Is AMD's AM3+ the end of the line for the AM3 Series of sockets?

[SparksIT]

Is AMD's AM3+ socket already a "Dead" or EoL socket? Similarly to how it was rumored\known (at that time) that second generation of Intel's Core i series would be would be on LGA 1155, despite the recent launch of LGA 1156.

I only ask, as I was looking at AMD's road map, and Komodo, 2nd generation of Bulldozer\Zambezi will also feature an on chip DX 11 GPU. Given that Llano a tweaked K10\Stars CPU with an on chip GPU, despite the similarities on the CPU side, required a new socket, FM1. Presumable to accommodate a different pin layout to access the additional GPU portion of the chip. Since Komodo will have a gpu would it be presumptuous of me to assume that it to will require a different socket?

Additional would the powers at be at AMD be:

A: Forward thinking enough to design FM1 socket to work with Komodo. There by making FM1 socket their future?

B: Follow Intel's lead have 1 socket (FM1) for the mainstream market, and a different socket for the enthusiast market?

I would assume A given that Trinity is supposed to based on Bulldozers.

 

[SickBeast]

AM3+ is far from dead AFAIK. Bulldozer will use it which is huge. There is even the chance that BD will work on old AM3 motherboards.

 

[Ares1214]

This is extremely stupid. How can a socket that hasnt even be released yet be dead?! AM3 being dead, thats worth discussion, but this isnt.

 

[SickBeast]

AM3 is sorta dead, but Asus and MSI seem to think otherwise.

 

[SparksIT]

This is extremely stupid. How can a socket that hasnt even be released yet be dead?! AM3 being dead, thats worth discussion, but this isnt.

Maybe dead is the wrong word...Is AM3+ a one and done...they release bulldozer series chips, then 5-6 months release a speed bump. Then release a 2nd gen bulldozer aka Komodo on a new or different socket. Similarly the way Intel handled LGA 1156.

Let me ask this another way...is AM3+ the end of the line for the AM series of sockets?

 

[podspi]

I think OP is asking whether we think AM3+ will be a one trick pony.

OP, I believe there are going to be a few generations of BD and BD+ that are not APUs. I would expect BD to slot into them, but we have no way of knowing for sure.

 

[itsmydamnation]

i would expect AM3+ to be short lived, why?

1. i expect the lower end to be a fusion based socket
2. i expect the higher end to go to tri channel memory with sepang

that leaves poor old Am3+ no where to go.

 

[Triskain]

Future Bulldozer derivates will have integrated PCIe that alone means a new socket is needed. So yeah, AM3+ has a lifespan of about a year until the upgraded Bulldozer arrives in mid 2012 .

 

[Morg.]

i would expect AM3+ to be short lived, why?

1. i expect the lower end to be a fusion based socket
2. i expect the higher end to go to tri channel memory with sepang

that leaves poor old Am3+ no where to go.

THIS IS AMDDDDDDDDD !!! quite hard to scream that tbh.

Anyway, I don't see AMD not thinking compatibility, they'll work something out surely...

Does tri-channel memory force a socket change ?
Does fusion force a socket change ?

I agree there's some remapping to do and stuff, but do you really need a physical change (thus loss of compatibility) to do that ??

 

[Tuna-Fish]

THIS IS AMDDDDDDDDD !!! quite hard to scream that tbh.

Anyway, I don't see AMD not thinking compatibility, they'll work something out surely...

Does tri-channel memory force a socket change ?

Yes, there needs to physically be ~250 more pins in the socket for the memory. There just aren't enough pins on AM3(+) to accommodate 3 (or 4) channels.

Does fusion force a socket change ?

Likely, because you need a video output for it, which also takes some extra pins.

Also, future CPU's will put a PCI-e controller in the cpu (for the video card, like in LGA1156), which will take more pins.

