thecompilr

Yeah, cloud is definitely an option too, but that might be a bit harder to pull off. Selling servers to AWS, GCC and Azure might be initially easier.

ARM64 servers.

This one is quite a low hanging fruit.

Apple has the most advanced ARMv8 based core, one that can rival Intel.

And the market is enormous, larger by far than iPad, or Mac.

Of course Apple being Apple they would ship entire boxes, and not just the CPUs.

That is probably the current distance. 1 AU is the mean distance.

I experienced nothing but exceptional service in Italy.

That makes me very sad. I believe nuclear power should be developed further, and California, as a state could find solution for the waste problem, without waiting for the government. After all it is the richest, and most innovative state in the US.

Nuclear could make human kind 100% green decades ago, if not for FUD surrounding it.

I love wind and solar, but they are still unable to supply stable power 24/7 (and I love what the Tesla batteries do to solve that eventually), and can't be used globally.

I don't have hard evidence, but

A) they are vulnerable to two out of the three attacks, which indicates they did not in fact consider speculative execution a danger

B) it is hard to believe that if they researched the topic at AMD, they wouldn't find this vulnerability in Intel processors a long time ago

AMD did things the "right" way, not because they understood the security implication, but because it was very hard to achieve such high level of speculation, and I think the performance gains didn't justify the effort for them.

You can either use lscpu, which is less accurate, or the best way is to check:

cat /sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_cur_freq

where instead of cpu0 you can write any core number, and it will give you the current frequency of that core in KHz.

VL - Extend AVX512 to operate on only 256-bit and 128-bits at a time. (vector length extension)

DQ - Extend AVX512 to Longs, Long Longs. (double word and quadword extension)

BW - Extend AVX512 to Bytes, Shorts. (byte and word extension)

Ads in print have a fixed cost, based on the size of the ad. But even printed ads target specific demographics. How many over 40 brides are there compared to 20-30 years old, that would read a wedding magazine is an extreme example of it.

Ads on TV have fixed cost, depending on the length of the commercial and the time it is aired. But even on TV you would target the show based on the demographics you prefer.

Ads on Facebook are pay per impression, so how can you make companies pay to show those ads to people who they know will probably not respond anyway?

I don't do any recruitment or hiring, and myself is well over 30.

I don't think discriminating during hiring is legal, or should be legal.

All I was saying that delivering hiring ads based on a certain demographic, that will bring you the most potential candidates, is a financially sound decision, as long as the hiring process as a whole is non-discriminating.

So the example is: "For example, an employer's reliance on word-of-mouth recruitment by its mostly Hispanic work force may violate the law if the result is that almost all new hires are Hispanic."

In this case the employers probably rely on many sources of candidates, and want to maximize each source. In the end the result is what matters.

I think that is perfectly legal, ethical and acceptable.

Those companies pay per impression, so they want to maximize the effect of the ad, and reach as many candidates as possible per each $ they pay FB.

This is just an ad. No one can stop you from applying to a job directly with the company, irregardless of your age, or any other demographics.

Just another sensational headline.

If that is an issue for you, you could try using the implementation I wrote for boringssl. It avoids SIMD multiplications altogether and only uses simple AVX2 instructions, so there is no slowdown (AFAICT) although it is not as fast as AVX512VL from OpenSSL in benchmarks.

You don't use FMA (or any multiplication) for SHA2. The throttling is a big issue.

Can't turbo as much is still a net loss of performance. Even on Platinum turbo frequency is almost 10% lower with AVX512 on single core, and 30% lower with all cores. So you really want the majority of your software to use AVX512 to gain net benefit. It takes the system 2ms to recover after an AVX512 instruction. But you are correct that the Silvers are way worse. I suspect Intel intentionally killed AVX512 performance on the Silvers. I tested power consumption, and there is no reason to reduce the frequency, except for the sake of it. The sad thing is there is no CPUID flag to distinguish good AVX512 from useless AVX512. Would really be better if they disabled it completely on Silver. The way it is now will just hurt adoption.

Are there really no anonymous benchmarks in the wild?

I am sorry, you are 100% correct. There are workloads where you can't beat Intel. Pure number crunching for example, where you can utilize AVX-512. I meant in the real world of (most) web servers in this case.

If you look at the C language then there are two compilers: gcc and clang. Clang uses LLVM as its back end, and is basically an Apple project. It is heavily optimized for arm64 (guess what: Apple uses arm64). LLVM is also the back end for Rust, Julia and many other languages. LuaJIT for example is pretty good on arm64 too.

This is the most meaningless comparison you could possibly make. By this logic Pentium 4 is also better, because it had 3.8 peak GHz, SMT and could do 4 uops/cycle.

In reality things don't work like that. First of all some of the uops can only be loads, others stores and other branches. Second, you have factors like branch prediction, cache latency, branch mis-prediction penalty that plays a huge role on performance.

I am yet to see a workload that can saturate 6 execution ports, even with SMT.

In the real world Centriq is better in performance/watt, and is even better in perfomance/thread that an SMT thread on an Intel Skylake.

What I would really like to see is a smaller version of the truck, and a collaboration with U-Haul. Now that has market value.

> If you make something that is performance critical, and in particular if it's concurrent, the normal rule is you avoid allocating on the heap at all.

This is a very synthetic workload, and all allocation were done by the standard library.

I recommend an app called Baby Tracker on iPhone (not sure if theres an android version). I recently found out it even has an apple watch companion app. Very convenient.

C language is as simple as it gets. The only concept people struggle with is pointers, but even that is a very simple concept. Teaching something that is simpler than C to computer scientists will breed very poor engineers.

> Jobs: Computers are actually pretty simple. We’re sitting here on a bench in this café [for this part of the Interview]. Let’s assume that you understood only the most rudimentary of directions and you asked how to find the rest room. I would have to describe it to you in very specific and precise instructions. I might say, “Scoot sideways two meters off the bench. Stand erect. Lift left foot. Bend left knee until it is horizontal. Extend left foot and shift weight 300 centimeters forward…” and on and on. If you could interpret all those instructions 100 times faster than any other person in this café, you would appear to be a magician: You could run over and grab a milk shake and bring it back and set it on the table and snap your fingers, and I’d think you made the milk shake appear, because it was so fast relative to my perception. That’s exactly what a computer does. It takes these very, very simple-minded instructions—“Go fetch a number, add it to this number, put the result there, perceive if it’s greater than this other number”—but executes them at a rate of, let’s say, 1,000,000 per second. At 1,000,000 per second, the results appear to be magic.

Wow, I don't know why but this explanation blows me away. So simple and yet so accurate.

Well, probably we would still see Apple develop their own CPUs. Intel just couldn't deliver on the low power promise.

So basically roomba with a driver?

Hear, hear

About time they open sourced it. Funny how it is Python according to github :)