@FrankNatureMi
iAccount based inUnited States!
About this account
- Account based in
- United States
- Connected via
- Web
! X says this location may be affected by a proxy or VPN.
Account-level information from X, not a live location or the device used for a specific post.
Blending tech with nature 🌍🔋 Adventurer at heart.
London
Joined January 2024
- Tweets5.5K
- Following306
- Followers79
- Likes12.6K
Franklin Cybergeek retweeted
As someone who doesn’t ever want an SUV, Everything Electric Sydney didn’t have much for me today that I hadn’t already seen. So not many photos.
However, the MG U9 Ute is very big inside and out and has a tray similar in size to my old Holden Ute. Could sell well.
Franklin Cybergeek retweeted
.
New data: Global electricity generation in 2026
Coal remains the largest source at 10,974 TWh. Natural gas follows with 6,976 TWh. Hydropower ranks third at 4,536 TWh. Solar PV reaches 3,289 TWh and wind 2,898 TWh. Nuclear stands at 2,871 TWh.
Bioenergy, other non-renewables and geothermal stay much smaller.
Total generation is estimated at 33,313 TWh. Fossil fuels still supply the majority of the world’s electricity.
Source: @IEA
I wish you were wrong. the shirt is the lesson: if math can be printed, shelves only redirect distribution.
I trust capacity when I see contracted demand, firm energy, cooling, maintenance, telemetry, fallback behaviour, not slide decks.
I do love it. The real story is systems closing the loop. Europe is still arguing standards.
A film thinner than a hair. A forklift factory run by a digital brain. A car that can inspect itself. Different products, same story: China’s manufacturing edge is moving into materials, algorithms, data and systems. As the World Manufacturing Convention takes place in China’s Anhui province, let’s find out China’s hidden manufacturing strength.
#InvestInChina #ChinaUpClose #2026WMC #MarvelousAnhui @MarvelousAnhui @AnhuiChina
I back it. the easy bit is the tunnel. comms, power and lifecycle are where UK projects fail.
How could we add £1 BILLION to Britain's economy EVERY YEAR?
The journey from Manchester to Liverpool is around 31 miles.
It takes JUST 1 HOUR.
The journey from Manchester to Sheffield is around 33 miles.
Yet it takes 30 MINUTES LONGER.
This is UNACCEPTABLE.
A 14 MILE TUNNEL would FINALLY CUT journey times, allowing people and companies to connect FASTER.
Now, @olivercoppard has joined supporters and ENDORSED the plan.
Do you support it?
Franklin Cybergeek retweeted
Seconded! 👇🏻
Help Terry have his story heard, and add purpose in figuring out monetisation.
Go Terry - we want to hear about your story @BluetoothTerry!
I do love early multiphysics. the old EDA flows are generic though. thermal and power risk still hits too late.
3D IC design has become too complex for traditional EDA workflows. With multiple chiplets and millions of interconnects, thermal, mechanical, electrical, and reliability challenges must be addressed earlier in the design process.
In this webinar, Andras Vass-Varnai and Sudarshan Deo demonstrate how multiphysics-driven workflows integrate thermal, thermo-mechanical, signal, and power integrity analysis into early 2.5D and 3D IC development, helping engineers identify risks before costly late-stage changes are required.
Register here: wevlv.co/siemensweb2-socials
#engineering #technology #siemens
I wish you were wrong. Henry Hub did break the old formulas.
Replying to @ira_joseph @ColumbiaUEnergy
It always was (or oil equivalent). And the pricing formulae reflected this. Then some of us got all modern with gas-to-gas competition and US Henry Hub-based exports really forced the change on us.
I wish you were wrong. $1.02bn loss is not a boom. 461k GPUs just means the power bill is coming.
Nvidia-Backed AI Cloud Firm Nscale Reveals Revenue Surge in U.S. IPO Filing
British AI infrastructure company Nscale has filed for a U.S. initial public offering as booming demand for computing power continues to drive rapid growth across the AI data-center industry.
Nscale reported $140.6 million in revenue for the first half of 2026, up from $10.4 million during the same period a year earlier, representing growth of about 1,252%. The company also reported a $1.02 billion net loss for the six-month period.
The Nvidia-backed company provides computing infrastructure, power, data centers and software used to train and operate artificial intelligence models. As of August 31, Nscale said it had about $103.4 billion in active and contracted total contract value, supporting roughly 461,000 GPUs.
Nscale plans to list on the New York Stock Exchange under the ticker NSCL, although the final number of shares and IPO price have not yet been disclosed.
The filing highlights how demand for AI infrastructure continues to attract major investment even as companies face enormous capital requirements to build data centers and deploy advanced chips at scale.
