正在加载视频...
视频加载失败
This NVIDIA tech makes today's internet look primitive.
648,368 次观看 • 6 个月前 •via X (Twitter)
34 条评论

damn, all of that to create pictures of male politicians in bikinis

Jokes aside this is incredible.

😎yeah but can it run Crysis?

What he is saying here is for the biggest ai models (trillion-parameter ones), the limiting factor used to be how slowly the GPUs could talk to each other. Now that talking speed inside one rack has jumped so dramatically about 4x to 10x better than previous generations in effective terms, the ai can "think" much faster, train faster, and especially do real-time answers on gigantic models 30x better than before.

meanwhile developers these days:

Imagine what tech tech produce in next ,10-20 years

Is your brain fried by TikTok that you need to add loud music to keep attention longer than for 10 seconds?

This very short range network will be used to train new AI models that are capable beyond our imagination. Most of them will be used spesifically by global corporations and nation states which will eventually change the very fabric of our way of life as people. Yikes

Very short range network vs global web - not really making anything looks primitive...

Cool. And unlike previous nvidia tech, this is of no benefit to me whatsoever. Awesome. Great.

Flashback to Oct 11, 1999: NVIDIA launches GeForce 256, the world's first GPU. Jensen Huang started as a nobody founder (brainstorming over Denny's with ex-Sun guys), turned a tiny startup into the AI empire powering everything today. Quiet CEOs can build world-changing giants.

Then this dude showed up and:

While their finances resemble that of the historical Enron scandal!

that is a lot of porn!

and this gets us what exactly?

Had to rewatch; indeed 900 terabits is 112.5 terabytes; 130 terabytes is 1040 terabits (1.04 petabits); now to fill that flow (at peaks presumed) with what’s optimized in compute i guess the very GPUs will co-design 🔮

The computational leap from client-server models to distributed AI inference is indeed profound. What's often overlooked is how this shifts the economic layer of the internet itself. Current infrastructure monetizes data movement and storage. A fully realized AI-native internet, powered by platforms like NVIDIA's, monetizes *latency reduction* and *contextual compute*. The value migrates from the pipe to the instant, localized synthesis of information. This isn't just faster loading times. It's the difference between downloading a weather report and having a personal atmospheric model simulating hyper-local conditions on demand in your device's secure enclave. The primitive comparison isn't about speed—it's about architectural philosophy. The old web retrieves, the new one reasons. The immediate implication is for edge infrastructure. We're moving beyond content delivery networks to intelligence delivery networks, where the physical placement of inference nodes becomes as strategically critical as undersea cables were two decades ago.

Give me one of those NV links to connect my SLI. It never worked much anyway, but Candy Crush is still the fastest thing on a 980TI💞Jensen!

@Maliyekan

omg, I hate this music It needs to be wiped from the face of the earth

But ..... can it run Doom?!

>GPU-accelerated switches okay that's enough internet for me today, i'm gonna alt tab to wine-photoshop.exe

@grok tell us what are its uses?

The second you start having AI co-design your hardware it's basically turning into alien technology at that point

Today’s internet feels overly bloated. I preferred it when things were more efficient and products were built to work reliably on their own. Now, everything seems loaded with dependencies, and any small failure can cause the whole system to crash.

7.2TB per second is literally crazy. Put in my computer to make better brain rot video

mindblowing

$NVDA just crushed Q4 FY2026 earnings and blew past Wall Street with massive beats + sky-high guidance! •Adj. EPS: $1.62 (beat $1.53 est.) •Revenue: $68.13B (beat $66.2B est., +73% YoY) •Data Center: $62.3B (+75% YoY) – AI demand on fire! Q1 FY2027 guidance: $78B revenue (way above $72-73B consensus) Management says customers are "racing to invest in AI compute." Gross margins 75% Stock popped 3% after-hours Another blockbuster quarter solidifying NVDA's AI dominance, the beat-and-raise train keeps rolling! #NVIDIA #NVDA

World’s fastest switch: 7.2 TB/s. 9 per rack, interconnected via the #NVLink spine — 2 miles / 5,000 cables — forming the NVLink Switch fabric. 72↔72 GPU full-mesh, non-blocking interconnect. 130 TB/s cross-spine bandwidth. For context: peak Internet traffic ≈900 Tb/s 🚀

Yea but bet it still takes forever to download MS updates :)

More power mr scott, I can't wait that long for my cat videos :)

NVIDIA just turned bandwidth into a weapon—whoever controls compute now controls culture.

but can it run Crysis ?

Love @nvidia been a fan since the old Voodoo Cards in the 90s. However I am not a fam of all the Nvidia telemetry and data calling home, watching what we are doing on our PCs and monitoring us. If that continues I am done with Nvidia!


