AMD Zen CPUs will have PGA mounting systems

[totjup5]

Good. I like PGA better.

[flibberdipper]

2 hours ago, Misanthrope said:

Your exaggeration is extremely disingenuous: it is far more likely for you to accidentally bend pins on the cpu by simply removing it from the box, switching to another rig, etc than it is for me to basically maneuver a mobo socket first into the corner of a case. 

Given that I store multiple AMD and pre-LGA CPUs in a box without any protection, and I have legitimately thrown the box across my room with no casualties, I think my point still stands.

 

And given just how much I handle PGA CPUs comapred to your barely-if-any (more than likely), I think that only re-enforces my answer.

[dalekphalm]

36 minutes ago, cayphed said:

I'm just saying, that'll impress me.

Also I'd like to see VM in VM's completely conflict free, now wouldn't that be nice?

And I completely agree with the RAM speeds, but I'm not sure even HMB or HMC could get us there either.

My whole reason for even entertaining the argument is simply because I just don't believe ANYONE is even trying to come up with something new, ground breaking, or even useful...

I'm curious, can you come up with any use case scenarios for multi-layered VM setups? Why would you need to run a VM inside a VM, instead of say, beside the VM in the same Host?

[deviant88]

when you have bent PINS:

[kameshss]

1 hour ago, i_build_nanosuits said:

Can someone with a 4K monitor count how many pins please?!

 

1331.

 

Think im lying?

 

check it urself.

[pyrojoe34]

Random thought regarding pin density/count:

 

Why has no one tried to implement a multi-layered pin system to increase connections without increasing physical pins. Something like a headphone jack where you have multiple connections on one pin by embedding pins within each other separated by insulating material. I'm sure that would increase the price of production a bit but you could double, triple, or quadruple the number of connections without actually increasing the number of pins. Seems like they'll eventually reach a point where they cannot realistically add more pins to a socket (just look at how may pins Skylake-e has) without exclusively selling the CPUs as permanently soldered to the motherboard.

[Curufinwe_wins]

12 minutes ago, pyrojoe34 said:

Random thought regarding pin density/count:

 

Why has no one tried to implement a multi-layered pin system to increase connections without increasing physical pins. Something like a headphone jack where you have multiple connections on one pin by embedding pins within each other separated by insulating material. I'm sure that would increase the price of production a bit but you could double, triple, or quadruple the number of connections without actually increasing the number of pins. Seems like they'll eventually reach a point where they cannot realistically add more pins to a socket (just look at how may pins Skylake-e has) without exclusively selling the CPUs as permanently soldered to the motherboard.

ANY switching mechanism within the pins would contribute to increased latency and interference cross talk. Not things you want to deal with.

[pyrojoe34]

11 minutes ago, Curufinwe_wins said:

ANY switching mechanism within the pins would contribute to increased latency and interference cross talk. Not things you want to deal with.

No it wouldn't require any switching, it would work just like a headphone jack, each layer would be constantly connected and able to transmit data. The connection would just be layered rather than a single "2D" grid.

[Curufinwe_wins]

2 minutes ago, pyrojoe34 said:

No it wouldn't require any switching, it would work just like a headphone jack, each layer would be constantly connected and able to transmit data. The connection would just be layered rather than a single "2D" grid.

Cross talk would still be a huge issue.

[dalekphalm]

9 minutes ago, Curufinwe_wins said:

Cross talk would still be a huge issue.

"could", not would. I'm assuming Intel/AMD Engineers have already examined such a possibility, but the reasons why it hasn't been tried yet may have nothing to do with crosstalk. It might simply be because it's more expensive or not a limitation yet, etc.

[Curufinwe_wins]

3 minutes ago, dalekphalm said:

"could", not would. I'm assuming Intel/AMD Engineers have already examined such a possibility, but the reasons why it hasn't been tried yet may have nothing to do with crosstalk. It might simply be because it's more expensive or not a limitation yet, etc.

no... it WOULD be a huge issue. Look at all the insulation added to cat cables in the quest for 600Mhz coherence. Now consider transmissions 5-10x more frequent than that, in a smaller space, with thinner conductors, with 0! tolerance for bit rot.

 

 

[JoeOCD]

Tibia honest, I don't care.

[dalekphalm]

1 minute ago, Curufinwe_wins said:

no... it WOULD be a huge issue. Look at all the insulation added to cat cables in the quest for 600Mhz coherence. Now consider transmissions 5-10x more frequent than that, in a smaller space, with thinner wire, with 0! tolerance for bit rot.

