[TBBT] Ch: 37 [The Academic Nemesis Reunion] (Patreon)
Content
[AN: Let me know if you find any mistakes.]
--
Dr. Gablehauser looked between us, clearly struggling to understand why two people who'd supposedly arrived for a professional meeting were pointing at each other like children.
"Well, this is certainly an unusual way to begin a research discussion," he remarked, lowering the folder in his hand. "Am I to understand that you two already know each other?"
"Unfortunately, yes," Paige said, dropping her hand and folding her arms. "We survived MIT together. Though I didn't realize Karl had moved on to building hardware."
"I've always liked hardware," I replied, picking up my Pelican case. "And last I checked, you were still torturing university servers with bloated math models."
"My models were perfectly reasonable until you started interfering with them!"
"Right, because cutting your runtime by forty percent was such a crime."
"Enough!" Gablehauser interrupted, holding up a hand. "Whatever disagreements you two had at MIT, I'd appreciate it if you left them outside this conference room. Now, please take your seats."
I pulled out a chair opposite Paige and sat down, while she remained in hers, watching me with obvious suspicion.
Gablehauser opened his folder and explained, "Dr. Peterson, I received your proposal yesterday concerning a specialized mathematical accelerator. As you know, Caltech has been looking into upgrading several of our computational research facilities. I'm currently coordinating with colleagues in Engineering and Applied Science on a cross-departmental computational research initiative. Your preliminary results caught my attention, particularly the potential applications in physics and engineering."
He turned toward Paige before continuing.
"Miss Swanson has also applied for a research position here. Her proposal focuses on numerical algorithms and high-performance scientific computing. Given the considerable overlap between your areas of interest, I thought it would be productive to discuss both applications together."
I glanced toward Paige.
"So you're looking for a job here too?"
"Yes, Karl. Some of us actually like doing research instead of playing pop star," she said, giving me a pointed look.
"Hey, my music pays a lot better than most research grants," I countered.
Gablehauser cleared his throat before either of us could continue.
"Dr. Peterson, perhaps we could begin with your demonstration? I'd like to see whether the actual hardware performs as well as your proposal suggests."
"Of course," I replied, standing and opening my Pelican case. "I've brought everything necessary."
I placed the FPGA development board on the conference table, connected its power supply, and plugged the interface cables into my laptop. Paige leaned forward from her seat opposite mine, studying the hardware with far more interest than she'd shown when I arrived.
Once everything was connected, I launched the demonstration software. Alongside the benchmark interface, I'd included a diagnostic panel displaying memory-transfer timings, processing-element utilization, buffer occupancy, and the current block scheduling configuration. It was useful for identifying performance bottlenecks during testing.
"The prototype uses an eight-by-eight systolic array containing sixty-four processing elements," I explained, turning my laptop toward Gablehauser. Paige leaned toward the laptop to get a better view of the screen. "Each element performs multiply-accumulate operations while passing data to neighboring units. The architecture reduces repeated memory accesses during matrix multiplication."
Gablehauser moved closer to the table.
"And you've implemented the entire processing array on that FPGA?"
"Along with the buffering and control logic, yes. The host computer handles the interface and transfers the data. For this demonstration, I've prepared identical matrix multiplication workloads for the CPU and accelerator."
I selected the first test and started the program.
The laptop processed the calculation on its CPU and displayed the execution time and verification results. Once it finished, I ran the same workload through the PMA.
The accelerator completed the operation considerably faster.
"My earlier benchmarks showed approximately 3.4 times my CPU's performance on this workload," I explained, showing them the results. "I've also included accuracy checks to verify that both implementations produce the same output."
Gablehauser examined the figures while Paige leaned further across the table, studying the performance statistics displayed beside the results.
"Impressive," he admitted. "What happens when you reduce the matrix dimensions?"
"The performance advantage decreases. For sufficiently small matrices, the CPU actually wins because transferring data to the accelerator introduces overhead. The current prototype works best when there's enough computation to keep the array occupied."
I ran another demonstration using smaller matrices, showing precisely where the accelerator began losing its advantage.
Gablehauser then asked questions about power consumption, reliability, and the modifications I'd planned for the second generation. I showed him the preliminary architecture diagrams and explained why additional processing elements, improved memory management, and floating-point support would be important.
By the time I finished, he seemed genuinely interested.
"Well, Dr. Peterson, I must say you've presented a convincing case," he said, returning to his chair. "Your proposed research fits our interest in improving computational capabilities, and I'm prepared to recommend a part-time appointment through our joint research initiative, focused on developing the second-generation accelerator."
That was encouraging news.
"Excellent. As I mentioned in my proposal, my existing intellectual property will remain mine. My attorney has already prepared documentation establishing the boundaries of any collaboration."
"Naturally, our legal and technology transfer departments will need to review those arrangements," Gablehauser replied. "However, I don't anticipate any insurmountable difficulties. Subject to final approval, I'm prepared to support your appointment and the laboratory resources you've requested."
I returned to my chair, ready to thank him, when Paige spoke. "Cute hardware, Karl. Too bad your software's choking it."
I turned toward her.
"Excuse me?"
Paige pointed toward the diagnostic panel on my laptop.
"Look at the utilization numbers. Half your array is sitting idle between stages while your software wastes cycles shuffling data blocks around. Your buffering isn't keeping it fed."
I leaned across the table to examine the figures she'd indicated.
"Those transfers are scheduled to avoid unnecessary memory accesses."
