Die Zukunft des Supercomputings: TotalEnergies arbeitet mit NVIDIA und Dell zusammen, um „Pangea 5“ zu bauen, einen über 100 Millionen Euro teuren KI-Supercomputer, der die Rechenleistung versechsfacht und gleichzeitig den Energieverbrauch um 40 % senkt.
https://interestingengineering.com/science/totalenergies-pangea5-supercomputer-ai-seismic-imaging
7 Kommentare
Submission Statement:
This €100M+ partnership between TotalEnergies, NVIDIA, and Dell to build the Pangea 5 supercomputer highlights a massive trend: the oil and energy sector is becoming one of the biggest drivers of AI infrastructure. While a 6x boost in computing power for seismic imaging and power grid modeling is impressive, the most futuristic aspect is the efficiency design. It cuts energy consumption by 40% and recycles the massive residual heat from the GPUs to warm the entire research facility hosting 2,500 employees. As AI data centers face severe criticism for destroying energy grids, could this type of circular heat recovery and hardware optimization become the mandatory standard for all future supercomputers over the next decade? Let’s discuss.
Adam Jensen is gonna be *pissed* when he hears about this…
Sixfold computing power increase is a lie unless it’s compared to some old system. NVIDIA has been mixing precisions to make their numbers look better than they are. They could have +50% gen performance improvement but they will lie and say it’s 3x because they will compare INT8 perf of previous gen with INT4 of next gen (example precisions – exact comparison depends on what was added as native in the new gen). Meanwhile scientific community at large has been very disappointed with NVIDIA’s FP64 performance.
Besides, TotalEnergies is an oil company. They will use this computer to figure out ways to drill more oil.
What’s interesting is that supercomputing is no longer only a scientific or government race energy companies, cloud providers, and AI infrastructure players are all converging into the same compute ecosystem now.
The energy efficiency part honestly matters just as much as the raw compute increase. AI scaling was starting to look physically unsustainable without breakthroughs like this.
In other words, all those big data centres being built now will be obsolete long before they have paid for themselves.
Correction: „that multiplies computing power sixfold while cutting energy use by 40%.“
This claim applies to the company buying the system. What it says is their existing compute is weaker and consumes more power. This isn’t a claim about some whiz-bang future tech invention.
From the article:
„Pangea 5 will multiply TotalEnergies’ computing power sixfold for AI and seismic workloads.“