeding me would be an excellent starting point," I suggested, eyeing the takeaway box in her hand. "Speaking of which, is that what I think it is?"
Penny lifted the box with a pleased smile. "Yep! Somebody canceled a party order right before my shift ended, so my manager let us divvy up the extras. I grabbed you the strawberry, since you practically inhaled the last slice I brought over."
"Have I told you lately that dating a woman with an inside connection at the Cheesecake Factory is the best decision I've ever made?"
"Yeah, well, just remember that when I make you come watch one of my plays," she replied, walking past me into the apartment.
"Deal. Although, if your next role involves dressing up as a radioactive badger again, I'm sitting in the back row so I can laugh in peace."
"Ha, ha. Very funny. FYI, it was a mutant badger."
Penny headed toward the kitchen while I closed the door behind her. She placed the takeaway box on the island, then set her purse down on one of the stools. I followed her inside, already thinking about getting something to eat before my stomach started making another embarrassing announcement.
Unfortunately, my phone chose that exact moment to ring.
I glanced toward the living room, where I'd left it beside my notebook, and went over to pick it up.
Leonard Hofstadter.
"Hey, Leonard. What's up?" I answered, returning toward the kitchen.
"Hey, Karl. We're putting together the dinner order right now, and I wanted to check what you wanted. You're still coming over, right?"
"Yeah, wouldn't miss it. Wait, what night is it?"
"Pizza. It's Thursday, remember?"
"Right, Thursday. Clearly I've lost all concept of the calendar today," I said. "Let me guess, Sheldon's already hovering over you with the menu?"
"Pretty much. He's getting his usual sausage, mushrooms, and light olives, Howard's demanding extra cheese, and Raj wants green peppers and onions. I figured I'd call you before it turns into a forty-minute constitutional convention over topping distribution."
I glanced toward Penny, who had opened the cheesecake box and was examining the slices inside. She looked up when she heard me mention pizza and gave me an interested look.
"Smart move. Hey, Penny just got off work and she's over here. Cool if I bring her along?"
"Oh! Uh, yeah. Yeah, sure, totally fine," Leonard replied, his voice pitching up slightly. "What does she want on hers?"
I covered the phone's microphone and turned toward her. "Hey, Penny, the guys are ordering pizza. You wanna join us for dinner?"
"Ooh, yes, I'm starving," she replied, popping the lid back on the cheesecake box. "Just get me a pepperoni. And none of Sheldon's weird mushroom stuff, please."
I laughed and uncovered the microphone. "All right, Leonard. Put us down for two large double pepperoni pizzas with extra cheese, and I'll cover our half when we get over there."