"I can see that. But your block sizes and scheduling strategy aren't particularly efficient. You're spending too much time preparing data between computational stages."
She took a sheet of paper from her folder and began sketching a simplified arrangement. Once she'd finished, she slid the paper across the table toward me.
"If you reorganized the data into larger reusable blocks and adjusted the transfer schedule, you could reduce some of that overhead. Your current implementation is leaving performance on the table."
I traced the arrows she'd drawn between the data buffers, checking her suggested changes against the limitations of my FPGA.
Damn it, she had a point.
The buffering system I'd developed was functional, but I'd concentrated primarily on getting the hardware working reliably. There were opportunities to improve the software.
"That might work," I admitted reluctantly. "Although you'd have to account for the FPGA's limited on-chip memory. Increasing the block size too much could cause other problems."
"Obviously, Karl. That's why we'd determine the optimal size instead of choosing one arbitrarily."
I looked at her drawing again before leaning back in my chair. "You know what, Paige? You're right about the optimization. But there's something you're overlooking."
"Oh? And what would that be?" she asked, putting down her pen.
"I came up with the idea for this accelerator on October twenty-fifth. By December first, I'd designed the architecture, written the software, assembled the prototype, fixed the hardware bugs, and completed the testing. That's thirty-seven days from an idea in my notebook to a working machine."
I gestured toward the FPGA board on the table. "And during those thirty-seven days, I was also juggling label negotiations, press shoots, radio promos, and getting fifty thousand CDs pressed."
Paige listened without interrupting, although I could tell she was preparing something.
"So yeah, I cared more about getting the board working than squeezing out every last percent of efficiency," I continued. "And yes, I'm boasting. I think I've earned the right after building this thing in barely five weeks while managing a music career."
Paige considered that before nodding. "Impressive, Karl. Truly. Still sounds like a fancy excuse for sloppy code, though."
I laughed, shaking my head. "Tell you what, Paige. If you build something better in thirty-seven days, I'll write you a formal apology. Until then, I'm going to enjoy this little achievement of mine."
"Well, perhaps I will," Paige replied, looking entirely too pleased with herself. "And when I do, I expect that apology in writing."
"I'll even frame it for you."
"Excellent. This is precisely the kind of exchange I was hoping to encourage," Gablehauser said, interrupting before we could start another argument.
Both of us turned toward him.
"Miss Swanson, the hiring committee signed off on your packet last week after reviewing your numerical analysis work, and I'm prepared to finalize your full-time appointment. Watching you dissect Dr. Peterson's software in under two minutes only confirms their judgment."
Paige sat a little straighter.
"Thank you, Dr. Gablehauser. I'd be delighted to accept, assuming the terms are satisfactory."
"Wonderful. There's just one small administrative matter we need to discuss."
Something about his tone immediately made me suspicious.
Gablehauser opened another document.
"Unfortunately, suitable laboratory space is somewhat limited at present. I've discussed the available facilities with our research coordinator, and there's currently only one laboratory equipped for the kind of work you're both proposing. It has the necessary electrical infrastructure, workstations, and testing equipment. Given the overlap between your projects, the arrangement we've been authorized to offer would require you both to share it."
I stared at him.
Paige stared at him.
Then we turned toward each other.
"Share a laboratory with her?" I demanded.
"Share a laboratory with him?" Paige exclaimed at practically the same moment.
Gablehauser looked thoroughly unimpressed by our synchronized objections.
"Yes. You'll have separate workstations and designated equipment areas. Your research appointments and intellectual-property agreements will remain independent, but you'll be expected to coordinate access to shared facilities."
"Dr. Gablehauser, with all due respect, I dealt with her for years at MIT," I protested. "Do you have any idea how many fights we had over lab space?"
Paige turned toward me.
"Because you couldn't keep your hands off my stuff! You moved my entire desk once because you didn't like my cable management!"
"Your cables were practically a fire hazard!"
"They were perfectly organized until you touched them!"
Gablehauser closed his folder with a firm thump.
"Then I suggest you both learn to communicate like the highly educated professionals your résumés claim you are. The laboratory is available under those conditions. Whether you accept the arrangement is entirely your decision."
I looked toward Paige, who appeared equally unhappy about the prospect.
Unfortunately, turning down a fully equipped laboratory simply because an old academic rival would be working there seemed rather childish.
And I really wanted access to Caltech's facilities.
After thinking it over, I extended my hand across the table. "Fine. I can share a lab, as long as you don't touch my gear without asking."
Paige shook my hand. "Agreed. And you're keeping your hands off my workstation."
"Wonderful," Gablehauser said, looking relieved. "I'll have the department administrator prepare the necessary paperwork."
As Paige released my hand, I couldn't help thinking that my peaceful return to scientific research had just become considerably more complicated.
...
We left Gablehauser's office and headed down the corridor to check out our new lab, my Pelican case bumping against my leg as we walked. Neither of us said much and she was probably already mentally dividing the room in half.
Paige slowed to a stop in the middle of the hallway.
I took another step before stopping beside her.
Twenty feet ahead of us, Sheldon had frozen mid-stride, a stack of manila folders clutched to his chest. He blinked once, tilted his head, and squinted as if trying to solve an equation that refused to balance.
"Paige?" he said, his voice pitched a little higher than usual.
A slow, genuine smile spread across Paige's face, the kind that had nothing to do with politeness and everything to do with amusement. "Hey, Sheldon. Long time no see."
--
AN: Yeah, I know what you're gonna say about the ending. I'm a cruel guy. See you in next chapter.