[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / r / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip / qa] [cm / hm / lgbt / y] [3 / aco / adv / an / asp / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / qst / sci / soc / sp / tg / toy / trv / tv / vp / wsg / wsr / x] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/g/ - Technology

[Advertise on 4chan]


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


File: 1593979187098.jpg (264 KB, 1300x698)
264 KB
264 KB JPG
If so why aren't we on 2000+ pins yet?
>>
File: 1524854485728.jpg (57 KB, 412x490)
57 KB
57 KB JPG
>>
>>78821366
Because we can't get the kind of PCB track density and layering required to plumb it all together. Tracks can only be so thin before signalling is impacted.
>>
If computers are all just 1s and 0s why doesn't the cpu just have 2 pins? 1 pin for 1 and 1 pin for 0 and then the cpu just sends whatver number it need to down the pin. Makes you think.
>>
File: 1597147343749.jpg (62 KB, 940x1024)
62 KB
62 KB JPG
>>78821366
Why do server chips have bigger dies and more pins yet are weaker than consumer enthusiast chips that are smaller and less pings?
>>
File: pentium2.jpg (17 KB, 480x360)
17 KB
17 KB JPG
What was wrong with cards Intel? Not damaged easily enough huh?
>>
>>78821366
You want 500w CPUs from intel?
>>
>>78821697
I can already envision it topping userbenchmark.
>>
>>78821680
You only get 270-odd pins and it was a design kludge because they could not reliably do cache and CPU on one die. Unless you want a foot long slot, it was not a practical form factor to run with.
>>
>>78821606
probably because they support more RAM, so more "lanes" to transfer bits.
>>
>>78821366
We are, and have been for the last 8 years
LGA2011, 2011 pins
LGA3647, 3647 pins
LGA4189, 4189 pins
SP3/TR4/TRX4, 4094 pins.

>>78821680
SECC was terrible but made sense when not everything was on the same die as the CPU core. There is many problems with using card's as a processor. Card creep being one of them. Manufacturing cost being another.
>>78821752
Whats pin pitch?
You can fit a lot more than 270 pins. Surely, they would not use the same slot 1 designs as from two decades ago anyways...
>>78821606
>>78821776
Not entirely, but yes. 4/6/8 channel CPU's require more pins. Not to mention QPI, HT, PCIX, power and whatever else... there is very little "unused" pins. The mechanical specifications can be found for all the sockets above. You can figure out the land mapping to them.
>>78821697
Theyre doing that already with 1200 pins
>>
>>78821901
>You can fit a lot more than 270 pins
Must be why we're still sitting at 288 pins for DDR5 DIMM's then. You can only go so wide with an edge connector pad before you risk alignment issues.
>>
>>78821996
>You can only go so wide
Hurp, only go so narrow.
>>
>>78821901
>LGA4189, 4189 pins
based intel
>>
>>78821901
wow! :o
>>
>>78821752

Intel got it right a little over a year after launching the P2 with the Mendocino core Celerons. I fuckin loved my Celery 300A. Overclocked to 450MHz it was ever so slightly faster than the big dog P2-450 back then. I was in Quakeworld heaven thanks to that processor.
>>
>>78821680
How would you do heatsinks on that kind of design?
>>
File: file.png (114 KB, 281x226)
114 KB
114 KB PNG
>>78821366
>why aren't we on 2000+ pins yet?
We are though. Its only the consumer trash platforms that are gimped deliberately.
>>78821399
Yes we can, in workstation and servers. Maybe not in overpriced chink gaymen boards tho.
>>78821680
too expensive to make
>>78821901
>Card creep
All boards I have have ways to mount the Pentium2 CPUs securely
>>
File: L_00007031.jpg (125 KB, 845x800)
125 KB
125 KB JPG
>>78822223
>>
>>78822232
Same as with any CPU. They were riveted to the PCB, no such thing as aftermarket cooling for these things.
>>
>>78822252
>>78822271
I just can't' see it scaling to the heatsinks we normally use today with fans and all. It would end up looking like a frankenstein design.
>>
>>78821589
Always going to need VCC and GND though.
>>
>>78821366
>>78821901
10000 pin cpu when?

I want an absolutely monstrously huge CPU
>>
>>78822271
>>78822286
The retail ones came with a riveted heatsink. The OEM ones came with a riveted metal heat spreader that you could stick a custom heatsink onto.
Heat really wasn't a big issue with these. The Athlon was hotter but apparently all of those could mount custom heatsinks.

