/tech/ - Technology★

>$900 CPU vs $400 CPU

>still loses

>l-look out Intel

But luckily there's always some new tech on the horizon to excuse the unbroken string of failures since the Athlon.

https://boards.4chan.org/g/thread/67185781

Sorry to link cuckchan but I made too many posts to copy paste here I'm the fag going on about how fucked intel are from 2020 onwards

Funny how the shills there literally have no response other then muh games and ipc and even amd beats the shit out of Intel in that + streaming with better thermals and low avg fps

>>957072

Yeah 2006-2018 was fucking brutal for amd but they finally turned it around late this year

>assrpae

<psp

yes

>>957099

So what do? Hit the wall in ten years waiting for new lithographic 1nm or smaller? Move to risc/away from x86? /g/ cuckchan couldn't give me an anw

>>957102

Seeing as performance is not getting any better and freetards are moving in the direction of RISC where performance will never be as good as CISC but is now adequate enough for most shit. I think its safe to say the enthusiast market is just dead now. Do whatever you want because its not going to get better or worse in terms of performance

>>957103

Sweet looks like I'll be keeping my 2700x for 5 years like I did with my 4690 and i7 920 before that

Get comfy

>>957106

>Get comfy

I am. In many ways we're finally at with computers what the automobile industry has gotten to in the 90s. A 90s shitbox is still a solid daily driver today. A late 2010s computer is going to be a solid daily driver for the next 20 years at the very least now as well. IT was hyperbolic to say the enthusiast scene is dead, just stagnant. Its likely they will be the only ones driving hardware growth in the same way the enthusiast scene does for cars nowadays

>>957108

Kek funny u mention that my 2001 shit box has a insane aftermarket even today (based Ls1 t56 + gm v) and a stock gen 3 ecu so I know exactly what u mean fam.

Didn't even have to turbo the fucker only needed 350rwhp detuned to spin the rear tyres up to 4th gear because torque

Anyway back to tech only thing I see really taking off is multi core module gpus and h/w acc raytracing is already here with the rtx

202x is gonna be nuts dual quad Oct core gpus woo

>>957123

I didn't say that I meant back to pc tech learn to reading comp

Right now it seems we've reached the current physical barrier

http://www.kit.edu/kit/english/pi_2018_097_smallest-transistor-worldwide-switches-current-with-a-single-atom-in-solid-electrolyte.php

>1 atom wide transistor

It doesn't seem to be done on large scale right now, but there might be a last spur on performance/energy saving "soon"(TM)

>>957162

>atoms

>limit of miniaturization

We can go much smaller.

>>957072

>Single-Threaded

lol

Meanwhile I'm just using a core2duo still and patiently saving for a Talos II.

>>957199

Why are you bringing up Talos II in a thread about AMD/Intel?

>>957205

POWER10 is the future. IBM revival to 80's IBM soon.

>>957261

I really don't understand the IBM fangirlism among Linux users--you do realize they are "patent bullies" just like Apple?

>IBM's patent department is actively lobbying Europe to legalise software patents. They have invested millions in fighting example cases to leading European lawcourts such as the EPO's Technical Boards of Appeal and the German Federal Court in order to soften and eventually remove European restrictions on patenting software.

>They have also threatened European politicians that IBM might close down local facilities if software patents are not legalised in Europe.

>IBM has also prevented the US government from conducting studies on the value of software patents for the national economy. In the wake of the Opensource hype, IBM's rhetoric has become relatively moderate, but nonetheless it is supported by real pressure.

>IBM has acquired approximately 1000 European software patents whose legal status is currently unclear. Given the great number of software patents in IBM's hands, IBM is one of the few software companies who may have a genuine interest in software patentability.

>Once software patents become assertable in Europe, an IBM tax of several billion EUR per year may be levied on European software companies."

Go ahead and denigrate me, call me an x86 "botnet" "shill" or whatever--the point is that IBM is not worthy of your trust, _please_ stop fellating them. IBM just wants people who will work for free (Open Sores). Freetards going apoplectic about Apple and then jumping up and down like a child when IBM endorses the penguin are retarded.

>IBM hopes to make Linux what OS/2 failed to become--a unifying architecture for desktop to mainframe, clear of Microsoft dependence or competition. What anyone finds liberating about this eludes me.

>>957291

He's a retard, go easy on him. Next he'll call Intel/AMD Jews even though IBM also has an R&D Center in Israel

>>957195

Very little effectively uses AMD's MOAR CORES approach. Rendering does but semi-pro 3D rendering these days just rents time on a render farm. That they continue to get shit on in single core is a problem as it is what really matters to most people.

>>957291

IBM was tremendously influential re making Linux a serious business OS in the '90s, provided a huge amount of the earliest commercial kernel work, and spent a fortune over a decade defending Linux against Microsoft, Novell, and SCO's "Linux licenses". They earned their love more than anyone.

