Live data from Hacker News

Intel's New Chimera: Alder Lake

agner.org

191–200 of 253 posts

Re: Intel's New Chimera: Alder Lake

#191

Earlier quoted context omitted.

OK, you win! Just a thought, if embarrassingly parallel, buy a 2nd desktop and farm it out?

Alas you can't do that for every workload, and even for those where you can that brings extra complexity. On a single machine I can just replace an `.iter()` into a `.par_iter()` in my Rust programs and boom, I parallelized my workload across every core with only a single line change. It's not that simple with multiple machines. (I might do that one day, but I'm not there yet.)

> On a single machine I can just replace an `.iter()` into a `.par_iter()` in my Rust programs and boom, I parallelized my workload across every core with only a single line change.

If you’re using Rayon for anything you expect to benefit from a lot of cores, my experiences have been very negative in the past. Rayon is often extremely inefficient once you allow it to use more than a handful of cores.[0]

If your unit of work is absolutely massive, you might not be as affected by Rayon’s scheduler, but it makes me sad that this issue still hasn’t been resolved. My comment on that thread (documenting just one instance where I ran into this issue) was almost 4 years ago! And the issue itself is almost a year older than that. Sure, I could try to contribute a general solution, but it’s just easier to stop using Rayon.

I no longer consider Rayon to be an advantage for Rust. There are plenty of good ways to wire up a parallel work pool, but they do require more than a one line change.

[0]: https://github.com/rayon-rs/rayon/issues/394

Re: Intel's New Chimera: Alder Lake

#192
post #171

Earlier quoted context omitted.

Because that’s what I had access when I did the tests. I’ll do a series of tests on our cluster and post these. Will these be meaningful for you? The thing is, M1 on air is a relatively low clock speed CPU with passive cooling. 10200K is actively cooled, unlimited beast of its generation. The libraries I use are X86 optimized. Yet the code I’ve written is 25% faster on ARM. This is significant .

X86 optimized doesn't actually mean that much on a lot of workloads because the memory hierarchy completely dominates performance. Apple knew this so the M1 has an extremely fast memory system.

It’s a scientific code which manipulates a lot of relatively big matrices, which saturates the FPUs pretty well (I made the analysis back then).

Intel’s memory architecture can keep up with the load until I utilize all cores. When I start to use HT cores, neither cache, nor memory can keep up.

I do the both tests (Intel and Apple) both with physical core counts to keep it fair.

So it’s not some run of the mill code.

Re: Intel's New Chimera: Alder Lake

#193
post #168

Earlier quoted context omitted.

"Dunno why every comment implies there's some design gap that leads to M1 or AMD being more powerful or efficient." That is because there is a huge design gap. Intel architecture and design teams have always assumed on the inevitability of their own manufacturing superiority and process advantage and thus never had to really consider where to place transistors for the most benefit, or think really hard on better ways…

Intel's chip is more powerful than M1 on an older process node. So where is the design gap?

LOL yeah. More powerful, i.e. it consumes more power. Fun fact: the M1 big core caps itself around the same wattage where most Intel SKU consider the core to be below stock wattage. As in, in order to achieve your “more powerful” claim, it takes Intel cores 2-7x the watts, depending on the workload, to eek out a marginal performance uplift… or sometimes just to match the M1.

Nobody in CPU design cares about unconstrained power performance. Even at Intel, they never use the PL2 power level as the design landing zone. All modeling is done at much lower power target, even below PL1.

That is to say: unconstrained power performance is not an engineering effort, it is a marketing and SKU differentiation effort. Your “more powerful” statement is a matter of how much power Intel/Apple/AMD/whoever is willing to shove into their piece of silicon, which in turn dictates the form factors and customers they can try to satisfy.

Apple only needs a few watts per big core and still get the peak perf it needs, and put that chip in all the form factors it sells. Obviously if M1 sucked down the kind of power as any Intel chip, it cannot go into an iPad, or a fanless macbook air, without totally compromising performance.

Intel on the other hand needs grossly more watts to hit that same performance, and therefore is losing badly on form factors that care about power: laptops and servers. Some form factors/customers just don’t care, which is why Intel cooks up marketing targeting the gamer market, or the 8-pound laptop market. That’s all fine, just don’t pretend Intel is not getting absolutely hammered on the CPU technical specs for the money-making markets that matter.

BTW, the DIY gamer CPU market does not matter. The 8-pound laptop market does not matter.

Re: Intel's New Chimera: Alder Lake

#194

Earlier quoted context omitted.

