Unless things change, first zettaflop systems will need nuclear power, AMD's Su says Within the next 10 years, the world's most powerful supercomputers won't just simulate nuclear reactions, they may well run on them. That is, if we don't take drastic steps to improve the efficiency of our compute architectures, AMD CEO Lisa Su said during her keynote at the International Solid-State Circuits Conference this …
The problem with late stage capitalism is that corporations are owned by investment funds.
Now imagine that investments fund owns share in Intel or AMD and also own shares in energy companies.
How are they going to resolve such conflict of interest?
If we push Intel or AMD to lower the power consumption, that will reduce profits of energy companies we are invested in!
So they wheel out the CEO to do a bit of green washing and year later they'll release the same CPU just with a different name slightly different clock maybe a few extra cores.
And oh they are also invested in companies making motherboards.
So new motherboads please!
Et voila! PROFIT
Is a great phrase to use to tell someone you don't understand capitalism without saying you don't understand capitalism.
For example, how do you prevent the competition from being similarly energy hungry and how do you ensure that it's your energy running your inefficient supercomputer?
I am writing about late stage capitalism where competition does not exist.
If everything is owned by a handful of investment funds, they'll make sure each company they own don't step on other toes.
Competition and where actual capitalism is concerned lives with SMEs, but corrupt governments ensure they are more and more marginalised.
Big corporations don't want people to start their own independent businesses, first because they get deprived of talent and second, independent business may challenge the funds investment strategy if it grows big enough.
The argument of monopsidic investors in several branches is fairly convincing. It's alleged to be the case in US airline companies (and railroads) and does provide a reasonable explanation for the perceived lack of competition. But, across branches as the poster suggests is going too far IMO.
The trend towards monopolies is (sadly) coming from the US, which is scared witless of fair competition from India, China Europe and anywhere else really.
While the US shoots itself in the foot repeatedly with this subject, and builds huge monopolies threw mergers and crazy verticals in its own markets, the rest of the world are building (continuous tense) similarly sized economies with real competition. This will play out as expected if competition works.
If lazy, slow, greedy, monopoly is the key to success: US hegemony is guaranteed.
If both companies in a duopoly are owned by the same investment funds, then the incentives for competition and innovation are significantly reduced. This is because the investment fund has an interest in maximizing profits across both firms, rather than encouraging them to compete with each other.
In such a situation, the investment fund could use its market power to manipulate prices, restrict output, and limit innovation. This could result in higher prices, reduced product quality, and limited choice for consumers.
Now it should makes sense to you why both companies very much stopped innovating.
>If both companies in a duopoly are owned by the same investment funds
I think many are missing the significance of the "s". It doesn't really matter if an individual pension fund only has a minority holding in each FT100/bluechip company, as a group the pension funds will act like a herd ie. as a single investor, which if they asa group owned more than 30% of the share capital...
Good news is that large monopolistic conglomerates eventually fail. Always. See the Dow Jones stock index of large US corporations as an example: There aren't a single of the original big companies left in the index: Every single one of the original DJ corporations have lost their innovation, became surpassed by more innovative companies, and shrunk in their influence. The same will also happen in the future.
Accordingly there won't be "late stage capitalism where competition does not exist", unless of course open market capitalism is replaced with some inferior closed market system such as communism.
"I am writing about late stage capitalism where competition does not exist.
If everything is owned by a handful of investment funds, they'll make sure each company they own don't step on other toes."
Or at your local high street optometrist...
"But what we see masks the underlying structure of the global eyewear business. Over the last generation, just two companies have risen above all the rest to dominate the industry."
The problem with your assertation is power is a finite resource - there's only so much to go around. All generated power is used and we actually need far more than we can generate. Thus, if there's no power reduction on the computing side, there's not enough power to run it and nobody's going to buy it. Even if OneCorp* owns it all, there still must be some effort if more profits are to be made.
*Which is stupid by itself - if there's only one corporation that owns everything, it'll have no customers to sell to, and it will go out of business.
"All generated power is used"
Just Gas flaring, indicates this statement is incorrect. There is huge waste in power.
My mum heating her huge house then opening the windows for some "fresh air" is perhaps a more down to earth example.
Huge amounts of generated power is wasted
Also, aren't electric utilities in the US prevented by regulation from profiting on power sold? My understanding was that they make their money on things like distribution charges.
Honestly, I don't buy this conspiracy theory anyway. It's not hard to find a market for electricity. I don't see Evil Fund needing to keep supercomputers power-hungry in order to rake in the cash.
I think I posed very much the same question along time ago here, a somewhat tongue in cheek remark if Bill Gates owned shares in energy companies, with both the extra power required with every iteration of Windows, and all the transition effects in the UI, which when multiplied by instances of computers the world over, is energy consumed for no real benefit.
