gautamcgoel

8 ms·

61.
by gautamcgoel 2y ago
Please don't move to Substack. You have a beautiful website which I enjoy visiting. I never enjoy opening Substack.
62.
by gautamcgoel 2y ago
The listing says that 1% of the nodes have RAM with memory errors. I assume this means hard errors since soft errors would just be corrected. Is this typical? Does RAM deteriorate over time?
63.
by gautamcgoel 2y ago
For those of you who don't know: this is a companion to the Princeton Companion to Mathematics, which is excellent but focuses mainly on pure math. Also, the editor of this new volume was a highly respected applied mathematician named
64.
by gautamcgoel 2y ago
This is the part I don't understand. You're saying that the interval from when the DRAM first receives a read request to when it sends the data back over the channel is about 10ns, at least in fancy gaming RAM. Ok, fine. Where is
65.
by gautamcgoel 2y ago
Interesting. If an L3 hit takes 15 ns, then based on your argument a hypothetical CPU with only one core (and hence no fabric) would be better off without L3, since a DRAM read can be performed in just 10 ns.
66.
G.M. Tricked Drivers into Being Spied on (Including Me)(nytimes.com)
36 points by gautamcgoel 2y ago|7 comments
67.
by gautamcgoel 2y ago
Just to make sure I understand: you're saying that checking L1/L2/L3 takes around 35ns, and then the CPU accesses DRAM which takes 10ns? If that's so, how is L3 cache any faster than DRAM? Also, can you explain why the m
68.
by gautamcgoel 2y ago
Thomas is indeed a beast - he was considered one of the smartest profs in the department when I was doing my PhD at Caltech. Just FYI, he left Caltech and moved to Israel a few years ago.
69.
by gautamcgoel 3y ago
Don't let the haters get you down, I love your articles.
70.
by gautamcgoel 3y ago
Why include self-attention layers at all? In other words, why not just alternate SSM and MLP layers?
71.
Remembering Simon Riggs(postgresql.org)
2 points by gautamcgoel 3y ago|0 comments
72.
'You Can Hear a Pin Drop': The Rise of Super Strict Schools in England(nytimes.com)
3 points by gautamcgoel 3y ago|0 comments
73.
by gautamcgoel 3y ago
Nobel Prize in what? There is no Nobel in mathematics or computer science.
74.
by gautamcgoel 3y ago
Can you talk about the interconnect? Is it fully custom as well? How do you achieve low latency?
75.
by gautamcgoel 3y ago
These ideas play a central role in the sci-fi novel Manifold Time . Highly recommend!
76.
by gautamcgoel 3y ago
It's not clear IMO. What you're saying would make sense given that the CPU only has access to 960GB of RAM despite there being at least a TB of RAM physically available. The article blamed the discrepancy on yield, which sounded r
77.
by gautamcgoel 3y ago
The interesting part of the article isn't the processor but rather the LPDDR5 memory, which apparently has ECC support. I'm not aware of any vendors offering such memory - does anyone know which exact memory they're using?
78.
Highlights of NeurIPS 2023 from Reading All 3584 Abstracts(alexzhang13.github.io)
3 points by gautamcgoel 3y ago|0 comments
79.
Some memories of Niklaus Wirth(odersky.github.io)
209 points by gautamcgoel 3y ago|61 comments
80.
by gautamcgoel 3y ago
Would love to know which soaps are good and which are bad, and also why hot temperatures help - I would have assumed that high temperatures stress the components more than low temperatures.
81.
by gautamcgoel 3y ago
LPDDR5 doesn't have nearly as much memory bandwidth of GDDR6.
82.
by gautamcgoel 3y ago
Author here. Thanks for this thoughtful comment. Regarding your first point: yes, this is sloppy writing on my part. The Kalman Filter is always mean-square optimal among all linear estimators , but as you say, it is only optimal along al
83.
by gautamcgoel 3y ago
Ugh, LaTeX is such a pain sometimes. Thanks for catching that.
84.
by gautamcgoel 3y ago
Yes. Also the universal approximation theorem won't directly yield the kind of "strong" approximation we prove, which is uniform in time and holds for an arbitrary (possibly adversarial) sequence of disturbances.
85.
by gautamcgoel 3y ago
Author here: couldn't have put it better myself! Yes, the result isn't super surprising, but it helps strengthen the case for using Transformers for signal processing and control. Also gives practical guidance on how such filters
86.
by gautamcgoel 3y ago
Uh... You're asking the wrong guy :)
87.
by gautamcgoel 3y ago
Author here, happy to answer any questions.
88.
by gautamcgoel 3y ago
Author here. Yes, this filter only produces point estimates; it does not also output a confidence interval. I suspect our construction could be modified to output a covariance/confidence interval.
89.
by gautamcgoel 3y ago
It's strange reading these comments. So many folks have discounted the results of the study by suggesting that of course childree folks will say they have no regrets about their decision because they don't want to admit how th
90.
by gautamcgoel 3y ago
What about the GR86 and the Supra? They are usually considered decent sports cars.

More ›