That's a fair point, although some people do constantly multitask. I have ADHD and I'm capable of doing work while focussing entirely on two tasks, often I'm looking at one screen where I'm writing code while reading an article on another screen. Sometimes I feel like "reality" is too slow and I've already figured out the code for the unit test (or whatever I'm working on) in my head and I'm impatiently waiting for m…

Your ADHD medication is performance enhancing.

Absolutely. But you first have to factor in that my ADHD is performance degrading.

As an example: Mirtazapine is an antidepressant used to treat major depressive disorder, but that doesn't mean you can give it to an otherwise emotionally healthy person and it'll give the same effect delta to them.

If you put a crutch on a healthy leg, it doesn't suddenly become a better leg.

My medication allows me to experience the world without crippling anxiety and constant distraction, it just so happens that 30 years living with those issues means that a large portion of my attention and time was devoted to dealing with them.

So, yes, my ADHD medication is performance enhancing, but that's an incredibly reductive way to look at it. That doesn't mean that anyone can take it and perform better, in fact studies of studies have found that the results of stimulative medication for people without ADHD (such as those who are using it to try and improve cognition during studying) can be similarly triggered by placebo effects:

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3489818/

    As it remains unclear whether stimulant medication has the same effect on healthy individuals as for those with ADHD, it is possible that many reported effects of prescription stimulants in healthy individuals may stem from placebo effects. Looby and Earleywine (2011) examined whether placebo effects influence reports of subjective mood and cognitive performance among college students who endorsed several risk factors for prescription stimulant misuse (e.g., low grade point average, fraternity/sorority involvement, binge drinking). Interestingly, participants believed that they had better ability to focus and persevere, particularly for a sustained amount of time, when they expected to receive MPH (Looby and Earleywine 2011). This is similar to circumstances in which participants may engage in nonmedical-stimulant use to study or cram for extended hours. On the other hand, when experimental participants did not expect to receive MPH, their attention appeared disrupted resulting in inconsistent reaction times throughout the CPT.
I take my medication because I have a disability. I was diagnosed later in life and have a well tested / carefully controlled history of use with this medication where I'm frequently checked against diagnostic guidelines. In almost every metric of my life the disability had a negative effect, you can see a history of that here: https://news.ycombinator.com/item?id=28739990#28741432

If my medication is performance enhancing, then literally any antidepressant, antianxiety, anticoagulant, anti-inflammatory, antihistamine, antibiotic, etc., is also performance enhancing by the same thought. What we need to consider is the baseline it lifts you from and the place it takes you to.

It's probably worth mentioning that I'm in the UK, not the USA, where access to stimulant medication is a _lot_ stricter and it can be very difficult to get a diagnosis of ADHD. I had to go through several hour long sessions with specialised psychiatrists to receive the diagnosis after spending years being tested for other conditions like Mixed-Anxiety Disorder.

An excerpt from my remission notice is here showing the clinically recognised reduction in symptoms, it mentions that the decision was based on both subjective and objective justifications: https://i.imgur.com/RJQCAs6.png

The diagnostic screening questions for the ASRS1.1 are here showing the areas where I subjectively improved: https://add.org/wp-content/uploads/2015/03/adhd-questionnair...

On a side-note, your comment came across very harshly. It's very easy to make a short, pointed statement like that about a topic you aren't very familiar with, but you should consider how informed your statements actually are - if they're backed in data or long term subjective beliefs. Sometimes, brevity imparts a lot of information in few words, sometimes it just comes across as rude.

Re: Intel's New Chimera: Alder Lake

#195

Earlier quoted context omitted.

Some of us write software for a living. You give me two more cores, I'll use them.

I'm a dev too, my machine is a dual core haswell. It depends very much what you're doing of course (compiling linux kernel repeatedly perhaps) but if you have a little you can get a heck of a lot out of it with experience, although you surely will hit a wall sometime. But 32 cores not enough is just unimaginable for me.

ai. Our models train in a reasonable time (or we push them onto gpu) but doing the pipelines and etl when developing is a big time sink.

We could sit there and optimize the approach, but the time that would take and the complexity added, it would be cheaper to just buy more powerful computers.

For example, on reasonably new 8 core 64gb ram, 7gbs HDD machine, I optimized a calculation for one of the features down from 2-3 days to about 15 minutes. Throwing cores or ram or disk at this algo before would’ve got nowhere. Now htop glows like a Christmas tree when it’s running. This was costly to do but it was the only way.

We could talk about python being slow and blah blah blah use C but once again, whenever we do tricky stuff with code that then gets shared amongst a bunch of python devs I need to think about the holistic cost including maintenance, ease of changes, bug fixing, resource that actually know c, client appetite, what the production load is going to look like etc

And finally, at some point a client will just come along with another million rows of data

Re: Intel's New Chimera: Alder Lake

#196

Perhaps the easier option would be DRM software and anything that needs instructions that are only on a specific core type, stick to those cores. It might reinvolve changes to DRM software but looking at the current compromises, looks like a better deal. Of course Intel should have thought of this and didn't pick this route. I'd love to see the reasoning though.