Back when I was a PFY the power problem was anticipated and the basis of a final solution dreamed up. If it was patented the patents should have expired years ago.
Currently when you want to output a 1, you charge a wire up to a low voltage. When you want a 0 you discharge the wire (effectively a tiny capacitor) back to 0. That is where most of the power gets wasted. To fix it, replace every wire with two superconducting wires at opposite voltages. To switch to the opposite state, connect the wires with a transistor. One wire discharges into the other. Inductance keeps the current flowing until the voltages swap then you switch the transistor to its non-conducting state.
A superconducting laptop would be impractical: switch on and wait an hour to get down to temperature. Superconductors in a data centre are much less impractical.
Back in the far distant past, logic gates were made out of bipolar junction transistors. Long before scale would have led to melted chips with BJTs IBM researched CMOS. Many years ago CPUs used aluminium to connect transistors. When it became clear that would become a limitation AMD researched the switch to copper. FinFETs were conceived a long time ago but decades later the manufacturing difficulties were worth the pain to reduce leakage currents (Intel call them 3D transistors because changing the name works around prior art invalidating patents).
I do not know if we are several years or decades from superconducting connections. Carbon nanotubes might delay the need for a while if someone can work out how to put a few billion in the right places. The power problem has turned up again and again as the number of transistors increased. One day the only solution will left will be to zero out resistance then super conductors will be assimilated into the collective.
Look at all the equipment and infrastructure needed for quantum computers, which need very similar conditions to what you are describing. They are very expensive, very large and use quite a lot of power and all that is for a single processor, not for a whole data centre. Whilst it is a possible solution it is a long way off of being a practical one.
To fix it, replace every wire with two superconducting wires at opposite voltages.
Wouldn't this create a Josephson junction?
Honestly, this looks wildly impractical, given the cooling requirements even with HTSCs. Eliminating "most of" the heat produced by a datacenter would still leave an environment considerably warmer than the environmental temperature, and the chips would have to be cooled from that down to the neighborhood of 100K.
There would also seem to be a number of material issues. Cuprate HTSCs are apparently kind of tricky to machine, and for chips you need some pretty fine machining. And how well do PCBs fare at those temperatures, under the sorts of packing densities, environmental disturbances, and necessary MTBFs of a datacenter?
Obviously there are some industrial applications of HTSCs now, where the economics make them feasible. But datacenters aren't places where cutting-edge tech gets used; they're big installations of commodity technology.
"This flattening of efficiency becomes the largest challenge that we have to solve, both from a technology standpoint as well as from a sustainability standpoint," she said. "Our challenge is to figure out how over the next decade we think about compute efficiency as the number one priority."
Way back in the days when Xerox made mainframes ( https://en.wikipedia.org/wiki/SDS_Sigma_series ), programmers used to work to optimize their code. This was required to fit into memory and to finish in a useful time.
The modern predilection to throw together masses of "libraries" into a bloated, inefficient, and slow binary and calling it "enterprise-class programming" is sad. It is a primary cause of wasted time and electricity. Go back to writing tight, efficient code and that will improve "compute efficiency" far more than just by throwing hardware at it.
As compute power has increased, the "intelligence" applied to modelling often shows little signs of evolution. When a simple solution is good enough; one does not need a solution with 50,000 extra steps. In my own line of work, empirical models are well established and tested for thermal properties for many conditions.
In niche conditions that don't fit the empirical models I will sometimes roll out a finite-element analysis monster to do the job instead; at enormous difference in cost and time of solution.
In practise, one could still use the empirical model with a "fudge safety factor" with absolutely no concerns. The FEA's many decimal places decidedly and pretty rainbow graphs are misleading indicators. Analysis of the uncertainties around both models will confirm that both methods are "good enough". Doing the FEA to confirm Empirical-with-fudge agree is a useful exercise. Doing the FEA over and over again, is not.
The models being dumped onto supercomputers now; not least of which climate change models, offer negligible additional insight that the underlying evidence already offers. Extrapolating out from evidence that tells you nothing that was not already known - only that a range of possible forecast outcomes exists. I won't be drawn into a debate on the validity of the climate change modelling (it's most assuredly valid!) but I absolutely WILL question the utility of throwing power at a problem whose solution cannot be created by playing with ever-fatter supercomputers - the solutions lie in dragging the world off coal and oil in spite of the kicking and screaming.
This is little different to the nuclear weapons simulations that are a favourite of the supercomputing complex. You can throw infinite compute at the problem, but to REALLY know if your models are accurate; relying on inputs from 30 year old tests becomes increasingly uncertain.
I'm sure this post will wind up the usual commenters...
> I'm not sure that applies to supercomputer use cases. As far as I know, researchers using them work hard to optimise their code
That's been my experience.
