AMD FM1 and FM2 Packages Pictured Side-by-Side, Incompatible

[btarunr]

Here is the first picture of AMD accelerated processing units in the existing FM1 package and future FM2 package on which the next-generation Trinity APU will be based on. Both packages are very similar, follow AMD's favourite yet archaic PGA design. The pins are physically arranged in a mostly similar fashion, though we don't have a pin map at hand. The difference comes with some of the blanked pins in the sub-central portion of the pin array. The FM2 package has 904 pins, compared to 905 on FM1. One pin is blanked, while a pair of blanked pins are arranged further away from the central cutout.

This makes FM1 and FM2 clearly incompatible. Neither will you be able to use today's A-Series APUs in the FM1 package on future socket FM2 motherboards, nor will you be able to run future FM2 APUs on today's FM1 motherboards. Yet, AMD will port the A75 FCH chipset to the next-generation "Virgo" platform. The FCH or Fusion Controller Hub, like Intel's PCH (Platform Controller Hub), is not much more than a glorified southbridge, and is portable between Fusion platforms as it's essentially a PCI-Express 2.0 x4 device. In the picture below, "Llano" FM1 APU is on the left, and "Trinity" FM2 on the right.



View at TechPowerUp Main Site

 

[Jstn7477]

Well, hopefully like Intel's transition from 1156 to 1155, it means that the chip has been reorganized and is *hopefully* going to be even better than the current APUs. We don't really want Bulldozer (in its current state) with graphics slapped on, do we?

 

[Super XP]

If FM2 only has 904 pins, then wouldn't Socket AM3+ having more pins be better? AMD wanted a 10-Core Piledriver in the FM2 package, but that doesn't make sense, there's not enough pins NO?

 

[Nesters]

****,I was going to get an Athlon X4 631 and OC the shit out of it and then drop similar kind of FM2 CPU/APU on the same board.
On second thought, I might still get that Athlon because I have no idea how much better is Bulldozer going to perform on FM2/win8.

 

[TheLostSwede]

Unexpected... NOT!

Also, don't forget the A85X FCH which should launch alongside the FM2 APUs

 

[qubit]

It looks like AMD are taking a leaf out of Intel's book and concentrating less on socket compatibility and more on improving performance. The right move, in my opinion.

 

[btarunr]

AMD wanted a 10-Core Piledriver in the FM2 package

Lolz where did you hear that? Trinity has two Piledriver modules (4 cores).

 

[mixa]

AMD's favourite yet archaic PGA design

And which package/design is not archaic?That's sounds way too ala fanboy you know.
And yes it makes sense these to be incompatible, there are going to be changes in the new APUs force AMD to make them incompatible, so ppl don't start to make experiments and end up with dead CPUs or MBs and then go QQ to AMD

 

[btarunr]

and which package/design is not archaic?that's sounds way too ala fanboy you know.

lga775.

 

[Disparia]

Good! Make less compromises for compatibility and more improvements for performance.

 

[NC37]

And which package/design is not archaic?That's sounds way too ala fanboy you know.
And yes it makes sense these to be incompatible, there are going to be changes in the new APUs force AMD to make them incompatible, so ppl don't start to make experiments and end up with dead CPUs or MBs and then go QQ to AMD

Agreed. I happen to not mind PGA or ZIF socket designs. Rather like them.

 

[Damn_Smooth]

I wonder if switching to piledriver cores will hurt trinity. Throwing 4 BD cores in there as they are now would be a definite downgrade from Llano.

 

[qubit]

Ok, just read the article properly and seen that AMD is porting over the A75 FCH chipset to the new socket. So, doesn't that suggest that the improvements are incremental and therefore the socket didn't have to change? Definitely need more info on this one to say for sure.

 

[Damn_Smooth]

Lolz where did you hear that? Trinity has two Piledriver modules (4 cores).

It was rumored that enhanced BD was going to be on FM2, and that there would be 10 core chips before we heard anything about Vishera. Not that it matters, I'm not too optimistic about it's performance anyway.

 

[devguy]

****,I was going to get an Athlon X4 631 and OC the shit out of it and then drop similar kind of FM2 CPU/APU on the same board.
On second thought, I might still get that Athlon because I have no idea how much better is Bulldozer going to perform on FM2/win8.

With so much of the chipset built into the APU, overclocking via the reference clock on APUs will not garner you much benefit. It's like trying to overclock an Intel 2500 (not k). I'd be interested to see how much the upcoming black edition Llano chips will overclock, though.

 

[Mussels]

i'm curious about FM2, to be honest. could make some fantastic portable/LAN systems, as well as HTPC's.


as long as its faster than FM1, everyone is happy.