Intel is in general heavily constrained by whatever Microsoft chooses to do with Windows. It's not like Apple where they can introduce a new hw feature and then immediately use it in the operating system. For example the M1 Macs have the big little architecture integrated with Grand Central Dispatch. Many, many interesting Hardware features have died in the Intel world because the Windows team simply did not add good…

I see your point.

Well it would be nice if Microsoft found a way to implement them and coordinate with Intel, with backwards compatibility of course (which is the key point of Windows).

Then this brings another question to my mind: if Intel is so contrained by Microsoft's decisions (which make sense) they should have been in a good relationship designing things that work nice together. Why did they design their CPU in a way that would break in Windows usage in the first place?

They must have much more R&D than saying "let's design this hybrid CPU and put a new instruction set on some cores and see if it works on Windows and DRM software else we disable those features" kind of approach.

Re: Intel's New Chimera: Alder Lake

#197

Earlier quoted context omitted.

Whenever I read stuff like this my inner optimiser starts wondering where the CPU is going and why - massive GC cycles, inefficient regexp stuff, poor algo choice, bad cache locality, etc etc. At the risk of ill manners but i must ask, have you looked at all this because I've got in some cases orders of magnitude speedup by popping the hood and instrumenting.

It's usually cheaper to buy more hardware than to burn engineering hours except at the very low scale (1-2 engineer) and at FAANG scale.

No. It depends. You weigh it up. And a little (cheap) nous can often beat a truckload of hardware.

Re: Intel's New Chimera: Alder Lake

#198
post #193

Earlier quoted context omitted.

Intel's chip is more powerful than M1 on an older process node. So where is the design gap?

LOL yeah. More powerful, i.e. it consumes more power. Fun fact: the M1 big core caps itself around the same wattage where most Intel SKU consider the core to be below stock wattage. As in, in order to achieve your “more powerful” claim, it takes Intel cores 2-7x the watts, depending on the workload, to eek out a marginal performance uplift… or sometimes just to match the M1. Nobody in CPU design cares about unconstra…

Ok, so when Intel and AMD get on 5nm equivalents and their chips are both more powerful and efficient than the 5nm Apple chips, you'll issue a mea culpa here, right?

The fanboyism runs so deep it's amazing. Let's compare PS5 to GameCube next and talk about Sony's great engineering advantage

Re: Intel's New Chimera: Alder Lake

#199
post #69

Earlier quoted context omitted.

It’s also not very intensive, especially since video decode usually runs on the GPU. (Or it should - browsers are not always very good about this but the tools are there, they just don’t want to pay licensing fees to use them.) I always cringe a bit when people insist they need 6 cores “so they can have discord and Spotify open while they game”. People don’t generally have a very good intuitive sense of just how litt…

> I always cringe a bit when people insist they need 6 cores “so they can have discord and Spotify open while they game”. I mean, they're not wrong. I have Discord sitting open at the login screen and it's burning 10% of a core. Run a few Electron apps and your core2duo suddenly becomes a core1uno. > Playing a video and writing a document Using what? VLC + notepad, or Chrome + YouTube + Google Docs? There's a huge di…

> I mean, they're not wrong. I have Discord sitting open at the login screen and it's burning 10% of a core. Run a few Electron apps and your core2duo suddenly becomes a core1uno.

That is what I said about “people look at the measurement at the desktop and freak out about single-digit utilization without even realizing their processor is sitting in the lowest frequency state”.

Is your processor even boosting all its cores when that 10% measurement is taken? In my experience usually it is not. Maybe different for Zen2/Zen3 but cpu measurement gets very complicated in general for those architectures since it can move between boost states so fast - it can boost so agilely that you can end up with garbage numbers because it’s moving between boost states faster than the task manager can enumerate the threads. It can move between boost states between the time you take the frequency measurement and the time you start enumerating the cpu load. This was a whole thing a couple years ago when people started looking at Zen2 and it turns out most third-party tools can’t handle this and only Ryzen Master gives you an accurate picture of what it’s doing.

In general it is such a small amount of cpu that it’s not worth caring about.

Also, typically while you game you’d have the overlay up and not the actual UI and that’s even faster - showing the login screen is likely slower than actually in game.

Load up a game and the cpu will be loaded up harder and will stay boosted and then take your measurement of discord.

People who are casually making observations about how intensive discord are never do that, it’s just “I see it using 6% at the desktop” [with the processor in the idle state rendering a full ui instead of just the overlay].

Post reply on HN