After Jacky was happily, as I was assuming, playing with the new VR mode for Envision, it was substantially harder to get additional test subjects.
In the end, I resorted to calling Ben and Michael, to have the garbage cans distributed among their office pool. I simply lacked the acquaintances to really organize such a test by myself.
Naturally, the special garbage can remained with me. Warden was, well, not busy, as it did not tax her resources in the slightest, but at least she was working on testing as much legacy software as we could. Not in VR, as that had to be done by a person sadly, but that they at least ran with the VR layer.
I dreaded having to test them in VR any further than I already had. The sheer quantity of applications to test alone was… daunting. After some navel gazing and soul searching, I decided that the tests I already had done would be enough.
It was just important that the apps still worked at all with the VR layer. Whoever was insisting on using them could use the mouse, keyboard, and holo-controls as far as I was concerned. They had lived without VR so far.
If it worked with their specific flavor of obsolete software, good for them. If not, they lost nothing.
Testing the legacy hardware was a bit more complicated, but I tasked a bot to change the configuration.
Sitting back and pondering what to do next, I decided to do what I thought Michael wanted.
Work on our CPU lineup. It was essentially mostly busy work. I mean, except for the successor of the Regulon, I already had done the hard work. With the already designed processors at the high end of the lineup, I only needed to par the design down to more reasonable levels.
I mean, while the Hyperion that I used in Glory, which I renamed Hyperion 12 1990G, had 128 256-lane Universal Equipment Bus controllers, 16 TB of unified cache, and 1024 cores with a maximum frequency of 44 GHz, more mainstream models would cut down on all of that.
The entry model, which I called Hyperion 3 1200 had to do with 32 UBE controllers, 4 TB cache, and 256 cores with 31 GHz maximum frequency. Yes, it only needed 27% of the power and generated only 18% of the waste heat, but it also had only 22% of the performance of the top unit.
All in all, the Hyperion lineup now consisted of two Hyperion 3, the aforementioned 1200 and the 1250, which boosted up to 32 GHz, two Hyperion 5, called 1400 and 1450, two Hyperion 7, called 1600 and 1650, two Hyperion 9, called 1700 and 1750, two Hyperion 12, called 1800 and 1900, and last but definitely not least, the Hyperion 12 1990G.
The 1900 had roughly 76% of the performance of the 1990G. Naturally, I envisioned the 1990G costing around 2.5 times as much as the 1900.
I did something similar to the GPU, which I now called Theia. Again, I placed the original design I had built into Glory at the very top. I created 9 steps, starting at Theia 10 and ending at Theia 90 with the latter sporting awe-inspiring 138000 cores. Of course, those were not the full-fledged universal cores that were used in CPUs. No, we were talking about significantly simpler specialized number crunchers. More important though was the bandwidth of 2.6 EB/s that fed the 768 TB of QDR.
Thanks to this downright insane bandwidth the Theia beat the previous top GPU by nearly 600000%. Or to put in more understandable words, a Theia 90 was nearly six thousand times more powerful than the Rigel 32 KRX GPU that I had used for Precious. Sadly, much of this performance increase was eaten up by other, much more pedestrian components., resulting in a Hyperion 12 1990G with a Theia 90 ‘only’ being around 120 times more powerful than a Cirrus 8K with a Rigel 32.
At some point, I would have to redesign the UEB standard to get more of the possible performance to the end user.
The Chimaera was at once simpler, and harder.
It was simpler insofar as I did not need as many steps, as it was more or less designed for embedded systems.
It was harder because I had to create a variant without the fuzzy logic core. Of course, compared to the Hyperion or the Grendel, the Chimaera was a very basic design, almost pedestrian.
That were three done, leaving the big parts. I had to design a big server processor that was, as Michael had termed it more ‘granular’.
Unlike the Hyperion lineup, I could not just take the Grendel and cut it down to size. The Grendel was a marvel of engineering if I may say so myself, but it was also extremely high-powered in all aspects.
The 32768 cores it had were served by no less than 64 Petabytes of cache. They had 8192 256-lane UEB controllers, of which at least 512 were reserved to address up to 2 Zettabytes of quad data rate RAM.
And yes, I know that this was seriously overkilling it, but please, I made it work.
My problem now was to cut it down to something more… well, ok, reasonable. But believe me, that was hard work.
I decided early on to keep the two gigabytes of cache per core and the ratio of one UEB controller per four cores.