The slot1 Pentium 3 was pretty easy to disassemble and with socket 370 we were back to fragile chips. But not as fragile as the later Athlons.
>>
>>78822366
https://www.youtube.com/watch?v=xQ3oJlt4GrI
>>
>>78821996
DDR is constrained to the same dimension AFAIK. And the pin pitch between DDR1-2 had decreases while 3&4 remained the same. One could anticipate the reason behind this being there doesn't need to be more at the present technology level. I have not looked over the spec for DDR5 though.

>>78822232
>expensive heavy card creep mitigation
Manufacturing cost has a huge play in this. No OEM is going to want to fork up all the money for such an interface that serves no functional improvement over PGA,LGA or BGA designs.
>>
>>78822437
Why don't we use this design on modern chips?
>>
>>78822635
Too expensive
>>
I had a spare computer in my basement with a 3600 that hadn't been turned on in like 6 months and when I tried pulling the cooler off the cpu got ripped out along with the fan and like three of the pins broke off.
I'm still fucking seething over that
>>
>>78822677
Preheat the CPU or give it a couple twists prior to lifting I guess.
No socket can prevent dumb apes in destroying shit. If you're not fucking up a simple ZIF socket, you're probably just fucking something else up.
>>
>>78822677
Should have heated up the paste first before trying
>>
>>78822688
>>78822690
yeah I read afterwards it's a fairly common issue on zen 2 for some reason if the paste is really cold, it just never even crossed my mind that something like this could happen.
Now I know at least lol, won't let it happen again and at least it happened with a 3600 and not my 5600X.
>>
>>78822703
Yeah I don't think their retaining clips are all that strong. 1331 pin PGA, you're kinda pushing it on how well they can retain the clip in there with a lot of pulling force on it.
AM1, AM2, AM3 all suffered from this problem as well. I started noticing it more with the higher density PGA sockets in the last 6 years or so.
>>
>try getting a slot1 pentium 1GHz with 100MHz FSB
>apparently they are super rare
>>
>>78822223
Easy. You design the card like M.2, but with the card-edge being on the horizontal edge. The CPU-Card is screwed into place with four standard machine screws, and you design your heatsinks to screw directly into the card mounting bracket, giving standard tolerances and measurements for mounting pressure and clearance.
>>
Why don't they just stack CPUs vertically so they can fit more more transistors and be faster?
>>
>>78821589
parallel address lines.
If you want 64-bit bus, you need 64 address and data lines minimum.
>>
>>78822775
Because the big issue is heat
https://en.wikipedia.org/wiki/Dark_silicon
>>
>>78822787
Just use graphene bro. Then stack them vertically for processor cube.
>>
>>78822809
The silicon industry complex will never allow graphene to take off.
>>
>>78822839
We need to gut big tech anyway.
>>
>>78821366
The more pins on a chip, the higher the complexity, therefore, the more error prone it is. Make things simple, and everything goes better.
>>
>>78822703
>>78822750
Its a common issue with essentially every heat-spreadered PGA CPU. Hell I've done it with pentium 4s.
>>
>>78821366
Technical reasons aside, more pins = thinner = easier to break, you have to insert them at some point.
>>
>>78822884
Absolutely agree, I have seen it done with a P4D before. But I've seen it more with the higher density PGA's like AM#. AM4/PGA1331 is really notorious for it. I assume the pitch is too fine for it form a solid grip.
>>
>>78822910
Its actually quite hard to break pins if you're not retarded. Both PGA and LGA are ZIF sockets.
I unfucked plenty of 2011 sockets where apes bent a bunch of pins. Managed to resurrect most.
>>
>>78823005
They're easy enough to break if you aren't careful. Anyway the point was there's an eventual limit to the number of pins you can reasonably have for a cpu.
>>
>>78823201
Sure. But the problem isn't with the pins breaking, its with the socket not being able to securely hold the package down to provide a stable signal/connection.
And they're not easy to break. Being careful is essential here, if you're a sloppy ape handed retard - just expect to pay twice or get someone to build it out for you.
>>
>1156 is better than 1151
>>
>>78821366
To add to this: Why are manufacturers so obsessed with making CPUs smaller instead of adding moar transistors and making it bigger and better?
>>
>>78823488
they are actually increasing in transistor count as they get smaller due to their size while also getting physically smaller.
>>
>>78823488
The failure rate increases.
>>
>>78823488
Bigger dies are harder to manufacture without defects, and they inherently use more power. Addtionally, bigger dies means more wafer space gets wasted.
>>
>>78821366
they change the pins so you have to buy a new motherboard as well
>>
File: 1604915636947.jpg (49 KB, 828x826)
49 KB
49 KB JPG
>>78821366
oh my fucking god what a cancerous, noob thread.