Again, since I am not a CPU engineer, I cannot just take your word for that. Ethernet Cables and CPU Socket Pins are significantly different. Ethernet Cables also need to run literally 20,000 times the distance over a Socket Pin (estimated length of a PGA pin = 5mm, maximum length of Ethernet Cable = 100m without crosstalk being an issue).

 

So no, I cannot say it would be an issue. If you have some sources that back up that, feel free to post them. However, I'm not even sure anyone's even tried a multi-barrel pin before, so I would be surprised if there was any data available on the subject.

 

As I said, crosstalk very well might be an issue - or maybe it isn't. If you do have a source to backup your claim, I would love to read it, and I don't mind being wrong when presented with evidence.

[Misanthrope]

1 hour ago, tmcclelland455 said:

Given that I store multiple AMD and pre-LGA CPUs in a box without any protection, and I have legitimately thrown the box across my room with no casualties, I think my point still stands.

 

And given just how much I handle PGA CPUs comapred to your barely-if-any (more than likely), I think that only re-enforces my answer.

Anectdotal points never stand to rational minded people so bye. 

[dalekphalm]

1 minute ago, Misanthrope said:

Anectdotal points never stand to rational minded people so bye. 

To be fair, no one countered his claims with a verifiable source, so the group saying "PGA Pins are durable as fuck" and the group saying "PGA Pins literally murdered my dog" are basically in the same group.

 

It's entirely anecdotal evidence all the way around. I'd be interested in seeing some RMA stats about bent CPU PGA pins vs bent Motherboard LGA pins, and the rate of occurrence... but I'm not holding my breath on those stats ever appearing.

[Curufinwe_wins]

2 minutes ago, dalekphalm said:

Again, since I am not a CPU engineer, I cannot just take your word for that. Ethernet Cables and CPU Socket Pins are significantly different. Ethernet Cables also need to run literally 20,000 times the distance over a Socket Pin (estimated length of a PGA pin = 5mm, maximum length of Ethernet Cable = 100m without crosstalk being an issue).

 

So no, I cannot say it would be an issue. If you have some sources that back up that, feel free to post them. However, I'm not even sure anyone's even tried a multi-barrel pin before, so I would be surprised if there was any data available on the subject.

 

As I said, crosstalk very well might be an issue - or maybe it isn't. If you do have a source to backup your claim, I would love to read it, and I don't mind being wrong when presented with evidence.

Cat 6a can't run 600 Mhz at even 1m. That was literally the point of Cat 7. 

 

Additionally, we have already seen memory stability differences in DDR4 going down from ATX to mITX motherboards purely due to the dimm slots being closer to the socket (hence why the world record mem speed is currently on a fairly mid-teir itx board), crosstalk and bit rot is already an issue literally everywhere today.

 

IT WOULD NOT WORK. END OF STORY.

 

http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=6670506

 

Here is a paper for you about it.

[mariushm]

59 minutes ago, pyrojoe34 said:

Random thought regarding pin density/count:

 

Why has no one tried to implement a multi-layered pin system to increase connections without increasing physical pins. Something like a headphone jack where you have multiple connections on one pin by embedding pins within each other separated by insulating material. I'm sure that would increase the price of production a bit but you could double, triple, or quadruple the number of connections without actually increasing the number of pins. Seems like they'll eventually reach a point where they cannot realistically add more pins to a socket (just look at how may pins Skylake-e has) without exclusively selling the CPUs as permanently soldered to the motherboard.

You can't. Like others told you already, it would make it hard to solder the pins onto the substrate and the insulation thickness between the segments would kill any added benefit and the socket would be several times more complicated to manufacture and you'd have capacitance between the segments of the pin, leaking etc etc ... too messy.

 

As it stands now, pins are spaced just enough to allow a few traces to go between the socket contacts and make it easier to route all the signal wires without resorting to 8-10+ layers on the PCB.  For a modern processor, we already have 6 layers or so on each motherboard, just to route all the wires to the pci-e connectors and memory sots and route the power planes to VRMs to pass tens of amps of electricity to the cpu.

 

[dalekphalm]

32 minutes ago, Curufinwe_wins said:

Cat 6a can't run 600 Mhz at even 1m. That was literally the point of Cat 7. 

 

Additionally, we have already seen memory stability differences in DDR4 going down from ATX to mITX motherboards purely due to the dimm slots being closer to the socket (hence why the world record mem speed is currently on a fairly mid-teir itx board), crosstalk and bit rot is already an issue literally everywhere today.

 

IT WOULD NOT WORK. END OF STORY.

 

http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=6670506

 

Here is a paper for you about it.