 

[Inceptor]

I doubt FM2 will be more than, say, 10% faster than FM1 -- on the cpu side (IF, a big IF, those two Piledriver modules perform reasonably well). The big improvement will be in the iGPU.

HTPC and extreme budget system territory, like FM1.

 

[btarunr]

It was rumored that enhanced BD was going to be on FM2, and that there would be 10 core chips before we heard anything about Vishera. Not that it matters, I'm not too optimistic about it's performance anyway.

No, those 8(+) core Piledriver chips will continue using AM3+. FM2 is a feature-limited socket for commercial desktops, home PCs, and HTPCs. You can't have more than 16 PEG lines using it.

 

[wahdangun]

Ok, just read the article properly and seen that AMD is porting over the A75 FCH chipset to the new socket. So, doesn't that suggest that the improvements are incremental and therefore the socket didn't have to change? Definitely need more info on this one to say for sure.

no, using same chipset doesn't mean the improvement is minor, its just seems the design in the cpu it self was not compatible with current motherboard implementation, maybe AMD will use triple or quad channel in trinity so it would not bottlenecked.

 

[Disparia]

I doubt FM2 will be more than, say, 10% faster than FM1 -- on the cpu side (IF, a big IF, those two Piledriver modules perform reasonably well). The big improvement will be in the iGPU.

HTPC and extreme budget system territory, like FM1.

And HPC

On November 2, Penguin Computing announced that they successfully deployed the first high performance computing (HPC) cluster, that uses AMD Accelerated Processing Units (APUs). The cluster consists of 104 new Altus 2A00 servers, interconnected with QDR Infiniband switched link. According to Penguin Computing, the cluster has peak performance 59.6 TFLOPs, or 59.6 * 1,000,000,000,000 floating point operations per second.

Each 2A00 server employs 1 quad-core APU from A6- or 8-Series, which integrates memory controller, PCI-Express controller, and 320 or 400 graphics stream processors. The server supports up to 64 GB of DDR3-1600 or DDR3-1866 memory, and can have up to 8 hot-swappable disk of solid state drives. The server provides several connectivity options, such as 40Gbps InfiniBand QDR, 10Gbs Ethernet NIC, or dual or quad 1Gb Ethernet connections. The server is housed in a 2U form factor chassis.

^ There's someone who wishes FM2 was compatible. Drop-in 104 Trinity CPUs next year for a nice boost.

 

[Damn_Smooth]

No, those 8(+) core Piledriver chips will continue using AM3+. FM2 is a feature-limited socket for commercial desktops, home PCs, and HTPCs. You can't have more than 16 PEG lines using it.

Thanks. I was getting FM2 mixed up with FMx. If they still plan on going that route sometime.

 

[Cuzza]

phew, glad I didn't buy FM1.

+1 to the PGA ZIF is NOT archaic argument. There is nothing wrong with it.

 

[Wile E]

I much prefer PGA over LGA. Bent/broken pins on a cpu are easier to fix than bent or broken pins in a socket. I wish Intel would go back to it.

 

[btarunr]

Socket pins can be made much more compact than package pins, and hence higher pin-counts can be achieved using LGA. For the same reason, AMD uses LGA and not PGA for all its packages with over 1000 pins. Intel switched to LGA from PGA because it wanted to maintain a package size similar to s478 package, while making room for 775 pins. LGA775 packages ended up being more compact than AMD's PGAs.

AMD has LGA1207 and LGA1974 (G32), which are both LGA, Opteron processors use them.

So the debate between PGA and LGA cannot be reduced to Intel vs. AMD by angry AMD fankids. PGA is archaic whichever way you look at it. It poses pin-density limitations, in turn pin-count limitations if you don't want to enlarge your package dimensions like no tomorrow.

If AMD wants more pins, for more memory channels, more PCIe lanes from the CPU, or simply more HyperTransport links, the transition to LGA is inevitable. Opteron products already made that switch five years ago.

 

[Xajel]

I was hoping AMD will make the next Bulldozer socket compatible with both Bulldozer and it's APU's but not the other way, it can make the Bulldozer socket larger but the central pins in both CPU's are identical so the APU can be used on Bulldozer socket without problem... it's hard I know its hard but it will be nice

The problem is both CPU's are targeted to different segmentation, Bulldozer is a high-end with high power and normally big packaging, and Trinity ( or what ever comes later ) is mid-end with medium power + the video things...

having a small CPU package on large CPU socket will result in excessive physical stress on both an might damage the socket, unless it was designed for such purpose !!