>>78821606
u must be fucking dumb as fuck. Why do server chips, because they are fucking meant for servers, not for playing ur stupid gay shit games.

Cores more cores motherfucker, ECC, memory cap and controllers, pci-e lanes, 24x7x365 for fucking years, cert fucking cations, enterprise bios, QPI fucking links - multi fucking cpu support.

old e5-2687Wv2 is pretty much on par with ryzen 1700 but has more pci-e lanes, bigger mem pool support and many fucking more... only con is TDP.
>>
>>78823833
>QPI fucking links - multi fucking cpu support.
True, the E5-2400 vs E5-2600 series has about 600 more pins and they add an extra channel, extra QPI bus.
>memory cap and controllers
False. The controller has been on the uncore (on die) since the Pentium 4 days.
>>
>>78824020

memory cap;
Sandy CPU's are capable of utilizing LRDIMM with 2 CPU's going up to 1.5TB of memory
Bigger controllers, supporting of up to 12 sticks per CPU for capacity reason, and total of up to 24 sticks per 2 CPU boards.

Consumer grade LGA1155 i7-2x00
Have only 2 mem channels, no support for ecc, up to 4 sticks of memory. Up to 4c, and lack of proper support for faster memory, and limit of 32G of ram.
>>
>>78824149
Thats just a perk of buffered ECC.
Consumer grade shit doesn't have buffered ECC support, amongst many other things that you did include. I was simply pointing out the fact the controller is NOT on the board.
>>
>>78821589
Not sure if trolling
>>
>>78824400
>>78824400
the mem controller is located on cpu on all sandy bridge cpu's. (haven't stated in my posts otherwise) - was describing feature sets differences in 1155 v 2011, server cpu's v consumer grade cpu's.


Additionally motherboards can have chips that expands the memory controller functionality further beyond by increasing io address pools.
>>
File: die.jpg (439 KB, 1818x542)
439 KB
439 KB JPG
>>78824512
reference on die.
>>
>>78822223
Ever heard of a GPU retard?
>>
>>78821366
i'm on 4094 pins
>>
>>78824512
Even westmere, fuck even woodcrest, include it on the die and those are many generations before sandy bridge.
>>
>>78824685
I was specifically refereeing to sandy, server cpu's were capable to compete with cpu's desktop cpus from 2017's.

But you cuck can't read fucking anything right? Like as if I was writing about any other fucking fuck CPU's than takin in example of difference between desktop and server cpu's for brain dead jokers.0

But seems you don't understand even that...
Please re-read fucking posts, eat shit, and die you little bitch.
>>
>>78824716
Ah, theres that autistic spirit. Well done. So tell more more how consumers can use server processors starting in 2017. How many FPS can I get on a Westmere-EP?
>>
>>78824739
Again you must be dumb as fuck... sandy e5-2687W, and e5-2687Wv2 stock can compete with 1700 on stock settings - almost 1:1

But i'll humor you with westmere/nahalem in single cpu config. (nahalem arch had big bottlnecks with qpi links)

If you take lets say x5660, overclock it using 1.4vtt for 200bclk you can still compete, and get quite decent fps. (likely around 5-10% over ryzen 1700)

// notes, the bigger, and 10c variants didn't overclock as well.

So eat you fucking shit you dumbass.
>>
>>78824860
You went about talking about enterprise shit, and now you're talking toddler shit like overclocking. I bet you put RGB's on servers. What are you doing with your life?
>>
>>78821366
That's alder lake, right? Rocket lake is still on LGA1200.
>>
just make it 2 big gold plates
>>
>>78822232
>LGA2011
Ohh, I remember why I hate intel now.
I bought into this "For enthusiasts, supported for
a long time" bullshit with 3930k and then those fags created LGA2011-0/1/2/3 which still required you to buy new MoBos for the newer CPUs.
>>
>>78825418
kek, classic intel
>>
https://www.youtube.com/watch?v=8JAWz9Da5og

You're welcome.
>>
>>78823552
Definitely. It's the only thing that explains LGA1156 going to LGA1155. Okay so you found out that you don't need that extra pin. Did you really have to make a new socket because of this? Just leave it NC and keep the same socket, that's what any sane engineer would do



Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.