With the reduced number of cores, I was also able to significantly cut back on cooling. Seriously, I reduced the number of cores and everything else with it by 94%. Down to 2048 cores in the top version. That gave me still colossal 4 PB of cache and 512 UEB controllers. Of course, with the much-reduced cooling, I limited this new processor to no more than 45 GHz. As I had designed the Grendel to be more or less directly cooled by liquid nitrogen, I had pushed it up to 52 GHz.
From there, I created another six tiers with successive lower performance. And with that, I was done with the server processor that I named Chronos.
Of course, it is relatively quick to tell about what I had done, but all in all, I had spent four days creating the new processors.
And had not even begun with the new mobile SoC.
Yes, in absolute time, that was lightning fast. Even if I had practically all of the hard work already done and was now just paring down the processors for cheaper products. But four days in real life are quite a bit more in VR. Even if I did not even come close to fully utilizing the theoretical maximum of 960 days, or two years and seven months. No, I kept it relatively slow with no more than two months per day. Or six hours in real-time.
And even with that in mind, it was very fast.
Frankly, I had enough of processor design for a bit. That was why I pushed the SoC design further back. Especially as the SoC was the one processor I would have to start from scratch.
In the meantime, the first results for the Envision VR had come back. Nothing too spectacular here. A handful of things in need to be optimized, a couple of bugs and glitches, or in other words, for a software project a spectacular success.
And I have to say, finding out that Warden already had solved the problems in question was a nice surprise.
That left me with a bit of a conundrum. What to do now? Sure, I could increase my studies of gravitics.
But I was slowly but inevitably running out of subjects to study. It would not be this day, or the next, or even the next year, but it would happen that I had taken every single course USW offered. And I was determined to delay that moment as long as possible.
I was not in the mood for music either, and I was by now no longer the only cook in the house, and it was not my day to be in the kitchen.
With a sigh and a shrug, I instead began to design the chipsets for the CPUs I had created. New processors would be nice, shiny, and expensive paperweights without the mainboards needed to make them into computers.
And historically it was a relatively bad idea to keep everything in computer design and production in one house.
We would have to deliver the chipsets, sockets, and design specs to OEMs all over the world. Fortunately, that was one area of computer production that had not been virtually wiped out by the universal chip fabber.
Oh, the OEMs had suffered, but there was enough demand for updated firmware that a handful of them survived. Not all of them were the better ones, sadly.
I personally found it astounding how at the age of electronic fabbers that produced components and chips cheaply and reliably, some manufacturers still cheaped out on small but essential components like power regulators or capacitors.
Seriously, they save a few cents on each mainboard and reduce the reliability, and such worth, of their product by half or even more.
But apparently, there was a need for low price electronics. I’ve never used an OEM-manufactured board in any system I built. Even before I had an industrial fabber I paid the maybe 10% more to have electronics manufactured to my specifications.
But I was also aware that if Enki was to produce the next generation of computer technology, we couldn’t keep the production in-house.
Oh, sure, at least in the beginning we would have to provide the chipsets, as I categorically refused to not use the 414 pm process. But when we would sell NADAs we could license those as well.
I kept the number of chipsets at three. The differences were mostly in how many of the UEB controllers I used to address the memory, or in other words, how much memory could be used on the board.
To be honest, I doubted that anybody but a hardcore gamer, or a Jack, would ever use all of the lanes that a 1990G had.
For an office garbage can, the 1200 would probably be overkill in the number of lanes. But that was nothing new. The current Cirrium 1d, widely considered the most basic desktop processor was in most cases three times the processor that an office drone needed.
That would change with the VR overlay. But even with VR, I doubted that we would sell all that many CPUs for garbage cans. At least in the beginning. Historically, every time computer power increased, OS quickly increased the requirements. It was as if the OS creators only waited for more power to be available to cram more mostly unneeded functions into their operating systems.
But that was not my problem. If Ralcon decided to beef up Envision so that it needed a Hyperion to run, who was I to complain?
When I was nearly done with the high-end consumer board design, I got a call from Michael.
It took me a moment to surface to his compression, but by now he was well aware that I was often not reachable at a moment's notice.
“Hey Michael. What can I do for you?”
“Well, you could loan me your Benz. And maybe come with me to Boston.”
Ok, the Benz was not a big problem. Over the last few weeks, he had used it a few times to travel to some corp or another for negotiations.
So much so that I was considering buying another one for Enki.
It was the other part of the request that mystified me. What would he be doing in Boston, and why did he think I should come with him?
“Boston? What do you want in Boston?”
“Well, there is this tiny insignificant little company in Boston called Enertech that I want to negotiate with.”
