More AMD Socket AM4 Technical Details Emerge

[btarunr]

More details of AMD's upcoming common socket for both its desktop APUs and high-end CPUs emerged from a recent article by Italian tech-site Bits-n-Chips. To begin with, AM4 will be an µOPGA (pin-grid array), in which the pins will continue to be located on the processor package, and contact points on the socket. The package will be square, and 40 mm in length, making it about as big as a current socket FM2+ package. It will have a pin-count of 1,331 pins, a big increase from the 942 pins of AM3+, and 906 pins of FM2+. AMD could continue to develop LGA sockets for its multi-socket capable Opteron processors based on the "Zen" architecture.

The AM4 platform layout will be functionally closer to that of the FM2+ than the AM3+. Besides the integrated memory controller, the northbridge will be entirely located on the processor die; and so the HyperTransport main system bus will be wired internally. Besides hundreds of electrical pins, the AM4 pin-map will consist of memory I/O, integrated graphics I/O, PCI-Express, and the chipset bus; besides other low-level system I/O interfaces. The memory controller on some of the first AM4 chips, such as "Summit Ridge," will natively support DDR4-2400 MHz, and DDR4-2933 MHz through overclocking.

View at TechPowerUp Main Site

 

[john_]

Let me add here that the socket will be able to support CPUs/APUs/SOCs with up to 140W TDP, so we can expect really powerful integrated graphics combines with maybe more than 4 cpu cores.

 

[PP Mguire]

Sounds good so far, if it's true.

 

[newtekie1]

Damn, I was hoping for LGA. Pins on the processor is so 1990's.

 

[uuuaaaaaa]

Sounds good so far, if it's true.

http://wccftech.com/amd-am4-µopga-socket-1331/

 

[Jism]

Damn, I was hoping for LGA. Pins on the processor is so 1990's.

Since the Socket 3 amd always used pins, except for the Slot A version.

However, 1331 pins ofcourse sounds tremendous. More power, less perhepials such as external northbridge, but less choice on motherboard vendors as well.

I.e a extreme motherboard would have the cherry picked northbridge chipsets, and now it's just a matter of silicon luck to have a NB that pushed beyond 300MHz FSB/HTT.

 

[Frick]

Damn, I was hoping for LGA. Pins on the processor is so 1990's.

A point that was made when I dropped a C2D a few weeks back. With 1331 pins on the CPU you don't really want that to happen.

 

[WaroDaBeast]

A point that was made when I dropped a C2D a few weeks back. With 1331 pins on the CPU you don't really want that to happen.

That's like saying glasses made ouf of glass should all be replaced with plastic cups, just in case you drop them.

 

[Jism]

AMD was always easy to 'insert'. You'd simply 'drop' the CPU without any force into it's socket. It was always one way by looking at the corner of the CPU and socket, so you'd know how to insert it.

Anyone not reading the manual shoud'nt be installing CPU's at all.

 

[straim]

Since the Socket 3 amd always used pins, except for the Slot A version.

However, 1331 pins ofcourse sounds tremendous. More power, less perhepials such as external northbridge, but less choice on motherboard vendors as well.

I.e a extreme motherboard would have the cherry picked northbridge chipsets, and now it's just a matter of silicon luck to have a NB that pushed beyond 300MHz FSB/HTT.

Hi, first post here a TPU! I think they mean the CPU/NB, which is different from the Chipset/NB The current CPU/NB and has an exclusive power plane in the CPU (the last statement according to the AMD "Scorpion platform" specs), and I think that could be the change here, Am I completly wrong?

 

[Jism]

Read source:

the northbridge will be entirely located on the processor die;

The CPU/NB was already located on the CPU since the Thuban if i'm not mistaken. Locating the NB on the cpu has some advantages, such as lower latencys, but more problems with getting a cherry picked NB for higher clocks.

 

[Frick]

That's like saying glasses made ouf of glass should all be replaced with plastic cups, just in case you drop them.

I don't think so. Is there a reason for the pins to be on the CPU?

 

[WaroDaBeast]

I don't think so. Is there a reason for the pins to be on the CPU?

Is there a reason to use glass as a materiel to make glasses?