>>957291

>an IBM tax of several billion EUR per year may be levied on European software companies

I read that as "America gets to rape Europe" and as an American all I've got to say is: buckle up.

>>957291

>>957261

ibm is pajeet approved ibm has more indian employees than american. india will be the #1 superpower by 2025

>2950X with in 3% reach of Intelel's 8700K in CB ST

>2700X almost even with a 7700K

brb selling my Haslell i5 setup for a 2700X

Bought an AMD CPU thinking that inter-CCX latency wouldn't be an issue. Boy was I wrong. Monolithic CPUs are more expensive but are worth the extra money.

>>957629

>CCX latency

It'd be interesting to benchmark Factorio as it's possibly the only game optimized so deeply that it would be a big factor. Overclocking ram actually gives that game a huge boost as the inner loops are so tightly grinding arrays.

>>957312

They only did that to kill Sun and other UNIX competition.

Who cares? I'd rather have processors that I can actually trust.

>>957102

<1nm or smaller

Don't be stupid, a silicon atom is .3nm. You aren't going to build 3 atom thick walls that last for long enough to sell them to consumers.

>>957676

Then don't make them out of atoms.

Do more basic research on quarks. There's no reason to stop at atom scale.

>>957687

On those scale uncertainty principle starts becoming a real issue

>>957334

>india will be the #1 superpoo by 2025

India will poo in the loo by 2225

>>957708

>superpoo

>not Pooperpower

missed opportunity fam

>>957177

Sure there's quantum computing next, but so far the current state doesn't seem to be stable/usable yet.

There's a good chance it'll be usable somewhere in the future since research is still being done in quantum physics.

>>957719

Since you posted a different technology, you just undermarked your own point. Tubes were unscalable for switching, and thus were quickly obsoleted in favor of transistors. (and soon, even transistors were obsoleted for field-effect transistors)

Now, for our trust vlsi mos gates, the width of the latice at room tempertures is the limit. The diamond lattice will not exist for a thinner silicon waffer, which makes it impossible to suspend boron and phosphorus since the lattice isn't even there. This is the end of the line for reducing the width of CMOS.

>>957743

>a different technology

lol

We've been through gears, vacuum tubes, transistors, and silicon transistors in the past 100 years to solve the same problem. Next will be subatomic. The fatalism about there being no smaller we can go is ridiculous as we're fully aware today of smaller constructs than atoms. It will take research but so did everything else. Man up.

>>957748

>deterministic machines with subatomic building blocks

Good luck with that.

>>957749

Nigger, everything we have is made of those building blocks. There are surely many fun things we can do with them.

there is no Intel 7nm

its 10nm thats BROKEN and LOW YIELD

and delayed for 3 years

what's all this shit about IPC? is that when i run a "frontend" for gnupg and it tries to talk to gnupg in english and then it fails to verify a single message and my OS crashes?

>>957046 (OP)

this thread is about the unprecedented level of botnet they will have in the next generations right?

>>957838

Are you implying Intelel is any better than AyyMD? I'd take AMD above Intel nowadays because 8 cores, soldered IHS and less """"""exploits""""""

>>957846

Both are shit, it's just that Intel is the better buy for anyone doing realtime work.

>>957838

>Don't buy AMD

>I can feel the microseconds of latency through the 10s of milliseconds on in my input devices and 10s of milliseconds on my monitors

There's always some sort of impossible to prove, "just trust me bro" thing wrong with AMD when they have a competitive product.

I still remember Anandtech telling readers that cores counts that weren't power of two hurt performance and caused weird latency issues when the 6 core Phenom came out. And then Intel releases a 6 core desktop part and the power of two no longer matters.

>>957851

This is a troll or marketing thread, the only benchmark they use is single thread rendering, which has no real world use case and negates the benefit of any additional cores. Not to mention they're all overclocked. Most people fall for this type of bullshit benchmark, you can tell by some people in this thread thinking that this benchmark means anything beyond single thread performance. This benchmark could easily be won by an unlocked pentium, doesn't mean that's the fastest CPU on the list.

>>957099

>for raw performance CISC is just better than RISC

Isn't IBM Power a RISC architecture?

>>957851

>Intel may score better perfs than AMD

Is that even true any more, with all the mitigations Intel's had to do?

>>957308

yes, because people only run one (1) program on their computer at a time and OS schedulers that split work across cpu cores don't exist :^)

goddamn you Intlel shills are getting desperate lmao

Intel(R) Meltdown(R) and the new Spectre-NG exploits got you working overtime to salvage your reputation, kek

>>957854

When there are hundreds/thousands of interrupts, those microseconds become milliseconds. So yes, system responsiveness does change.

