Technically, began was not quite the right word, seeing that I had played with auto surgeon designs on and off over the last year. As a result, I had dozens of concepts already in my files. Some of them were actually quite advanced. Sadly nearly all of them predated me building a NADA and were obsolete before they were ever built for that reason alone. On the other hand, I would not have to start from scratch, and most important I had the bare bones of the software available. Sure, what I had I would have to adapt to the new platform but at least I wouldn’t have to pour another 5 virtual years into it.

That of course led to the most important question, where to start. It took only a short brainstorm for me to decide that I would start with the computer-architecture. Sure, I could use slightly adapted commercial components as I had with the cluster and Precious, but that would be a horrid waste.

Don’t get me wrong, both the Tesseract VI I based the cluster on as well as the Cirrium k8 that beats inside Precious were state of the art. Created in the three-nanometer process that marked the endpoint of what is possible with lithography, they were the best one could get for their respective mission.

But neither had Q-links integrated, and neither took the possibility of being constructed atom by atom into their design. In other words, both designs were essentially two-dimensional, badly cooled, clunky designs with snail-speed pipelines compared to what I could create. That, of course, did not mean that I would throw out the basic architecture completely. I viewed the processor design as akin to a building block system.

The massive multi-parallel design of the Tesseract was optimized for supercomputers, and what I needed was more along with a few more versatile cores instead of the swarm of small, fast optimized number crunchers. With that decision made I started out with a Cirrium core.

Immediately I replaced the pipelines with Q-links, separating the distinct elements of the core physically. Then I stacked them into the third dimension with cooling layers in between.

I was debating if I should jump from 256 bit to 512 bit architecture, but decided that would cost more in adapting than it was worth.

Still, I quadrupled the bandwidth of the data pipelines, followed by increasing the number of instruction pipelines to eight.

Meanwhile, my simulation showed that in a 414 picometer architecture with integrated cooling layers my new processor would be able to run in excess of 40 gigahertz without getting over 40°C.

I looked at the results for a few minutes, trying to find words for them. In 250 years nobody had cracked the 10GHz barrier without massive liquid nitrogen cooling, and only liquid helium allowed even coming near 20GHz.

Instead of the floating-point operators of the Cirrium, I adapted the basic Tesseract core into my design concept and replaced the 64 Cirrium operators with 512 Tesseract operators. That would give this new processor nearly 13% of the parallel processing capability of the server processor without sacrificing the versatility of the desktop processor. At a hunch, I added a dozen fuzzy logic cores to make the work of a possible VI more smoothly.

When I came to the cache I was thinking hard.

Level one cache is generally rather small to integrate it better into the processor, and such shortening the response time, but my Q-link took care of that.

A side effect of that was that the data could be read and written exceptionally fast though.

For more data level two and three caches were added. In the end, I decided to experiment and integrated all three levels of cache into one, 256 GB-sized module connected by another quadruple bandwidth Q-link. The result was, if my simulations were to be trusted, rather fast.

This new core I had designed was somewhere between a thousand and two thousand times faster than the Cirrium core it was loosely based on.

That came not without a price of course, as it was also five times bigger and used nearly twice the power. But honestly, that cost was insignificant compared to the jump in processing power.

To make things easier with the conversion of the programs I had already written I combined the instruction sets of the Tesseract and the Cirrium lines.

As usual, I combined 16 of my cores into a single processor that was light years ahead of anything available anywhere else. With a sad feeling, I knew then and there that I would have to replace Precious soon, and that the replacement of the cluster with something closer to my new base would have to wait until I had designed a similar substitution for the Tesseract.

As the auto surgeon did not need any fancy graphics I took much less care with designing the GPU. Sure, I put the standard building blocks into the same 414pm process and connected the distinct operators and cores by Q-links while integrating the cooling layers, but I left the basic design as I found it. That netted me only roughly 300% performance improvement.

When I called up the motherboard design my alert triggered. Apparently, I had spent nearly 25 hours on designing the processors, virtual of course, and the bots had finished with the pylons.