“Oh. Yes, that makes sense. But why do you want me to come with you?”
He sighed.
“Because now we begin to play in the big leagues, and I want somebody at my side who can answer the hard questions.”
I had to blink, even in VR.
“Just to make it clear, you want me to sit by your side and somehow convince the old, privileged executives from one of the triple-A corporations that this tiny girl that looks as if she just now entered high school is some sort of expert?
You’ve met me, right? Do you really think I can pull that off? Or will it more likely result in them not taking us seriously?”
“You are not wrong. I mean, you are our CTO, as well as the majority owner. Fuck, Enki exists primarily to market your inventions. But you are right, you don’t look like the supersmart, experienced mad scientist.”
He paused for a moment before he continued:
“But I still think I might need your advice. The other corps were easy. Here are the jacks, have fun with them, and if you agree to play nice with us, we give you a rebate. Or we will help you in incorporating them into your business. For all it counts, that was the kid's table.
Enertech and Ralcon are… different. Enertech is not interested in our cyberware. At least not directly. We have to give them something more tangible than a few jacks and a rebate. And yes, we have the new nano-fab design for them. But I am not certain that I can sell them the idea.”
I pondered what he had said for a moment.
“I am sure you can do it. It should be easy. Especially as a couple of days ago, Burgmeister, Xiao Ping, and Dalgon managed to have the shared patent that Enertech and Ralcon had on nanotech declared to be expired.”
I could hear the surprise in his voice.
“Really? Fuck, then the new design is worth much less to them.”
I had to chuckle softly.
“On the contrary. So far they have been able to get away with using their so-called 10th gen franken-bots to make quite a bit of money. Now they can already see this business completely getting away from them.
The real 10th gen that have sprung up are cheaper and better. But so far they were ‘illegal’. Now they are legal and worse, there will be legal 12th gen bots flooding the market.
We can assume that they will use the open-source designs to keep a tiny market share, but seriously, it costs around $20k to make one of the 8th-gen fabs we offer them the plans for. After that, each batch of 12th gen nano-bots costs them a couple of bucks.
If the markets all over the world are an indication, the other three triple-As will mostly set the price at $1000 for a batch of 12th-gen bots.
Enertech still has the connections in the market here. If they manage to get a 12th-gen lineup fast enough, they will still maintain the lion's share of the market in the US.
Each nano-fab they build will amortize itself within two or three days, and after that, it is almost pure profit.”
His pensive answer came after a few seconds:
“So, they now can either find a viable 12th-gen bot lineup, together with the 8th-gen fab design, or they can watch their market share evaporate pretty quickly. And every day counts for them.
So the schematics I will offer them will mean a few billion dollars for them?”
“Yep. They should jump on the chance. Of course, you will have to make sure that there are some safeguards for us, but as the only thing we want in payment is their neutrality, and we barely scratch their business, it should be pretty easy.
We can offer them a deal that we sell them 12th-gen designs for a discount in the future if they need them. But you should only mention that when the negotiations begin to stall.”
“Yeah, I get it. What about competition? I mean, if we sell them the schematics, it is pretty obvious that we can sell the bots to the public as well.”
“They have the market penetration we would have to create first. Don’t get me wrong, selling specialized schematics is a very lucrative business and something we will probably keep doing. But for general use nanobots? It is honestly not worth the pushback we would get.
We would have to compete against Enertech, Burgmeister, Xiao Ping, and Dalgon, and that is if we don’t provide Ralcon with schematics. Could we make a business out of it? Probably. But let’s be honest, we have the cyberware, where we have an exclusive market, we have the new cloning tech, and the meat vats that are essentially ready for production at any time. As soon as we think we are ready, we will have the Q-links and the NADAs. Following that, we have a completely new lineup of CPUs and GPUs.
It will take a few years before we have enough capacity to satisfy those markets. Why make enemies by going into a market with no less than four triple-As fighting for the lion’s share, just to get a few scraps? Taking away from our ability to utilize our technology to get other, even more important markets exclusively?
The Q-links will dwarf nanobots as a business in a few months, and while the margins might be smaller, the volume will be staggering. And I don’t think there will be a workshop, factory, or any other fabricating business that will work without a NADA.”
“You are right. I still feel like the little orphan kid begging for scraps on the grownup table. So, if they ask if we intend to go into the nano-bot business I can say that we don’t? Maybe even sign a contract?”
I had to think for a moment about that.
“No, that is not quite right. We will go into the business of very specialized nano-bots and nanofabs.