AM3 is the continuation of the AM2 line from 2006, and frankly, the ~940 pins available have been pretty much tapped out. AMD currently clearly wants to maintain socket compatibility as long as possible, but if they want to have more things go trough that socket, they will simply have to make it bigger.

 

[nanaki333]

i just hope we don't end up with another s754 to s939 to AM2 fiasco again. did they even wait a year between 754 and 939?

 

[Tanclearas]

i just hope we don't end up with another s754 to s939 to AM2 fiasco again. did they even wait a year between 754 and 939?

Well, in the case of 754, everyone knew that 939 was coming. I waited for 939 and was not disappointed.

I view AM3+ as a way to get a cheap PhII now (I picked up a 965BE for $140) and then move to BD down the road. However, my guess is that AM3+ doesn't have a long future ahead.

 

[Morg.]

Yes, there needs to physically be ~250 more pins in the socket for the memory. There just aren't enough pins on AM3(+) to accommodate 3 (or 4) channels.

Aha .. I thought it was totally possible to do it with less pins if required... like how SAS has so many less connectors than any SCSI or even IDE for that matter.

Maybe at that level it's just too much of a performance hit ?

 

[Phynaz]

Socket compatibility is a red herring. It's motherboard compatibility that matters.

 

[Tuna-Fish]

Aha .. I thought it was totally possible to do it with less pins if required... like how SAS has so many less connectors than any SCSI or even IDE for that matter.

Maybe at that level it's just too much of a performance hit ?

Yes. You can get much more bandwidth per pin if you move from a synchronous bus into async and LVDS, like PCI -> PCIe, ATA -> SATA, old amd bus -> HT, and many other examples.

However, this would require a new memory standard (and DDR4 isn't it), and much more importantly, it would considerably hurt latency. And while the GPU doesn't care about latency, the CPU performance would be seriously degraded.

There has been a clear update path all the way from AM2 to AM3+ -- you can always update either the MB or the processor. This is laudable, but to put it very simply, you can't keep it up forever. Eventually you will hit a wall, and based on my layman analysis, AM3+ is it.

Also, DDR4 looms on the horizon, and unlike the DDR2 -> DDR3 transition, it's bus is not electrically similar enough that you can painlessly make a processor that supports both. Also, as there can only be one DDR4 dimm per channel, I can't see AMD making a DDR4 workstation processor with only 2 channels. So in the future there will at least be new sockets for:

- DDR3 fusion (FM1)
- DDR4 fusion
- DDR3 processor with PCIe on chip for discrete graphics (how many channels?)
- DDR4 workstation with PCIe on chip (more than 2 memory channels)

 

[nanaki333]

Well, in the case of 754, everyone knew that 939 was coming. I waited for 939 and was not disappointed.

I view AM3+ as a way to get a cheap PhII now (I picked up a 965BE for $140) and then move to BD down the road. However, my guess is that AM3+ doesn't have a long future ahead.

i had to order a bunch of desktops for work and ended up having to go with 754. making them obsolete within a year and hard to find parts for. funny thing though... they certain last. i just decomissioned them all a few weeks ago and they're sitting in storage now. all still fully functional.

my first a64 was a s939 variant, then the dually.... then eventually accepted the fact am2 was the new boss.

 

[PreferLinux]

Socket compatibility is a red herring. It's motherboard compatibility that matters.

Yes, and that is what everyone forgets too.

Yes. You can get much more bandwidth per pin if you move from a synchronous bus into async and LVDS, like PCI -> PCIe, ATA -> SATA, old amd bus -> HT, and many other examples.

However, this would require a new memory standard (and DDR4 isn't it), and much more importantly, it would considerably hurt latency. And while the GPU doesn't care about latency, the CPU performance would be seriously degraded.

There has been a clear update path all the way from AM2 to AM3+ -- you can always update either the MB or the processor. This is laudable, but to put it very simply, you can't keep it up forever. Eventually you will hit a wall, and based on my layman analysis, AM3+ is it.

