Linus Gabriel Sebastian & Linus Media Group / Linus Tech Tips - Narcissistic corporate shill YouTuber driving his media empire into the ground. KILL COUNT: 2

[Kuritan Deplorable]

Why is this a problem?

Because it ruins his totally cool manpower saving plans. He wanted 12 perfectly identical machines, all running separate tests so he could do twelve times the work at once. When the real way to deal with this variance issue, with the level of precision a youtube video might need, is to run all 12 machines on one test, and identify an average and outliers. That's both useful information for people who can't guarantee they got flawless silicon, and could potentially provide insight into unusual bottlenecks.

 

[cia-log]

I have friends like that. While I took the short straw on comp uni, I have friends who makes 6+ figures and live from paycheck to paycheck and don't see anything wrong with it

Yeah same. I don't care about Linus or any other gaming hardware channel because it's just marketing in the furtherance of this stupid culture I don't respect. By that I don't mean gaming, but specifically this niche of tech collectors who buy tons of dumb shit that they barely use like retro arcade cabinets or NAS hardware.

At roughly eleven Billion transistors for that cpu, a 0.25% deviation rate is frankly remarkable, not terrible. At that point your likely contending with unironically quantum bullshit effects and other "beyond the reasonable grasp to ever address" quirks of nanoscale engineering, across billions of failure points - give or take a few atoms here or there over that entire count and it can be significant. How is it that Linus, in his pursuit of the lab, consistently and universally proves himself more and more ignorant to the hardware that he claims he's going to have the expertise to rate at a industry certified level? I get that he doesn't have to know it personally, but does he not ever talk to any of these folks he has?

People got mad at me the last time I said this but I'll say it again, this type of testing is completely pointless for the average consumer. Every single person's home computing environment changes so rapidly within a given month that there's simply no way to establish a meaningful baseline with any sort of consistency. That's why it makes more sense to just ask people what their personal observations are with a given piece of hardware, at least that way you can aggregate real-world results and get an intuitive idea of what to expect. He basically admits this at the end of the video, which makes you wonder what the point of any of this is.

Okay, so, actually watching the video, this is even more fucking retarded than I thought it was. Linus acts like this shit is all a giant mystery, but you can measure the voltage each core is receiving, its reported clock speed, its actual clock speed, and then correlate that data with the performance data in gaming/productivity... And then maybe come to some kind of conclusion about what's "wrong" with the outliers. Ryzen CPUs have this cute little thing called "clock stretching" where the real-life clock speed is slowed, and voltage is increased, if possible underperformance/instability is detected.

This is all compounded by the fact that he's using Windows to do all this testing. I wouldn't be surprised if the majority of the variation he's seeing in performance is due to the OS, but in either case he's introducing a variable he admits he has no control over, can change at any time, and he has no meaningful way of quantifying its influence on test results. With the amount of time he's spent doing this channel he could've enrolled in a decent CS program and learned about this kind of thing.

 

[Jeff Q. Anime]

He wanted 12 perfectly identical machines

So... You know, other than your idea, which sounds really complicated, if you're testing something that shouldn't generally be CPU bound, but may be influenced by slightly faster boost clocks... Limit the max clock speed to something they can all handle. Wild idea.

Anyway, this is all old fucking news. GN already did this a few months ago, found variance up to 3%, and said everything is fine. Boost variance is just part of the consumer experience, for CPU tests. It's generally marginal. They've gotten really good at binning and automagically overclocking CPUs.

 

[jalopy]

Linus has created an update video on the money hole that makes no sense for his company, "The Lab".

Maybe he can publish ranges rather than fixed numbers to represent the differential. The fucking leaf.
He made an entire 20 minute video bitching that reality doesn't make it easier to post benchmark numbers.
The reference to car companies at the end makes no sense. Car companies pay to have their vehicles tested by third parties or the government itself to meet a standard. The fuel economy and crash safety information comes from government bodies in most countries.

They also thought some random soyface skit near the end was necessary

 

[CYKA RUSH B]

If you actually watched the video you'll know that the variance was 4%, which is a lot for a product that should be identical. Only through binning did they get that down to 0.25. You disingenuous retard.

fuck off linus

 

[i440BX]

Maybe he can publish ranges rather than fixed numbers to represent the differential. The fucking leaf.
He made an entire 20 minute video bitching that reality doesn't make it easier to post benchmark numbers.
The reference to car companies at the end makes no sense. Car companies pay to have their vehicles tested by third parties or the government itself to meet a standard. The fuel economy and crash safety information comes from government bodies in most countries.

They also thought some random soyface skit near the end was necessary
View attachment 5669636

I hate Soyfaces. Seriously. Don't they had parents to tell them it is not appropiate to do that. Its like they are injected with crack.

 

[Nabeshin]

how they were slightly different (by 0.25%)

While it was 4% it's an known fact that the margin of error that most companies uses for consumer/prosummer hardware to be 5%, so it's within margin.

