adrian_b

33 ms·

481.
by adrian_b 4mo ago
If you can avoid GPUs, that is preferable. The reason is that with GPUs it is far more difficult to reach a great percentage of the maximum theoretical throughput. Most GPU programs reach only a very small fraction of what is theoretically
482.
by adrian_b 4mo ago
Yes. The fastest US supercomputer, El Capitan at Lawrence Livermore National Laboratory, reaches only 1.809 exaflops, while this reaches 2.198, 22% higher. On the HPCG benchmark, which is more strongly influenced by memory bandwidth, the ad
483.
by adrian_b 4mo ago
No. The AI oriented GPUs or TPUs have either weak FP64 throughput or they may not support FP64 at all. They can compete neither with CPUs nor with GPUs that have good FP64 support, like the AMD CDNA datacenter GPUs, which occupy all the top
484.
by adrian_b 4mo ago
No, that is too slow. Most claims about the cost of emulating FP64 on GPUs are wrong, because they assume that only the significand of floating-point numbers must be extended. In reality it is even more important to extend the exponent, bec
485.
by adrian_b 4mo ago
This seems like a "sour grapes" comment. The new Chinese supercomputer beats all US supercomputers also in HPCG, not only in Linpack. What is remarkable is that this was done despite the US attempts of sabotaging HPC in China by &
486.
by adrian_b 4mo ago
Additional details in: https://arxiv.org/abs/2605.08633v1 https://www.servethehome.com/arm-cpus-take-number-1-in-lates... https://www.top500.org/news/lineshine-debuts-no-1-top500-e
487.
by adrian_b 4mo ago
I have used only HiDPI monitors with X11 on Linux for about a dozen years and it has always worked perfectly. However I use XFCE, where you can set directly the DPI values of the monitors, so that everything will work fine. I have heard Gno
488.
by adrian_b 4mo ago
> but i didn't find them addressing it explicitly in the text Yes, this is exactly what bothers me about this article and about a few similar articles published in the past, that they do not contain any evidence that their claims ab
489.
by adrian_b 4mo ago
If you self host then you can audit the open-source llama.cpp or whichever other program you are using for inference, to see exactly what it does, and also whichever open-source harness you use for implementing a coding assistant or other a
490.
by adrian_b 4mo ago
No, the law of the excluded middle is not relevant for a demonstration by "reductio ad absurdum", when it is performed correctly. If P is a proposition and it is demonstrated that "P implies not P", from this it can be c
491.
by adrian_b 4mo ago
Another poster has provided a link to a definition of "proof by contradiction", which I assume that it is the one that you mean ("A proof by contradiction is a proof of a positive by refutation of the negative."). Unlike
492.
by adrian_b 4mo ago
"What can’t be false must me true" is what classic logic called "Tertium non datur", and which has absolutely nothing to do with a demonstration by "Reductio ad absurdum", i.e. with "assuming a fact, showi
493.
by adrian_b 4mo ago
You might be able to put more in series if the tolerance of the capacitance value is low enough. Otherwise, you can use a step-up voltage converter to generate a voltage as high as you want, in order to distribute power at greater distances
494.
by adrian_b 4mo ago
The usefulness of the smuggled NVIDIA GPUs has greatly diminished for AI purposes, because the elimination of NVIDIA as a competitor has allowed the growth of the production of domestic GPUs. Moreover, China has just demonstrated a supercom
495.
by adrian_b 4mo ago
For now, "training on frontier model conversations" are just allegations for which no evidence has been provided, while their research publications are certain evidence about their innovations. The Americans should wake up to real
496.
by adrian_b 4mo ago
Additional details at: https://www.servethehome.com/arm-cpus-take-number-1-in-lates... Also at: https://arxiv.org/abs/2605.08633v1 This uses a custom Armv9-A CPU with SME (scalable matrix extension), w
497.
Chinese LineShine Supercomputer Debuts at No. 1 in TOP500(top500.org)
3 points by adrian_b 4mo ago|1 comments
498.
by adrian_b 4mo ago
The "à" between Québec and Montréal should have "accent grave", not "accent aigu".
499.
by adrian_b 4mo ago
Both 10 Gb/s Ethernet and 10 Gb/s USB have bit data rates that are 3% lower than 10 Gb/s, due to encoding (64/66 bits for Ethernet, 128/132 bits for USB). So the their maximum speed is approximately 9.7 Gb/s. T
500.
by adrian_b 4mo ago
"No runtime abstraction" seems the correct approach for good performance. Run-time abstractions should always be avoided when the problem can be solved at compile-time. While the conditional compilation syntax of the C/C++ pr
501.
by adrian_b 4mo ago
I have been using only 4k monitors in Linux, with X11 (and XFCE) for more than a decade. During the years, I have used a great number of text editors, e.g. vim, jedit, kate, geany, Visual Studio Code, neovim, etc. Only this year I have swit
502.
by adrian_b 4mo ago
While C and C++ have ugly syntaxes for "malloc", "std::shared_ptr" and the like, it is quite easy to mask that ugliness by using macros and writing thus only cleanly-looking programs, without any noise words and symbols
503.
by adrian_b 4mo ago
The full GLM-5.2 has 1.4 terabytes of data. Even in quantized form, reducing that size by 2 times or 4 times, a Raspberry Pi 5 could not run it otherwise than by reading the weights from an SSD. Even thus, I do not believe that a Raspberry
504.
by adrian_b 4mo ago
Hiding evidence is the normal manner of conducting business for a lot of company managers. Therefore the sentence seems unusually harsh in comparison with the lenience in other equivalent cases.
505.
by adrian_b 4mo ago
Most of my mini-PCs have been Intel NUCs (or more recently an ASUS NUC). I also had some Gigabyte and Zotac mini-PCs and a few others from less well-known vendors. IIRC almost all had one such 20 Gb/s USB Type C port, unless they had o
506.
by adrian_b 4mo ago
USA is hostile to the entire world, because the US actions already for several years, but especially during the last year, have caused global price rises in more and more product categories, starting with smartphones, then with SSDs, then w
507.
by adrian_b 4mo ago
By "hostile superpower" you mean USA? As an European, today I certainly classify USA as a "hostile superpower" because the actions from the last few years of both the US government and of certain big US companies have st
508.
by adrian_b 4mo ago
Yes, Capitalism is working by a mechanism where some people have access to huge fictitious resources, a.k.a. money, for whom they have never given anything equivalent in exchange, and they use those fictitious resources to outbid the other
509.
by adrian_b 4mo ago
No, you pay the most if you believe that you might get the most value out of it. Moreover, the AI companies have not bought anything with their own money, but with the money of naive investors who believe that their money will be used by
510.
by adrian_b 4mo ago
This phenomenon of "hardware companies are increasing their markets" is just a consequence of the fact that the memory market is now dominated by quasi-monopolies. Decades ago, when memory production still existed in many countrie

More ›