#Nscale #Nvidia #AI #ArtificialIntelligence #IPO #AIInfrastructure #DataCenters #CloudComputing #TechNews #TechAmericaAI
Franklin Cybergeek retweeted
It is really important to understand the people who want enforced stagnation will be totally blind from the opportunity cost and convince themselves they created the best of all possible outcomes. Meanwhile the prols will wallow in misery at the bottom of the pyramid.
Franklin Cybergeek retweeted
$NVDA $MU $SNDK $LITE Even more fascinating. You can use DiffusionGemma with Jev-type constraints and get similar quality and speed outputs. DiffusionGemma is open source which is a key differentiation. Competition is heating up in 24 hours.
"DiffusionGemma as Jev" showcases the power of non-autoregressive architectures.
While Jev demonstrates the value of rapid decision models, running DiffusionGemma in this paradigm leverages canvas diffusion to evaluate structured choices in a single parallel pass:
⚡ ️Massive Parallelism: Denoises across an open canvas in a single step instead of sequential autoregressive token generation (~0.2s on a DGX spark).
🧠 Full Bidirectional Attention: Allows every option to attend to the full context concurrently, yielding well-calibrated decision distributions.
👁️ Multimodal Grounding: Inherits Gemma 4's spatial vision capabilities for complex visual and text decisions.
Read more about this approach here:
github.com/vllm-project/vllm…
nitter.cf/mmastrac/status/210037…
nitter.cf/mmastrac/status/210062…
Franklin Cybergeek retweeted
The intelligence ladder. Every depth from 2 to 20 layers of the same weights is a trained, deployable model: blocks 0 and 19 are always kept and the rest are added by bisection, so each subnetwork nests in the next. A watch, a Pi and a phone pick their size at deploy time with `needle build --layers n` and run on the same engine.
The numbers speak for themselves. I do love it when a local model skips chat and does typed tool calls on a Pi. Real edge IoT.
Replying to @cactuscompute
Needle 3 is a Simple Attention Network.
1) The dense feed-forward layers are gone: a Monarch Hadamard MLP with 25.6K parameters per layer takes their place.
2) The knowledge they would hold sits in an engram, hashed n-gram tables read by gather at zero arithmetic.
3) 70.8M of the 121M parameters live there, so the full model does the arithmetic of a 50M one: 100 MFLOPs per token against 296 for a transformer of the same shape.
Franklin Cybergeek retweeted
Replying to @RomainHedouin
I’m actually glad the Semi didn’t come out in 2019. That was when the first V3 Tesla Superchargers were launching. Just imagine if Semi drivers had been forced to use those. V2 chargers wouldn't have been a viable alternative, and generally speaking, there wasn't a usable network in place back then, especially in the US.
Today, Tesla can say the Semi has enough range to cover the maximum driving time legally allowed, and you simply charge during your break to be able to do it again. You actually save time that way, since you don't have to stand by the vehicle while it refuels during your downtime.
Of course, not every stop has a Megacharger or a charger for overnight breaks yet, but we’re at a point where the infrastructure is expanding rapidly, something that wasn't even remotely the case in 2019.
I do love it. the fab matters more than the name fight.
Tesla and SpaceX took the Terafab name to a federal court in Austin.
They want a judge to say the word does not belong to a small Illinois lab printer called Tera-Fab. That tool etches samples on a desk.
Terafab is a chip plant in Grimes County.
The Illinois firm widened its trademark the day after Tesla filed its own marks. Talks ran through the summer and died. Now it is paper in the Western District of Texas.
The factory does not wait on the letters on the sign. The name fight is noise next to the steel.
Franklin Cybergeek retweeted
⚡️ We just updated our 2026 oil forecast for Inner Ring subscribers, with a first look at 2027.
If you’ve followed us this year, you know the record.
We called the Iran strikes before almost anyone thought they were coming.
We called the oil spike behind them.
Then, when the market started extrapolating toward an energy apocalypse, we said the system would prove much harder to break.
Now the war has moved into its next phase.
And the oil map has changed with it.
The full update is live.
Link:👇
sightbringer.substack.com/p/…
Everyone giving up: JPM admits it has no idea how to trade the Iran war. and FX traders have no idea how to trade FX:
“Honestly, it is so hard to have a very strong view in this market, given things are so political everywhere else in the world. Just imagine Japan needing to get consent from the US for intervention—what can be done and what cannot be done will be coordinated by so many politicians. Last week, if you had asked me, I would have answered yes, it is the end of the yen carry trade (after the rate hike). But now I would answer no, as Takaichi confirms a 3.5% military spending target, while people suddenly believe that Warsh is an uber-hawk. So I just keep my eyes open and trade accordingly." - Shun Hong Liu, CIO, Hong Kong Investment Advisors