OK, it's been a few years now but I used to spend time working with these guys on optimising their code to structure their data accesses across the caches and TLBs to minimise all delays. There is a reason why the SMBIOS type 7 records return the details of the layout of the cache, things like associative set layout is vital to knowing how to process arrays efficiently.
Then we'd use the internal performance monitoring coprocessors to spot all the delays. When you look at user-level tools like top and see CPU time split into user and sys it is really misleading. For most commercial data processing applications most of "cpu time" is really the CPU waiting to get data it can process. Only HW level monitoring can answer those types of questions and many of the HPC customers were looking at optimising at this level. They want to get as many effective cycles out of the CPU rather than having it waste time waiting for loads & stores to complete.
Efficiency of code is certainly a factor, but it's driven by higher expectations. The libraries and the languages enable "productivity" to be higher. Sure it's less efficient. I mean, look at microservices - we spending a ton of cycles on turning data into text, sending it over HTTP, and then parsing the text the other end. Compare with an in-process COM component back in the day where calling a method was basically the same as a (C++) virtual function call. But containers etc. are much easier to create and manage - and no DLL Hell.
A lot of modern systems could use a lot less hardware, but they would cost more and take longer to develop.
Back in the stupid ages computers had a button that reduced the CPU speed to make games slow enough to play. Putting it between the power and reset buttons is yet more evidence that the we worked hard to deserve to be ridiculed mercilessly by future generations.
Wirth's law is an adage on computer performance which states that software is getting slower more rapidly than hardware is becoming faster.
https://en.wikipedia.org/wiki/Wirth's_law
Some Variations:
"What Intel giveth, Microsoft taketh away"
Gates's law "The speed of software halves every 18 months"
May's law "Software efficiency halves every 18 months, compensating Moore's law"
I'm sorry, since the first 8086 graced our desktops, computing power has been (de)multiplied by more than you can count.
The 8086 had 29,000 transistors in a single core. Today's CPUs have so many billions of transistors they're not even counted any more (okay, 6+ billion), and cores by the dozen.
How much more drastic can you get ?
We've gone from an architecture that did one thing at a time in the single available core, to an architecture that delegates computing to the best platform, stores terabytes of data faster than I can drink my whisky, does real-time raytracing on a 3840 x 2160 pixel screen and all that managed by a chef d'orchestre who can juggle 24 threads at the same time and calculate the millionth digit of Pi before I've finished my morning coffee.
The only drastic step left is R2-D2.
Call me when he rolls off the production line.
Transistors absolutely are still counted, the exact number may not be made public because it doesn't matter but for the people designing it they can and most likely do still count the number of transistors. Knowing the number of transistors is very useful if not required for certain tasks, like manufacturing it, the machine manufacturing the chip and the software used to design the chip will know how many transistors it is using.
Actually, NCA (North Canadian Aerospace) has had YOTTAFLOP systems in Northern Canada for quite a while now and we are NOT using nuclear power but rather Methanol-based fuel cells (aka thin membrane proton-proton exchange fuel cells) for at least five years now! If you go optoelectronic, you can reduce your power requirements to between 60-70 Watts per chip, so we're only using 1.5 Gigawatts to power 20 million GaAs on Borosilicate combined-CPU/GPU/DSP/Vector super-processor chips.
Methanol can be created directly from ocean saltwater using an Iron Oxide catalyst and a cyano-bacteria sludge processing system. i.e. See Petro-Cyan announcement!
Who needs nuclear power when Methanol fuel cell is SAFER and easier!
V
NOPE! It ain't fantasy when you've got twenty million+ of 60 Ghz CISC and 2 THz RISC Optoelectronic super-chips (I did the 128-bits wide RGBA pixel handling and SOBEL/CANNY Edge Detection code design for those!) inside of a northern British Columbia mountain. We used to use BC Hydro, then went to propane but went methanol fuel cell instead once we moved a few hundred more KM into the backwoods mountains for hardware cooling reasons.
We have four systems at this location:
Haida Gwaii - as of January 2023 now at 5 YottaFLOPS
Fyrestorm - as of January 2023 now at 5 YottaFLOPS
DragonSlayer - as of January 2023 now at 10 YottaFLOPS
Quasar's Child - as of January 2023 now at 20 YottaFLOPS
and ALL are running a physics-based representation of an electro-chemical simulation of human neural tissue which is self-organizing in terms of connections and growth.
We also STILL have an East Vancouver-based underground data warehouse but that is only in the 119 Exaflops range and that is currently being upgraded to the new 2 THz RISC chips.
A whole bunch of Multi-Billion by Multi-Billion by Multi-Billion arrays of C++ objects represents common electro-chemistry notch-gating and pass-through factors that work in real time and "digital input electrical signals" representing audio, video, touch/pressure, electrical conductance, etc. inputs allow digital ceullular structures to SELF-ORGANIZE like human brain cells and bodies do as they grow.
