@Krewell Some of it goes to recycling when practical. The small, nonrecyclable stuff ends up in the trash. I've had to make peace with it. (And even then, I still have shelves and drawers full of tech stuff)
Well, this is certainly quite a significant day for the tech industry. Jensen Huang's footprint extends well outside of NVIDIA; the full repercussions of his resignation remain to be seen, but there *will* be repercussions across the industry.
https://t.co/rsjgdPvIRn
Zen 6 is here at last.
Even though AMD has not made any paradigm-shifting architecture changes, Zen 6 is still AMD's biggest gen-on-gen IPC uplift ever for the Zen architecture. The company has put TSMC's N2 process (and their own brains) to very good use here
I spent a good bit of time trying to work out all the interconnects and bandwidth on this one. Kudos to Patrick for finding the PCIe link between the NIC and GPU. That was a very neat find for a desktop workstation, and makes it even more like its server counterpart
For those who prefer to read, we just did an almost 10K-word deep dive into the Gigabyte NVIDIA DGX Station and tested performance across components. If you have not seen it, the NVIDIA Grace CPU and Blackwell Ultra have very different network connectivity https://t.co/gSDRTInjNE
@dylan522p@Patrick1Kennedy None that AMD has clued us in about so far, at least. I was as surprised as anyone; I figured SP8 would be half of SP7, right on down to half as many IODs.
5GHz is not a barrier. It is a threshold. A barrier repels you.
(I know that's QC's phrasing, not yours, Jim. But it's still a poor choice of words on their part)
I had assumed this was just a new SKU using the original Orin SoC (NV has several SKUs already), so a new Orin architecture SoC is a much more curious development. Sadly, NV isn't saying just what they altered, besides making arch/Β΅arch changes to the tensor cores
NVIDIA's upcoming Jetson Orin Nano 2 board is rather interesting. 4 years after Orin launched NVIDIA is minting a new SoC for it, with some major changes to the tensor cores to improve real-world throughput: nearly 2x perf on only 16% more on-paper TOPS
https://t.co/l95KIfAK1D
@DouglasABailey I would argue that it is not. Because servers get hijacked much too often.
To be sure, it is *some* data, which is better than none. But it is not enough. Users need to be able to see the complete URL hierarchy
Well this is a nasty surprise this morning. This change to Google search results makes it impossible for users to see the full URL of a result they want to click on - to look before they leap. It makes things both more dangerous and less informative for users.
Google has confirmed the rollout of the goto URL parameters and told me "We have a long history of deploying technical measures against evolving forms of abuse, and we regularly take steps to protect our services and users." Story updated at https://t.co/48u7Rvh3h5
@cavemanjim In practice, I'm sure there are more than a few engineering teams that need the help (good comms is very hard, as you know all too well). But it can definitely go too far, particularly when 25 minute talks don't leave any room for marketing fluff.
@WreckItROFL Yeah. Hot Chips posts the decks morning-of. The original Vera deck was less heavy on the marketing and better organized as a technical presentation.