It took me nearly two minutes to extricate from cyberspace. Theoretically, I should have been able to get out in an instant, but for reasons nobody could explain yet, it did not work that way. Whenever the surfacing was too fast it scrambled something in the Jack's brain. We only knew two of the factors determining the time needed for a safe return to reality. One was loosely based on native intelligence of the Jack.

Basically, the higher the IQ, the faster a surfacing the Jack could stand. While the first factor favored me, the second more than equalized any gain I had.

The second identified factor was the compression. The higher the compression the longer it takes. An average intelligent Jack with a 25:1 compression would need around 20 seconds.

At 30:1 it would have already increased to 45 seconds. All that real-time, unfortunately.

At 57:1 I should need around twelve minutes to surface, but my Pure enhanced brain was better at resisting the scrambling than the average Mongrel brain.

I came back into reality to Wallace still reading and the bots standing at attention waiting for new orders. Unplugging the OPB cable was way easier to accomplish left-handed than plugging it in, and I inspected the tank and the pylons, most of it from the wheelchair, but a few connections I had to stand to reach. The bruised hip would be painful for every normal human. For me, it was considerably worse.

The average Pure felt pain between 20 and 30 percent stronger than the average Mongrel. Unfortunately, I was many things, but average was not one of them.

My Delta was 133, with 24 being the average Pure. In other words, I felt anything around three times as strong as a normal human, including pain. And that was before Frankel had the brilliant idea to enhance me further.

I did not know how much this asshole had increased my pain sensitivity but from what I felt I guessed that he at least doubled it. And thanks to the sanctimonious prick that was lead designer on the immune system upgrade they gifted us Pures with and his teetotaler agenda I could not even take most pain killers.

For all purposes the only option I had was Epzitecan, and that had some unwanted side effects. It was, of course, addictive, and had the tendency to increase pain sensitivity with prolonged use.

It interfered with the neural signals up to convulsions if given at a strong enough dosage.

All that I could somewhat live with. An addiction can be overcome, the increased pain sensitivity would be of no concern over the time my hip ached and the, at the dosage I needed relative small decrease of dexterity was of no concern until I got my arm out of the cast.

It was the last side effect that prohibited it for me now. It had the unfortunate tendency to impact cognitive capabilities. Essentially it made me feel as if I was swimming in jello. Dumb, slow, and anything but happy. Nothing I could allow for the time being. And unlike opiates, I would not even get a nice trip out of it. I could afford to take a dose when I was going to bed. That would only increase my sleeping time to six hours, maybe seven. But for now, I needed my brain working.

Long story short I clenched my teeth and inspected the upper pylons.

Naturally I found a few small problems there, as well as at the lower pylons. Nothing major, just a handful of a tad too loose connections, an insufficient vibration dampener, and one connection switched. Except for the connection, nothing of it would have prevented the NADA from working but would have made a few problems later on. Ordering the bots to correct the issues made short work out of it.

The next step was installing the raw material storage. I had included the absolute minimum in my equipment when I escaped Seattle, so at least I could build the most basic parts. I would have to get, or build, additional containers for the more exotic materials. Thankfully this was a job the bots could do without supervision.

After I sat back down I contemplated diving back into cyberspace, but I decided against it. I was frankly still pretty beat, and the pain in my arm and hip was getting way too distracting. Sure, in cyberspace I would not feel anything of it, but the price was that at least the hip would be considerably worse when I got out.

Add in that I was still not quite over the concussion and the nearly six hours since I woke up in the clinic were nearing the limit I could do that day. I longed for a shower and some painkillers.

Luckily nothing on my to-do list was that time-critical at this moment anymore. Until the cluster cracked Giorgio’s files and completed an in-depth comparison with the probable frame-up, a process that would, depending on the quality of the encryption and the quantity of files, take the whole night or even longer, I could do nothing on that front.

The installation of the material storage would take a couple of days, give or take a few hours, none of the Pulse III would arrive before the next day, and it was irrelevant if I finished the design for the auto-surgeon now or in a week.

After another less than satisfying meal, I skipped the shower as I did not think I would be thankful for standing even that short. Instead, I took my medication. It took not long before I felt my thoughts grinding to a near stop in the Epzitecan created molasses that became my mind, but at the same time, the pain lessened to a degree that made it possible for me to sleep.