 

[john_]

A point that was made when I dropped a C2D a few weeks back. With 1331 pins on the CPU you don't really want that to happen.

There are cases with bent pins on the motherboard and cases with bent pins on the cpu. In every case you have to be careful. Then there is also the Skylake chips that were bending under force. That wouldn't have happened if the pins were on the cpu.

 

[Frick]

Is there a reason to use glass as a materiel to make glasses?

Many, and few of them are technical. Is there any technical reason why the pins should be on the CPU? It's a serious question I don't know the answer to.

 

[newtekie1]

Since the Socket 3 amd always used pins, except for the Slot A version.

Except they've been using LGA on their servers for generations.

I.e a extreme motherboard would have the cherry picked northbridge chipsets, and now it's just a matter of silicon luck to have a NB that pushed beyond 300MHz FSB/HTT.

Doesn't matter, they are going the way of Intel, so no real overclocking without an unlocked multiplier.

That's like saying glasses made ouf of glass should all be replaced with plastic cups, just in case you drop them.

Ironically, glass in eye glasses has almost entirely been replaced by plastic(Poly Carbonate), largely because of the glass being easily broken when dropped...

There are cases with bent pins on the motherboard and cases with bent pins on the cpu. In every case you have to be careful. Then there is also the Skylake chips that were bending under force. That wouldn't have happened if the pins were on the cpu.

It is far easier to bend pins on the CPU than pins on the motherboard.

A big problem with the Skylake bending was the very thin PCB on the processor. The other issue was that there are no pins, and no support, under the middle of the processor. If the pins covered the entire base of the processor/socket, skylake would have not issue with bending.

 

[john_]

It is far easier to bend pins on the CPU than pins on the motherboard.

A big problem with the Skylake bending was the very thin PCB on the processor. The other issue was that there are no pins, and no support, under the middle of the processor. If the pins covered the entire base of the processor/socket, skylake would have not issue with bending.

I haven't had an Intel cpu to know the difference, but I have seen cases with people who managed to bent the pins on their motherboards. I also had seen cases from at least one retail shop that bent pins on the motherboards(based on what people where saying) so it can refuse RMAs. Now in 17 years of AMD cpus(my only Intel CPU didn't had pins, neither the motherboard) I have never managed to bent one pin.

If the pins where on the Skylake it wouldn't matter if there where no pins in the edge or in the middle.

Anyway with 1331 thin pins my opinion in this matter could change. Hope it never happens of course.

 

[newtekie1]

I haven't had an Intel cpu to know the difference, but I have seen cases with people who managed to bent the pins on their motherboards. I also had seen cases from at least one retail shop that bent pins on the motherboards(based on what people where saying) so it can refuse RMAs. Now in 17 years of AMD cpus(my only Intel CPU didn't had pins, neither the motherboard) I have never managed to bent one pin.

Which is more likely? You drop your CPU, or you mis-handle it and touch one of the pins and bend it. OR You drop something onto your motherboard, in the small time the socket is uncovered and it just happens to hit that small little 1"x1" square target and bends some pins? It doesn't matter that you've never bent a pin on a CPU, I've never bent a pin on a motherboard, the fact is the pins on the CPU make them more prone to damage.

If the pins where on the Skylake it wouldn't matter if there where no pins in the edge or in the middle.

Yes, it very well would have mattered. The thin PCB is what causing the bend. Pins might help a little, but not much. Pins are solidly connected to anything, they float in the socket with a very small force holding them in place. That is why they are so easily pulled out with the heatsink when the TIM has a strong enough bond between the CPU and Heatsink. I've warped CPUs with pins in the past.

Not having support under the middle of the CPU, where the majority of the force from the cooler is focused is what caused skylake to bend.

 

[Am*]

For the love of God AMD...do no mess this launch up. The below quote is already making me nervous:

The memory controller on some of the first AM4 chips, such as "Summit Ridge," will natively support DDR4-2400 MHz, and DDR4-2933 MHz through overclocking.

It needs to support 3200MHz out of the box, no excuses...no more repeats of history please, like the slow-as-hell DDR2/3 mixed on-die controllers of the AM3 days where Sandy Bridge was mopping the floor with the Phenoms with 2-3x the bandwidth/performance using the same memory. And no bloody BIOS flashing to support higher frequencies either...