>>957978

Is your mouse polling at 1Ghz or something?

>>957913

>because people only run one (1) program on their computer

Basically, yeah. How often does a desktop have two programs simultaneously pushing the CPU as hard as they can over a few seconds or more? Roughly never. What do you hope to improve, the user experience of systems infected with bitcoin malware?

>>958210

>conviniently forgetting about gamers who also multitask by having a browser and/or chat program open

Face it man, your 4c4t quad core is obsolete if you do anything more than gaming at the same time.

>>958252

>implying that browser or chat program is "pushing the CPU as hard as they can over a few seconds or more" while in the background

Shit's irrelevant. Benchmarks show that the desktop experience is all about single core performance. And I'm sure if AMD ever surpasses Intel in single core performance you'll be repeating my words to me.

>>958256

"benchmarks"

Not representative of the real world most of the time.

>>958259

Because real world users are rendering a project in blender while they play Witcher 3, right?

>>958276

No, real users are checking emails while writing some shit in word at the same time as they download shitloads of documents from the internet, opening window after window after window without ever closing it again to get faster access meanwhile there's the spyware the boss installed on the machines to keep an eye on the employees, or the remote desktop software from the IT guy who needs it to fix the users fuckups as quickly as he can.

>>957303

IBM is inherently anti-jewish because they were instumental during the holocaust

try again

intel is only for gaymers and literal uphill gardeners

>>958282

Unless they have multiple programs that cap out the CPU for sustained periods it doesn't matter. Desktop users only have one program at a time that does that. Even the "power user" types don't have more than one. It's not until you hit developers and workstations that you get those workloads.

>>958292

I don't know much about gaymen workloads, but I upgraded my dead Xeon E3-1265Lv2 earlier this year. My options were basically a similarly priced system to the price I paid for the old one 5 years ago or however long it was, which was basically no better in any meaningful way, or something interesting. So I upgraded it to an EPYC 7351p.

Gotta say, AMD is on to something here. It's an order of magnitude cheaper than the same power from Intel. Eight channel memory and it doesn't ever seem to get bottlenecked on the RAM, and when saturating all the threads it just powers through it at 60-80% of single core performance per thread like a beast.

For a dev system/server, it's truly a modern marvel. I can see it giving Intel some serious trouble in the server hosting market, even though the single core performance is a tad lower than several generations old Intel chips.

There's also just something magical about compiling a fairly hefty linux config in 40 seconds.

>>958206

1 MHz? Yes, and my graphics card, and my audio device, and my keyboard, and I/O, etc. Optimizing the OS and hardware for low latency starts with the CPU, and I've concluded that Ryzen can't be used in realtime systems. Sure, workstations won't be affected, but audio and gaming won't benefit from more cores if they are extremely far apart.

>>958307

>mouse

>keyboard

>1 MHz polling

Yeahhh no. You're way off.

The others aren't causing frequent interrupts, either. Check /proc/interrupts. A system spamming you with interrupts is broken.

>>958309

Well, I'm on a Windows system. In the span of 10 seconds I got 6169 DPCs and 18102 ISRs. So 617 and 1810 per second respectively, on average. That's with audio playing and me not moving the mouse.

Now, with a game in the background and me moving the mouse, I got 93068 DPCs and 34749 ISRs over 20 seconds. So 4653 DPCs and 1737 ISRs per second. That's a lot, and that's where the microseconds become milliseconds. In gaming, it is critical that latency stays low for smooth frame delivery and responsive input. For audio production, that means being able to monitor audio. I don't feel like restarting my system and enabling a CCX, but usually the average interrupt to process latency is 1.25-2x larger. Not worth the extra cores.

>>958316

>4653 DPCs

Irrelevant.

>1737 ISRs per second

>That's a lot

No. It's also considerably less than a million per second as in >>958307

>>958316

>that's where the microseconds become milliseconds

Are you trying to count total time spent in a second? Why? What matters for latency is how long the longest interrupt takes to service and if the rate was so high that interrupts were arriving while interrupts were disabled as that will also result in latency. With something pathetically low like 1737 per second that isn't happening, and anything slowing down a single interrupt is going to be a difference of nanoseconds.

>>957308

>most people

who cares, I just want to install gentoo faster

>>958276

Real /tech/ users are. The rest can die in a gas chamber as far as I'm concerned.

When you guys say real-time, do you actually just mean low latency? Because I have a hard time imagining anyone doing actual real-time stuff would use Windows and a modern x86 processor and not, you know, an RTOS on an architecture either built for it or at least analyzed well enough that guarantees can actually be made.

>>958391

Good luck doing that on Windows without either manually setting thread affinity of both programs or suffering stutter.

>>958395

Yes low-latency.

