kolbusa

Also try playing with the can and the trash bin.

Not sure why the article does not reference the following paper which is a must read for anyone working with floating point: https://docs.oracle.com/cd/E19957-01/806-3568/ncg_goldberg.h... (original: https://www.itu.dk/~sestoft/bachelor/IEEE754_article.pdf).

I am visiting Boston and I cannot stop wondering why SF is not like that. The city feels so much more livable almost everywhere I went. I'm sure there are shady parts, but every time I need to go to SF for some reason I get really depressed.

Copying is usually not necessary. Often times you can swap data and/or shape arguments and get a transposed result out. While it is true that Fortran BLAS only supports col-major, CBLAS supports both row- and col-major. Internally, all the libraries I worked on use col-major, but that is just a convention.

Those who do have same qualifications are still payed much much less. Anecdotally, when Intel still had offices in Russia our salaries including stock awards were 3x or lower than those of the US personnel. Not because of difference in qualifications but because of the labor market (the salaries were very good considering other opportunities in my home city).

Nitpick... This paragraph is somewhat confusing. I think it is worded incorrectly:

> Let's simplify the problem and implicitly transpose the matrix multiplication. Both A and B (our inputs) will have K (our reduction dimension) as the leading dimension. This doesn't really matter much in practice, but it simplifies our code a lot.

The code is

  C[n * 16 + m] += A[k * 16 + m] * B[k * 16 + n];

Which means that actually *m* is the leading dimension of A with stride 16, and for B it is *n* with stride 16.

In my experience, EIGEN's threadpool is decent. But OpenMP (edit: I mean Intel's implementation donated to LLVM) is often faster especially if threads are allowed to be affinitized to HW processors. Another promise of OpenMP that is not made by various threadpools is cooperative execution; in threadpools tasks are usually independent.

However, if any part of an app uses affinitized threads, the whole app needs to be using the same thread pool, as otherwise perf will go down. In this regard, OpenMP is less composable.

They support esim.

It will be very relevant in any field that does performance testing. And that includes compilers.

My guess would be that this is because of thread migration.

(After reading the TFA: that's what Agner says right there in the 4th paragraph.)

(After re-reading the comment: I guess that the OS changes would need to be extensive with little to no benefit: running AVX2 on all cores will likely be faster than running 2 P cores with AVX512. The only thing that is really affected is the code that could use AVX512_FP16, but I doubt there's a lot of it outside of Intel.)

This is completely off topic here, but the first promotion of Int64 -> Float64 is also iffy as not every Int64 is exactly representable. The Int64 -> Float16 promotion also looks weird to me because Float16 range is so small

It really depends on the domain. HPC is more frequently SOA (think CSR sparse matrices), while AOS may make more sense in other cases.

I think that this 'incarnation' of Novaya is also not based in Russia.

You have Russia actively trying to overtake Ukraine and yet somehow lack of military support for Ukraine should help negotiations? That seems to be true only if you want Russia to prevail. How can you force the sides to negotiate? There’s very little leverage.

> And - speaking as an American - of course the American people are morally responsible if their government does something similarly horrible as what Russia is doing in Ukraine. That very obviously also goes for what happened in Vietnam. The US Government and its presidents are the product of the American culture, which the American people are responsible for.

It's not an 'if'. The US did a lot of stuff on par with what Putin is doing now in Ukraine. The people and government of the US never were held responsible. Yes, they went through the motions of getting the UN approval, etc., but that does not make the actions morally justifiable. And this is what Russian propaganda uses to justify its actions.

I don't like pinning Putins actions on ordinary citizens. Putin is a dictator. In 2012 there was a big rally against him, and it was suppressed pretty violently, and since then there were no attempts on the same scale. Look at how Belorussians tried to topple Lukashenko. They made a much stronger attempt, but sill it did not work out. Are they morally responsible for the Belarus participation in this war? Are the people like me who moved out are responsible? When does that moral responsibility start? 2000s or 2010s?

Sure, there is a large chunk of population that supports Putin. They maybe are morally responsible for supporting the war. But I don't think they will actually be swayed by economic sanctions. Their culture won't change. They will be happy to severe ties with 'rotten West' -- they will feel like they are soldiers of the economic war. What will happen, I am afraid, is Russia turning into a second North Korea (or Venezuela).

> If the other empires of Europe changed (Germany, Spain, Britain, France, etc) - their people changed those empire, conquest obsessed cultures - then Russia too can and must change. Russia is the last major power in Europe still clinging to those decrepit, backwards ways. Russia must give up the notion of empire culturally and that means its people must fully abandon all the related ideas that propel and sustain that ideology (which endlessly spawns monsters like Putin (who is just another Russian Tsar type)). Until the people of Russia change their beliefs, the authoritarianism will just keep repeating.

In my rather uneducated opinion, the culture did not go away. The US is an empire. It is built differently, but it effectively is. China is or is becoming one. European culture is dominating the world in many subtile and not so subtile ways. Russia wants to be an empire, but it fails to recognize that empires are being built differently now, and tries to build a 19-th/20-th century one.

Disclaimer: I am from Russia, and live in the US. Have extended family in Russia. Unfortunately, some of them are brainwashed by the propaganda. Some are unable to leave as their whole life is to work in government-funded research (thankfully not military in any way). Fuck Putin and fuck the war. My best wishes to Ukrain and its people -- you will will regardless of how this war turns out.

PS. Apologies for the long rant...

Thank you and all the OVD Info for your work.

They have offices in Ukraine. I got this e-mail as well. This is collateral damage. You can get an another domain. Many people won't get their life back. Fuck Putin, fuck the war.

> Norton CD

Out of curiosity, I started googling, and this seems like a good place to read about ncd and more: http://www.softpanorama.org/OFM/norton_change_directory_clon....

Probably, if killed responsibly. The scene in the movie is just animal cruelty in my opinion (edited my comment for clarity).

Tarkovsky is a genius. The original novel for Stalker is a must read (esp. if you can read Russian). But I cannot watch some of his movies after I learned that he _cruelly_ killed a horse in https://en.wikipedia.org/wiki/Andrei_Rublev_(film)#Depiction.... I cannot justify this for the sake of art.

The setup I saw is: there is an IP plan that documents whatever 3rd part IP you are using in your product (open-source or not). Someone has to sign-off on that plan, and sometimes developers do self-attestation that they have not deviated from it. Additionally, the binaries are scanned for certain things to avoid escapes of pre-release information, etc.

I don't think so. The new ISA that is in the SPR is mostly about deep learning: it supports int8 and bfloat16 (https://fuse.wikichip.org/news/3600/the-x86-advanced-matrix-...). You can emulate higher precision using bfloat16 (https://arxiv.org/abs/1904.06376), but I have not seen this used in the wild.

There is a way to train with mixed fp16/fp32 precision.

Skip the dog part, but please please please try put as much details in the first IM as opposed to starting with 'hey'.

I had in mind a scenario in which a cloud provider would have access to everything but would be disabling things and selling those nodes cheaper (or, rather, sell nodes with extra features for more), while maintaining a uniform fleet of nodes.

Am I completely misreading this? Theoretically this is not only rent seeking but also ability for cloud provides to be able to configure the same piece of silicon with different features based on demand / etc.

C++ is a mess from the backwards compatibility and portability point of view. Just try shipping a binary C++ library on Linux.

Authorized repairers often don't do repairs on component level, and sometimes ship your device to some place else to have the whole assembly replaced. Which is always much more expensive.

Intel contributes optimizations to gcc and llvm.

And to add the insult to the injury, the genuine parts are not available, which screws 3rd party repair shops even more.