We make cyberware CRS-free with nanotech after all. And I don’t think it is viable to keep the CRS-free myosynth in-house. That is something that has to be done for each recipient with their own DNA. That means we will probably sell the bioreactors. And those contain a specialized nano-fab.
Later, well, the NADA is in essence a very specialized nano-fab as well. But you can sign a contract to the effect that we won’t sell generalized nano-bots for the next… let’s say ten years. But you have to make sure that you explicitly include our right to produce and sell specialized nanobots and nanofabs.”
“Ok, I can do that. That should be enough to mostly ignore us. And that is just Enertech. Ralcon will be much harder.”
“For Ralcon what we are offering them will change their stance from strongly opposed to in favor of our core business. As soon as Envision does VR, they will want to propagate anything that makes VR easier.”
“Yeah, but as soon as we come out with the new processors, we will cut into their market.”
“They lost that market more than 100 years ago when the patent protection on the current processor generation ran out. Yes, they sell CPUs and GPUs, but no more than anybody else. And they make no more money than anybody else. No, with a new generation of CPUs coming, they can justify packing more ‘functionality’ into Envision and sell a new version.”
“If you say so. By the way, how are the new CPUs coming along?”
“Except for the successor for the Regulon, I’m done with the processors. And yes, I have made a new server processor that is more reasonably dimensioned. I am now working on the mainboard chipsets. After that… well the currently available holo-storage won’t cut it. I will need to design something along the same lines that I used for Glory.”
“Hmm, and the Regulon? How long do you think it will take you?”
“No clue. For the new SoC, I need to start from scratch. And honestly, at the moment I am a bit burned out on processors.”
“All right. It’s not as if that was an urgent matter anyway. But another question. I’ve looked into the specs you’ve given me. The Hyperion is what? 3000 times more powerful than the Cirrium 8k? And the GPU you’ve designed is even more so. And Glory is still only 120 times more powerful than your Precious. How can that be?”
I growled in frustration for a bit.
“Standards. I used UEB 7.3, which is the last version of it. It was overkill for the old tech. Even though it is more than 100 years old. The problem here is, that the new tech quickly hits the wall of what it can offer.
It is limited to 256 lanes per controller, and no device can be addressed by more than one controller. That means I can’t get a higher bandwidth than 24 PB/s. I’ve gotten around that a bit with direct access for some of the devices so that the GPU for example can get the data directly from RAM or the mass storage, without going through the CPU, but it is a hard limit.”
“That sounds troublesome. Do I understand you correctly, that this UEP is more than 100 years old? Then why is there no newer standard?”
“It is called UEB. Universal Equipment Bus. And there was no need. They defined this standard before they created the old processor generation. There were no new processors because they had hit the hard limit on what was possible without a NADA. To get more functionality out of a processor, they would have to make it bigger. But they were already at the limit of what was possible with processor size.
Any more and things like internal power support, cooling, and signal speed would have made the CPU actually slower despite increasing the amount of cores and controllers. That meant that for around 100 years what we have is more or less the best that we could get.
With enough hard work, you could squeeze a few percent more performance out of the graphene, but we are talking about low single-digit percentages. Even the biggest Tesseract did not use more than 50% of the available performance reserves of UEB 7.3. There simply was no need for something better.”
“Ok, so before you made everything before obsolete, there was no need for something new, I get it. But why do you still use it? It is a clear bottleneck after all. Why not make something new?”
“Because it is the standard that everybody uses. You can go to any single electronics shop and get equipment compatible with UEB. If I make something new, nothing that I have not created will fit.”
“Hm, I see. And if you create a new standard? The old one is clearly no longer sufficient.”
“The last UEB forum took eight years to create the 7.3-standard. Heck, it took them nearly a year to just agree on naming it. I don’t want to waste my time doing that tedious work.”
“You got me wrong. I don’t mean why don’t you propose a new standard at some international forum, but instead, why don’t you create a new standard and publish it?”
“Then it wouldn’t be standard.”
“Of course, it would be. Provide the specifications for free and let them create the new devices of their own design. Hell, you could even make the new standard an extension of this UEB. Make it UEB compatible for old tech, but make it able to use more powerful new tech.”
That… might actually work. I would have to think about it for some time.
“Not a bad idea. But… that is not some quick thing to do. Maybe we can do that for the next generation of processors.”
“Yeah. I think it would be better if our first generation of computer tech would use standard equipment wherever possible. I have to hang up. The skimmer is here. See you Tuesday at the latest.”