>>958386

>Irrelevant.

why?

>million per second as in

where do you see million per second?

>Are you trying to count total time spent in a second? Why?

Yes, the average interrupt to process latency significantly declines with a CCX disabled. The average can be multiplied by the number of DPCs to give an estimate of driver latency.

>>958452

>why?

DPCs don't run while masked, they're like Linux's tasklets. They aren't contributing to interrupt latency and the number of them is meaningless in this context.

>where do you see million per second?

I asked if you think your hardware is generating a billion interrupts per second >>958206 and you thought it generates a million interrupts per second >>958307 . I'm well aware the real number is usually in the thousands on a desktop.

>Yes, the average interrupt to process latency significantly declines with a CCX disabled

I doubt it affects interrupt latency at all as it shouldn't need to communicate with the other cores. Even if it did, you'd be talking about a difference of nanoseconds. It also wouldn't matter how many interrupts per second you have as the latency is independent unless you have so many that they're getting blocked in which case your computer is likely unresponsive.

>>958460

>I asked if you think your hardware is generating a billion interrupts per second >>958206 and you thought it generates a million interrupts per second

Oops, I meant 1 KHz mouse polling.

>I doubt it affects interrupt latency at all as it shouldn't need to communicate with the other cores.

Well, that's not the case if you look at the interrupt to process here >>957838

>>958463

It shows there's possibly ~200 nanoseconds of difference in the handling of the interrupt before it's offloaded. That's so small that I'm sure the result will differ each time you reset it and recheck those numbers. The rest is purely inside of Windows, and at most about 1 microsecond of difference. The "highest" results will be heavily affected by luck as DPCs can be interrupted (same as tasklets) and are irrelevant in this context.

1 microsecond of difference, if there actually is one (reboot, play for a bit, recheck the numbers, and see how stable they are), is way below what you can perceive and way below other sources of latency like the scheduler.

>>957866

Exactly

>>958479

>That's so small that I'm sure the result will differ each time you reset it and recheck those numbers.

They do differ. I cycle the test to try to get the lowest result, which is what I work off of. The highs vary too wildly, whereas the lows I can get consistently.

>and at most about 1 microsecond of difference.

Actually, I was able to shave off about 2us by optimizing Windows and UEFI settings. There are many guides out there. Most of the savings come from increasing the timer resolution to .5ms, and disabling the power saving bullshit. On Windows 8+ they added tickless, which can be disabled.

I'm not really arguing with you that this should be imperceptible, but for me it is perceptible. The system is a lot more responsive on one CCX than two. The average lowest DPC latency on 1 CCX is .5, where on 2 CCXs it's .7 microseconds. The interrupt to process varies a lot more, from 1.1us to 2.3, and on 2 CCXs it's 1.6-3.1us, usually in the 1.6us range for 1 CCX and 2.7us for 2 CCXs. It makes tracking targets a lot easier.

Obviously microseconds aren't perceptible, or even a few milliseconds to most, but somewhere down the line they add up. Even on a blind test I would notice the difference between 500 and 1000Hz mouse polling, as far as latency goes (stability issues are another thing).

>>958597

>I would notice the difference between A and 2 times A

No fucking shit, how big is the fraction of lost time?

>>958598

Why are you so hostile? It's roughly 6/10 time lost or 1.67x faster.

>>958597

>but for me it is perceptible

Ask someone to help you test if you can tell if a system has it enabled or not.

>>957199

>Meanwhile I'm just using a core2duo still

same here, fam. Just got myself a fresh SSD since my old one still works but is worn out (6 years old). Ubuntu + xfce is running amazinlgy fast on my c2d machine. Wouldn't even consider the ~300mb RAM after boot too bloated. Seems like I will get comfy again after considering the new ThinkPad A485.

>>958601

Would be nice if I knew someone with a slo mo camera, 10000 FPS for the ease of math.

>>958398

>Real /tech/ users

>Windows

>>957199

>>957261

Tell me more about this Talos shit. Is it really free from proprietary code?

>>962608

It's free from all code as no one uses it.

>>962608

Remove the ethernet controller and yes, it is completely free of any sort of propietary code. Its fucking awesome.

>>962617

>as no one uses it

unlike ur mom, whos seen plenty of DNA code in her time

<https://youtu.be/UxejE5JZNvI?t=7

<Tor users can't post images

>>957308

>Very little

Absolute kek

>>957102

Perhaps a switch to magnonics instead of electronics if anyone manages to make something that can facilitate them without blowing up the second it gets exposed to oxygen.

>>958282

You're describing a RAM problem, not a thread problem. All of those tasks use very little CPU time, they benefit from a couple threads but there is no normal situation where you actually need 16 threads. They are quite limited by RAM though, particularly the browser tabs.