That variance is why big companies like Passmark and others when doing testing for industrt, uses 100+ machines, so you can run the test on more than one machine and compensate for that variance by doing a generalized value with the proper margins.

 

[Administrator Flotsam]

I said this but I'll say it again, this type of testing is completely pointless for the average consumer

You mean testing for variance between copies of the same CPUs or do you mean detailed benchmarks in general? Because detailed benchmarks are relevant for plenty of reasons.

If you actually watched the video you'll know that the variance was 4%, which is a lot for a product that should be identical.

Because it ruins his totally cool manpower saving plans. He wanted 12 perfectly identical machines, all running separate tests so he could do twelve times the work at once. When the real way to deal with this variance issue, with the level of precision a youtube video might need, is to run all 12 machines on one test, and identify an average and outliers.

It's generally marginal.

As other users have pointed out, my only real question is why Linus did this test in the first place. Before AMD pulled themselves out of the depths of the FX series with Ryzen, both Intel and AMD were used to releasing CPUs with mad OC headroom and benchmarkers like GN and LTT would include OC'd performance in their charts as well because OCing made such a huge difference (though not enough for AMD FX to ever catch up with Intel). And in the charts you'd see variances in the frequencies the outlets were able to reach.

And even before that it was well known that no two pieces of silicon were the same; that's why competitive OCers search for the most pliable CPU and RAM stick copies from many of the same model.

Having lived and worked and benched through that period and GN having done a more detailed comparison already, why did Linus bother with a vid like this is all new stuff to him? If this video was pitched to Anthony he would've probably shot it down in a single sentence.

And I don't think Linus would care enough about variance even if it was higher because I don't think LTT tests with enough competence to spot wayward results even if they were to present themselves. Remember the chart in the GN expose where they showed one card having like double the performance of the next competitor.

 

[AnOminous]

If you actually watched the video you'll know that the variance was 4%, which is a lot for a product that should be identical. Only through binning did they get that down to 0.25. You disingenuous retard.

When haven't they done that? That's been standard practice as far as I've known. People who worked retail would always go through the CPUs to find the best of batch for their own overclocked system.

If you know a time when these products were in fact actually identical let me know when and where that was because I think it was a dream.

I hate Soyfaces. Seriously. Don't they had parents to tell them it is not appropiate to do that. Its like they are injected with crack.

Anyone who does this should be fitted with a shock collar. The voltage should increase every time they do this. Eventually if they're incurable it should jack up the voltage to the point their vitreous humor boils and shoots out their eyes while brains spew out of their ears and nostrils.

 

[TheRaptorJesus]

People who worked retail would always go through the CPUs to find the best of batch for their own overclocked system

To this point; cpus have become so consistent that this is really pointless. Silicon lottery shut down because there was no reason for them to bin and sell higher bins because everything is basically the same now. Modern overclocking isn't worth it and that's why undervolting is the new thing.

 

[AnOminous]

Modern overclocking isn't worth it and that's why undervolting is the new thing.

When you have variance like 0.25% or even 4% it's not worth it hunting for outliers, but it used to be dramatic, with some CPUs not even functioning stably at their rated speeds (early Cyrix was like this as well as being super hot). I haven't really had to mess with my GPU much for heat because usually if the fan even comes on, it means something's wrong.

Even when I was doing some very minor ETH mining the fan would only come on when it started, using almost default settings.

 

[cia-log]

You mean testing for variance between copies of the same CPUs or do you mean detailed benchmarks in general? Because detailed benchmarks are relevant for plenty of reasons.

Well that's just it, what is a "detailed" benchmark? Because I wouldn't consider his method of simply firing up a game and eyeballing the FPS to be rigorous benchmarking at all. Which obviously raises the question, why go through all this trouble of building twelve identical rigs and sticking them in the fridge if you're just going to have to admit that your results are mostly guesswork?

 

[Administrator Flotsam]

Well that's just it, what is a "detailed" benchmark? Because I wouldn't consider his method of simply firing up a game and eyeballing the FPS to be rigorous benchmarking at all. Which obviously raises the question, why go through all this trouble of building twelve identical rigs and sticking them in the fridge if you're just going to have to admit that your results are mostly guesswork?

Ok I thought you were talking about the whole scene, including people with proper testing methodology and sanity-checked charts. "Detailed" benchmarks as in performance numbers collected in controlled environments with reproducible results that give the reader / viewer info about if and how some products may be better suited to his needs than others.

Linus I'm sure is too far detached from both the budget gamer and the enthusiast who knows his stuff to think of anything beyond Mr. Beast-style spectacle vids but for hardware.

 

[BigGuyOS 6 : Jovial Janny]

If you actually watched the video you'll know that the variance was 4%, which is a lot for a product that should be identical. Only through binning did they get that down to 0.25. You disingenuous retard.

While it was 4% it's an known fact that the margin of error that most companies uses for consumer/prosummer hardware to be 5%, so it's within margin.