It's all or nothing for AMD now -- and if this will be another stagnant launch like the Faildozer, I will put my PC building hobby to rest for at least another 5-8 years (or until Intel release something that's at least 3-4x faster than my current CPU, whichever comes first)...which won't be good for either Intel nor AMD. DX12 and especially Vulkan, along with the current gen consoles -- will make it that much easier to hold out for a real upgrade.

 

[cdawall]

It is far easier to bend pins on the CPU than pins on the motherboard.

While I agree with this theory I cannot tell you how many motherboards I have straightened pins on recently. The number is uncountable and the user that does it typically blames everyone, but themselves. I have yet to receive an AMD CPU with bent pins at least in the shop.

It needs to support 3200MHz out of the box, no excuses...no more repeats of history please, like the slow-as-hell DDR2/3 mixed on-die controllers of the AM3 days where Sandy Bridge was mopping the floor with the Phenoms with 2-3x the bandwidth/performance using the same memory. It's all or nothing for AMD now -- and if this will be another stagnant launch like the Faildozer, I will put my PC building hobby to rest for at least another 5-8 years (or until Intel release something that's at least 3-4x faster than my current CPU, whichever comes first)...

Intel only technically supports 2133 out of the box

http://ark.intel.com/products/88195/Intel-Core-i7-6700K-Processor-8M-Cache-up-to-4_20-GHz

 

[Am*]

Intel only technically supports 2133 out of the box

http://ark.intel.com/products/88195/Intel-Core-i7-6700K-Processor-8M-Cache-up-to-4_20-GHz

I know -- the same way Sandy Bridge only technically supported 1333MHz max, but as always Intel tend to understate their figures, while AMD tend to overstate. Last time, both supported up to 1333MHz and yet Sandy Bridge was getting 16-18GB bandwidth while Deneb/Thuban got 7-9GB using the exact same modules at same specs. I don't want to see them skimp on anything this gen and they'll need all the future-proofing they can get to persuade people to buy into their platform this time (especially after AM3+/Bulldozer fiasco).

From what AMD's roadmap stated, they plan on making at least 3 generations of Zens and they don't have the luxury that Intel do by just bringing out a new socket with a new line-up of chips -- so AM4 needs to have enough features over Skylake to make it a far better option and hopefully prompt Intel to launch a new socket...otherwise this will all be a huge waste of time for them.

They NEED a platform & CPU line-up that beats Skylake enough to prompt a new socket out of Intel -- to give them the much needed time to claw back market share and sell enough motherboards to hold people over from switching to Intel. Otherwise if it doesn't -- Intel sit back, introduce some minor price cuts in response and rub salt into the wound with their next socket release (putting the final nail in the coffin for AMD's CPU division).

 

[PP Mguire]

For the love of God AMD...do no mess this launch up. The below quote is already making me nervous:



It needs to support 3200MHz out of the box, no excuses...no more repeats of history please, like the slow-as-hell DDR2/3 mixed on-die controllers of the AM3 days where Sandy Bridge was mopping the floor with the Phenoms with 2-3x the bandwidth/performance using the same memory. And no bloody BIOS flashing to support higher frequencies either...

It's all or nothing for AMD now -- and if this will be another stagnant launch like the Faildozer, I will put my PC building hobby to rest for at least another 5-8 years (or until Intel release something that's at least 3-4x faster than my current CPU, whichever comes first)...which won't be good for either Intel nor AMD. DX12 and especially Vulkan, along with the current gen consoles -- will make it that much easier to hold out for a real upgrade.

THey don't really need high memory bandwidth to target us regular joes in the crowd. They need single threaded performance.

 

[Am*]

THey don't really need high memory bandwidth...

Umm yes they do. Average Joes will not get them very far -- they need OEMs and the rich server-buying crowds on their side, far more than they need the Average Joes.

 

[alucasa]

The problem is the pins. Whether it's on CPU or on mobo, it does not matter. They need to invent a new way that does not involve fragile parts (pins).

 

[eidairaman1]

id be more afraid of bending pins on a mobo than a CPU