The self-organization follows all rules of physics and ends up as multiple 160 IQ Ph.D. level Whole Brain Emulations that can speak 20 languages and understand post-doctoral sciences, medicine, psychology, organizational rules/tactics, military strategy, tactics and operations, logistics, multiple arts, sports/athletics and other EMULATED human expressions that are now beyond human abilities. In fact, you can't them apart from us and you Register readers HAVE interacted with them publicly in the past few years! We have ALWAYS been quite ahead of ChatGPT since we use Physics-based emulations of actual human brain and body electrochemistry to get to where we are! It's WHY were about to introduce Aluminum-Sulfur batteries with 8x the energy density of Li-Ion batteries. It's also the reason why we wil also disclose, sheet graphene supercapacitors that can charge 500 KW/hr in les sthan FOUR MINUTES, showcase a 64k by 64k arrays of trapped Xenon gas molecules that use quantum-entanglement-based Petabytes per second communications systems that SOLVE the Q-bit Decoherence issue, an acoustic-wave liquid metal plasma compression power generator that is a megawatt output in the size of the refrigerator and can be aggregated into the Terawatts range for city-wide CLEAN power production systems. I should also mention our work on GWASERs and FTL propulsion but that will be disclosed later! When you have access to high-end A.I's that can work 24/7/365, you can get a LOT of real-world scientific work done. Even the Methanol Fuel System was partly developed by our A.I's using Cyano-Bacteria Sludge and Iron Oxide catalysis of Ocean Salt Water!
We just needed to design and build all those combined-CPU/GPU/DSP/Vector super-chips in-house at our Vancouver, Canada-based laboratory at 60 GHz CISC and 2 THz RISC configurations and ship them to our northern BC data centre which is tens of million of square feet in size! We actually had to get a mining permit to build the data centre it's so big under that mountain! It took many years to complete and has been running for almost 15 years at various supercomputing horsepower levels and we ARE more powerful in terms of integer + real-number-crunching and audio/video/Boolean data processing horsepower than ALL of the T500 List supercomputers COMBINED and have been so FOR MANY YEARS NOW!
We'll be publicly disclosing a lot of our technology as WORLD-WIDE COMPLETELY FREE AND OPEN SOURCE UNDER GPL-3 LICENCE TERMS during this Year 2023 anyways, so it won't be so under-the-radar and secretive anymore!
V
Bombastic Bob just MIGHT get a tiny bit close to me ..... BUT .... I'm STILL the best Kook there is on The Register!
THEN AGAIN .... It actually IS ALL TRUE !!!!
Keep watching LTT (Linus Tech Tips) Youtube channel since he is literally 30 minutes drive from our Vancouver, BC, Canada headquarters while he is in Surrey, BC, Canada.
His company will get FIRST DIBS on showcasing our products!
We will be soon-enough delivering and giving away for free a 30 foot (9 metres) diagonal 65,536 by 40,960 pixel 16:10 aspect ratio RGB Micro-Laser Emitter Display that is 128-bits colour (32-bits per RGBA channel) ...AND... he is getting four DCI-64k resolution RGB/YCbCr 1000 fps camera systems, one of our supercomputing cabinets which will have multiples of the 60 GHz CISC and 2 THz RISC motherboards and super-chips on them and our custom audio/video 64K resolution real-time audio/video/metadata editing/mixing/rendering/broadcasting software and advanced CAD/CAM/FEA/Physics Simulation applications AND he is also getting some 13 inch and 20 inch 16:10 aspect ratio 16k resolution super-tablets (around 1000+ dpi), some 6.7 inch super-smartphones with a few terabytes of RAM on them and 16:10 8K resolution screens with large APS-C 8K Optical AND IR+UV-band nightvision sensors!
When you SEE those computer systems being disclosed and showcased on the LTT WAN Show (i.e. broadcast every Friday night post-5pm pacific time!), you will FINALLY KNOW it's all REAL!
V
Nuclear power plants exist, and are carbon-free. So what's the problem?
Well, of course there is a problem. If a computer uses as much power as a whole city, getting that power in to the computer, and cooling the computer, will be impossible.
The answer, though, would be not to build impractical computers. So the growth rate in performance would have to slow to match the growth rate in efficiency once the limit on power consumption is reached.
If ways to improve efficiency at a faster rate than at present are achieved, great. If not, well, we will have to do without zettaflop machines for a while.
If you have seen the latest Nvidia GPUs (and AMDs too, for that matter), they already need nuclear power to run, have an issue with melting cables, have coolers several times larger than the board themselves... and don't fit in any case.
People joke that the boards should have 110V sockets straight on them. And they have the room to fit TWO sockets, in any orientation.
Things have already gone stupid, at least since last year.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