My bad, I did miss that that was with binning. I'll post the graphs and conclusions for posterity.

Spoiler: Gaming benchmarks

Gaming benchmarks (only the important ones)




(they explain why they think CS:GO is no longer a good benchmark, and drop it)

They then drop 'Corsola'

Spoiler: Productivity benchmarks

Productivity benchmarks (He mentions 4% variance in some benchmarks, but doesn't show it like in the overall gaming benchmarks)

Spoiler: Conclusions

Note that Euclidean distance is multi-dimensional measure of how closely the CPUs are to each other. It measures uniformity as a distance unit (not a %) (0= identical, no upper bound)


I missed this initially, but he makes his final conclusion used a binned subset.

 

[TheRaptorJesus]

When you have variance like 0.25% or even 4% it's not worth it hunting for outliers, but it used to be dramatic, with some CPUs not even functioning stably at their rated speeds (early Cyrix was like this as well as being super hot). I haven't really had to mess with my GPU much for heat because usually if the fan even comes on, it means something's wrong.

Even when I was doing some very minor ETH mining the fan would only come on when it started, using almost default settings.

I remember the barton days. But since they started boosting shit to the max automatically, manual overclocking has been dead.

I've never heard of a company that fucks with its entire network during business hours like LTT does. It seems like every other month they are taking it down and fucking with shit, and for some reason when people are in the office working.

 

[mandatorylurk]

I don't think I can watch videos with Jake anymore, it's like watching someone pretend to be retarded get away with being a dick and getting too handsy with women because "he doesn't know any better." Just the constant "let's spend money on this and this and this" is even getting to a point where a whore like Linus is having doubts.

You see Dan and the Labs guy doing it right, then they cut to Linus and Jake being fags and not telling Dan about just not turning the network back on, I would have a stern "Linus what the fuck" moment with him. Dan really needs to demand a raise or go somewhere else because I guarantee in a year or two there will be a video where he gets blamed for something when it will turn out to be Jake's gay tinkering with everything.

 

[The Great Milenko]

Linus has created an update video on the money hole that makes no sense for his company, "The Lab".

In it, he goes over how they attempted to speed up their automated testing

Updates on "The Lab" should come from Gary. I didn't want to watch the video, but simply having 12 computers isn't "automation."

 

[Kheapathic]

I've never heard of a company that fucks with its entire network during business hours like LTT does. It seems like every other month they are taking it down and fucking with shit, and for some reason when people are in the office working.

That's because no one does. You do all the required prep-work, send a mass email out saying "We're bringing shit down, SAVE EVERYTHING, NO EXCUSES!" and when people start leaving, you pull the plugs and start work. Then after you check everything, you might do a test, to make sure everyone stays online when you unplug one of your switches, just to make sure, because test/theory to live environment doesn't always work perfectly.

Though part of that is LTT needing their own operations (non-camera) facing team to do shit when no one is looking; but Linus can't have that, because everything is content... and no one wants to put in a full day of shooting videos, then do 12+ hours more of work and videos to fix shit.

 

[Kuritan Deplorable]

That's because no one does. You do all the required prep-work, send a mass email out saying "We're bringing shit down, SAVE EVERYTHING, NO EXCUSES!" and when people start leaving, you pull the plugs and start work. Then after you check everything, you might do a test, to make sure everyone stays online when you unplug one of your switches, just to make sure, because test/theory to live environment doesn't always work perfectly.

And this is why a good boss doesn't hound his network team for what they're doing every moment of business hours - Because half their work is occuring after 5pm, and if you want to actually keep them around without paying FAANG salaries, you don't question them 'making up' the hours they spent later.

Not that Linus ever knows what his people are doing at the best of times, though.

 

[Kheapathic]

And this is why a good boss doesn't hound his network team for what they're doing every moment of business hours - Because half their work is occuring after 5pm, and if you want to actually keep them around without paying FAANG salaries, you don't question them 'making up' the hours they spent later.

Not that Linus ever knows what his people are doing at the best of times, though.

Which is something else I'd do, going off my previous idea in the thread where he uses his massive resources to make a multi-hour actual tech tip stuff. Have Dan and the other guy do "How to build your network closet." How to choose a room, what the room should have, what to look for in a rack system, your hot/cold aisles, heavier stuff at the bottom, punchdown blocks, your TIA-568A/B, and whatever else. Instead we get guys who ordered shit that couldn't fit through the doorway normally, and then mostly Linus and Jake fucking around, and then suddenly Dan and the other dude are all "Yeah, so this is looking pretty, damn we're good." Give them a fucking budget and let them make something actually fucking useful for designing network shit for your at home hobbyists.

"Hi it's Dan and [OTHER GUY] and today we're gonna show you how to make a at home network stack using Ebayed SuperMicro gear."

"How much did I spend on this?"

"None of your fucking business Linus, we're being useful here."