I'll have to take your word for that article, since it's behind a paywall, unless you want to quote sections of it that specifically reference your claims? The abstract references how pin location and consideration are very important (No disagreements there), but obviously, you need to access the full text to actually see their findings.

 

Anyway, we'll leave it at that. I didn't even claim we should even be trying to use this method of pin configurations. Simply that engineers may not have ever even explored the possibility due to other concerns which would prevent them from even doing any testing on it.

 

Your point about Cat 6a vs Cat 7 though, I'm really not sure what your point is? Cat 6a could not achieve 600MHz, and Cat 7 did. A clear evolution of dealing with crosstalk (Albeit, in a way that likely has little to no relevance on CPU Pins), and achieving a higher transmission frequency.

[spartaman64]

9 hours ago, Dabombinable said:

Still, they aren't actually specifying what Zen is 40% faster at-because it won't be 40% faster at everything.

40% more IPC

[spartaman64]

i think it would actually be easier to bend the tiny pins on the intel motherboards than the broader ones on amd cpus and it shouldnt be an issue for either one unless you are throwing your cpu and motherboard around your room

[flibberdipper]

1 hour ago, Misanthrope said:

Anectdotal points never stand to rational minded people so bye. 

You're just mad you don't have anything to counter with. So bye.

[Trixanity]

48 minutes ago, ScootsMcgoots said:

 I dunno much about this but having " my cpu is more likely to be damaged than the motherboard" ( I'm referring to how the pins are on the CPU) is a less convincing argument when compared to saying " LGA would possibly allow a more advanced piece of hardware".  Yeah that's cool that your CPU has pins on it and might be more resistant to shock but that's an odd quality.  That's sorta like saying "Well your laptop may have better battery life but mine has a larger space on top of it to place my coffee cup.  I guess it's cool if the feature matters to you but I'd think if the feature was really important reviewers would actually criticize that quality of hardware.  I haven't heard someone reviewing a mobo complain of how fragile the pins are on LGA, I've just heard them  say "be careful etc" since it's common sense.  it's not exactly something manufacturers should have to worry about whilst designing their hardware.  I understand breaking the pins by accident once in awhile ( I dropped my cpu from a couple inches up down into the socket) but if you treat your hardware soo badly that being able to be rough with it is a quality you desire than maybe you should expect broken components.        

I don't think there's any reason to believe LGA allows more advanced hardware. There seems to be speculation that LGA allows more pins. Are more pins necessarily more advanced/better? Intel's consumer platform reduces and increases the number of pins by a single digit number from year to year with no apparent difference. Intel's prosumer/enterprise products have a lot more pins. Zen would supposedly have some 1300 pins. More than Intel's consumer products, less than enterprise. Does it then need LGA? No. It already has more pins than the consumer products.

Are all those extra pins necessary on the enterprise stuff? Perhaps. But supposedly Zen for enterprise will have LGA boards, so it will use LGA when necessary. Problem solved. PGA won't hold the platform back.

 

At best one could question why AMD would go for both approaches instead of sticking to one and how feasible it is but I doubt it changes much practically and financially. Both Intel and AMD already has to do BGA in addition to whatever they're doing anyway.

 

I think we've come to the point where people obsess over the pin count and pin placement. It would seem that people idealize the LGA approach because Intel does it. Do people think Intel owes the performance of its Core processors to using LGA? Seems ridiculous but that's the sense you get in this thread.

[Dabombinable]

10 hours ago, ace_cheaply said:

Um, yes they are?  Zen has 40% faster ipc.

Considering a Phenom II P920 is still faster in some areas despite having the same IPC as the A8 4555M in my new laptop......AMD's stated performance is never all across the board.

[Misanthrope]

3 hours ago, dalekphalm said:

To be fair, no one countered his claims with a verifiable source, so the group saying "PGA Pins are durable as fuck" and the group saying "PGA Pins literally murdered my dog" are basically in the same group.

 

It's entirely anecdotal evidence all the way around. I'd be interested in seeing some RMA stats about bent CPU PGA pins vs bent Motherboard LGA pins, and the rate of occurrence... but I'm not holding my breath on those stats ever appearing.

Conceding that, CPUs are generally more valuable anyway so assuming both cases are equally as likely, for a consumer LGA it's still preferable. 

[Misanthrope]

2 hours ago, spartaman64 said:

40% more IPC

Than AMD's own lack luster chips which are now a good 20% or worst behind Intel in ICP today? 

 

Meaning that after the usual hype BS of cherry picked tests and such and consider the would compete with not current but future Intel chips, suddenly it's about the same performance if not a little worst than Intel. 

 

You can quote me as predicting Zen will, as usual, only have a price to performance edge but not actually better or equal to high end Intel.