While the analysis is definitely intresting, OP misses completely the point. And worst, this work becomes a joke. OP is not losing $780 because OP, thanks to Anthropic, is making far more money than without it. This does not wanna say that Claude should have current level of uptime, on the countrary, they should improve it but the point OP makes is plain wrong.
As a European, this is genuinely alarming. Headlines like this make me feel that NATO is effectively over at best, and that war may be approaching at worst.
`”fundamental disagreement" between the Trump administration and European allies.`
The Peace President is about to bring war on Venezuela and destabilize the country (and probably with far reaching consequences in the region). Who would have thought?
Wow! As others have said, deal of the century!! As a side note, a few years back, I used to scrape eBay for Intel QS Xeon and quite a few times managed to snag incredible deals, but this is beyond anything anyone has ever achieved!
I have also bought Crucial for decades. Great quality and reliability for a fair price. Anybody doing anything semi-professional will be impacted by this questionable decision.
Check the specs again. Per chip, TPU 7x has 192GB of HBM3e, whereas the NVIDIA B200 has 186GB.
While the B200 wins on raw FP8 throughput (~9000 vs 4614 TFLOPs), that makes sense given NVIDIA has optimized for the single-chip game for over 20 years. But the bottleneck here isn't the chip—it's the domain size.
NVIDIA's top-tier NVL72 tops out at an NVLink domain of 72 Blackwell GPUs. Meanwhile, Google is connecting 9216 chips at 9.6Tbps to deliver nearly 43 ExaFlops. NVIDIA has the ecosystem (CUDA, community, etc.), but until they can match that interconnect scale, they simply don't compete in this weight class.
Google's real moat isn't the TPU silicon itself—it's not about cooling, individual performance, or hyper-specialization—but rather the massive parallel scale enabled by their OCS interconnects.
To quote The Next Platform: "An Ironwood cluster linked with Google’s absolutely unique optical circuit switch interconnect can bring to bear 9,216 Ironwood TPUs with a combined 1.77 PB of HBM memory... This makes a rackscale Nvidia system based on 144 “Blackwell” GPU chiplets with an aggregate of 20.7 TB of HBM memory look like a joke."
Nvidia may have the superior architecture at the single-chip level, but for large-scale distributed training (and inference) they currently have nothing that rivals Google's optical switching scalability.
Dr. Avi has a pretty clear point and he goes into details on the JRE released just yesterday. https://youtu.be/EaAun27gftk
What most stand out is the sheer amount of closed mind people in the accademia, Avi is not afraid of making suggestions of what it might be and even saying “if it turns out of being a rock, so be it”.
Meanwhile, in third-world, overly bureaucratic Italy, one has to wait several months to get all the paperwork in order to take advantage of a solar installation. Self-deployed solutions are also limited to 800 watts, which is peanuts in today's world.
I'm also an AirGradient user. AirGradient prioritizes accuracy and it's fully OSS. The sensors are fully replaceable; not if but when they eventually lose accuracy. I'd argue that monitoring PM2.5 levels indoors is even more crucial than outdoors. In our home, we have good air recirculation and HEPA filters in every room. Since we began keeping indoor air clean, our spring allergies (and general allergy issues throughout the year) have become much less severe. IMO AirGradient is the best way verify current levels and ultimately ensure the filtration works.
https://imgur.com/a/UIkkANL