Returning to Dominate The World With My Knowledge System
Chapter 64: Two More To Go

Chapter 64: Chapter 64: Two More To Go

Tyler stood in front of the lab table, his hands in his pockets and a satisfied a smile on his face.

On the table before him, ten non-volatile DRAM-based NVMe units were lined up.

They looked sleek, compact, but they are unimaginably powerful.

Each of them a storage miracle, each of them the kind of advancement that shouldn’t exist for at least another three to four decades.

While he had been deep in his office all day finalizing the Heimdall motherboard schematic, his team had quietly pushed the boundary again.

Six more of the advanced storage modules had been fabricated. Coupled with the four completed the day before, that brought the total to ten—just two shy of the twelve he needed.

Tyler’s gaze swept across the modules, then shifted to the team still packing up their equipment and running final diagnostics.

None of them were resting, despite the hour. They didn’t have to be told this was history. They could feel it in their bones.

The room held a quiet charge, as even the air felt denser.

Someone unfamiliar with what Tyler was building might assume this was overkill.

That ten Valkyrie-X GPU chips, 64 terabytes of volatile memory, and 192 terabytes of non-volatile DRAM storage were excessive or even insane.

But they’d be wrong. Dead wrong.

In 2025, the training of large frontier AI models—like GPT-4 and beyond—required clusters of hundreds to thousands of high-end GPUs.

These were chips like NVIDIA A100s or H100s, which hovered around 300–400 TFLOPS each.

Just training GPT-3 required several thousand PFLOPS cumulatively across months of runtime.

Tyler’s Valkyrie-X chip, by contrast, could push over 700 TFLOPS each.

Ten of them gave him an output of approximately 7,000 TFLOPS, or 7 PFLOPS of raw compute.

And not just general compute. These chips were specialized, as they were built from his own [Computational Mathematics] knowledge—to handle inference, training, data parallelization, attention span optimization, and real-time sensor fusion. These weren’t your typical GPUs. They were each a multi-purpose AI engine.

With ten of them, he could segment training across specialized model blocks, optimize weight propagation, and assign isolated task groups to each chip without causing congestion or thermal choke points.

In short: Ten chips wasn’t extravagance. It was the absolute minimum for stable, high-speed AGI model training with internal autonomy.

As for why the 64TB of volatile DRAM wasn’t just justified but absolutely necessary?

A few years from now, supercomputers like the Fugaku, Frontier, and El Capitan would use 100TB to multiple petabytes of RAM.

Why?

Because the depth of modern AI models—especially those aiming toward AGI—demands entire training datasets, instruction sequences, and contextual memory to remain in-memory at all times.

With that much memory, they won’t experience any bottlenecks, swapping or delay.

Tyler’s 64TB memory bank wasn’t for show. It was to load entire datasets for real-time training, enable continuous long-context analysis, provide persistent memory-state for simulated consciousness and optimize latency for micro-adjustments in task response

So, no matter how advanced his GPUs were, they’d choke on read latency if the DRAM wasn’t equally capable.

64TB volatile memory meant the system could think without pausing.

For the 192TB of non-volatile DRAM storage, most people think of storage as just a warehouse for files.

But Tyler needed much more than a warehouse. He needed instant access to prior model checkpoints, logging of every interaction, version cloning for rollback events, low-latency, high-bandwidth inferencing output and multi-instance autonomy to allow AI replication across subsystems

By creating the non-volatile DRAM based NVMe, Tyler had leapfrogged the world by at least four decades.

Tyler turned back to the team and gave a single nod, the corners of his mouth lifting slightly.

"Good work."

They didn’t cheer or clap. They just smiled back, tired but proud. Some gave brief thumbs-ups.

One guy gave a small bow.

Another chuckled softly and said, "Boss, you keep thanking us like we’re doing you a favor."

Tyler raised a brow.

"You are," he replied.

"Not really," the man laughed again. "You’re paying us well to watch miracles happen."

This was true as most of these men and women had walked into this job expecting high-level assembly work.

The team wanted to tell Tyler that they could continue and finish another of the DRAM or even the remaining two before they leave.

Still, as much as they wanted to finish the last two before closing, no one moved. They remembered yesterday and how Tyler had refused to allow overtime.

And sure enough, as he glanced at the clock and then looked back at them, he said, "We’re done for the day. Get some rest. I need you fresh."

They hesitated, but just for a second, before nodding.

A few began cleaning their stations. A couple lingered a bit longer, eyes drifting to the remaining substrates they could’ve worked on.

But Tyler’s decision was final.

He couldn’t afford errors at this level. The fabrication of these components was surgical. Any minor defect in a DRAM substrate or logic layer could result in data corruption, electrical arc, or catastrophic thermal failure during runtime.

Better to go slow and perfect than fast and ruin everything.

The ride back to the hotel was quiet, but not somber.

Some texted family. Others listened to music or scrolled through the news, though nothing they read came close to what they had lived through in the lab.

As for Tyler, his mind was already at work again.

Tomorrow, the last two NVMe DRAM modules would be completed. And once they were, he could begin with Heimdall creation.

That would be the true test, as it would be not just of hardware design, but of every optimization algorithm, thermal logic, and predictive firmware model he’d worked out.

He would be merging the Valkyrie-X GPUs, 64TB of volatile DRAM, and 192TB of advanced storage into a unified architecture designed from scratch.

And it all had to work.

One thermal imbalance, one flawed voltage rail, one bad controller... and the system would fry itself within seconds of power-up.

Still... Tyler was all smiles. He was confidently looking forward to what the future holds for him and the challenges it might bring.