Also, DDR4 looms on the horizon, and unlike the DDR2 -> DDR3 transition, it's bus is not electrically similar enough that you can painlessly make a processor that supports both. Also, as there can only be one DDR4 dimm per channel, I can't see AMD making a DDR4 workstation processor with only 2 channels. So in the future there will at least be new sockets for:

- DDR3 fusion (FM1)
- DDR4 fusion
- DDR3 processor with PCIe on chip for discrete graphics (how many channels?)
- DDR4 workstation with PCIe on chip (more than 2 memory channels)

I think you mean from parallel to serial. Your statement is true, but isn't relevant to the question.

 

[Arkaign]

Socket compatibility is a red herring. It's motherboard compatibility that matters.

This_this_this. Once in a while, I'm able to put in a decent upgrade CPU into a mobo that was released a while before a CPU revision, but for the most part even if the socket stays the same, you end up needing a new chipset to really use the new chip. Examples :

Almost all 915/945 chipset boards would not run C2D. Almost all 965 boards wouldn't run 45nm C2D/C2Q. Many AM2 mobos wouldn't run PhII AM2+. Etc. Lately it's gotten more severe, with AM3 possibly not able to run BD adequately (unknown?), S1156/1366 having really only one real gen of chips, etc. We don't even know for sure if Ivy bridge will run in existing SB-oriented chipsets, or if the socket will stay the same but some boards be incompatible. Who knows.

In the end, if you're not doing multi-GPU, decent mobos aren't that expensive anyway. It's often better just to pair up a new CPU with a new mobo to match, particularly if you can just recoup some of your investment by reselling your old gear.

 

[MentalIlness]

You ask if "AM" is a one and done thing ? Or dead ?

Is it that you are asking if AMD will rename the sockets ? And quit adding the "AM" to it ? Different prefix ?

 

[MarkLuvsCS]

AMDs backwards compatability has not been bad at all. Most AM2 boards run the newer PhII just fine it's just they won't typically run 125w socket chips. It's true there definitely are some upgrades that simply aren't possible but most still are. I recently upgraded a pretty old Phenom 1.8ghz quad to a new 3.2 ghz propus core and it was definitely worth it. The 1.8 was holding back my 460 by a considerable margin @ 1920x1080. So the ~$100 upgrade breathed new life into a secondary gaming machine. I have typically purchased AMD in the past mostly for a total cost of the system and a pretty easy upgrade path.

I don't think a new socket is really going to affect that many people especially when it seems pretty solid amd boards always seem to round $100-150 and even cheaper budget boards without the bells and whistles. It would be nice if AM3+ is the budget series sort of like the 754s were and then release the "higher" end socket. At least if they tell everyone their plans around bulldozers release with future upgrades if people really think it'd be terrible to purchase a new board for a newer system later on down the road they can make their choice ahead of time.

 

[StrangerGuy]

Complaining about socket changes?

It's called "buy a new mobo when you buy a new CPU" and that works everytime. Decent ones are not expensive. Unless of course you insist always to use overpriced bling-bling boards...Which is like complaining about your fuel bill while driving a SUV.

 

[Concillian]

They did the same thing with Socket A. So many versions of Socket A though, still needed a new motherboard each new CPU.

Socket compatibility isn't everything.

There is always something new (and incompatible with the old socket) each generation. If you look back to the real roots, S939 is barely different from AM2 (1 pin different so you don't accidentally plug the wrong thing in). http://www.amd.com/us/products/desktop/processors/athlon/Pages/athlon-faqs-socket-am2.aspx

If you count that, Essentially the same socket all the way back to single core dominance days. It's time to make a new socket.

 

[RobertPters77]

S754 wasn't so bad. All you lost was dual channel memory. After 939 came out, the S754 chips were cheap as hell. I remember paying 120$ for an A64 3400. And the mobile chips fit in the desktop sockets which was a plus when the dual core Turions came out.

If I didn't kill that system with an overclock, I'd still be using it right now.