Socket M2 - the next Socket A... or not?

[xbdestroya]

Well, here's my situation - I have a socket 754 system that I put together early this February; kind of my "Finest tech of 2004" PC, as it were. Anyway my original plan was just to skip Socket 939 entirely and jump on Socket M2, but now Socket M2 seems to be coming a little sooner than I thought, and with features that to me are just sort of 'meh.'

I mean, I may yet be blown away by the eventual benchmarks, and I really do like the whole notion of a unified socket, but what's the lifespan of this thing going to be? I only ask that because we have DDR3 getting ready to enter the scene and 'The CPU formerly known as K9/K10' not too far off at this point. Well, supposedly...

What are peoples thoughts on how long M2 will last for, it's advantages - if any - over 939, and how soon we'll be seeing a move to a true successor to the K7/K8/K8+ architecture?

I know this is all specualtion absent some harder facts, but still...

 

[imported_Kensai]

It seems that the A64 sockets are shortlived.. I wouldn't assuming the M2 to live too long.

 

[Bona Fide]

Some people are saying M2 is a "transitional" socket, until AMD finalizes its K10 line, which should be out by mid-2007. Hence there are few really "new" features. I personally abhor it. I see nothing good coming out of it; just "different". The move to DDR2 is a nice idea to help set one standard for memory between Intel and AMD, but then again, when did that ever help? Also, the new M2 processors are gonna be getting more juice through the mobo, which could result in higher power draws. We'll just have to wait and see. My s939 X2 should hold me well into 2008.

 

[coomar]

its not going to be as long as socket a, at most 3 years, ddr3 and integrated pci-e bus will make the chipset after a significant step forward

 

[xbdestroya]

Yeah you guys are thinking along the same lines I am. I wonder if I should recreate my system as a socket 939 then or if I should hold out until 2007+. I mean, 2 years isn't that long to wait, right?

The reason I would want 939 is of course to go dual core when the prices come down a little more - I'm really eager to get that 'smoother,' heavily multi-tasked computer environment. Ugh. Oh well, at least I'm of the sense that I'm not alone in thinking M2 might be a pass.

 

[Furen]

I'd say M2 has potential for 3-4 years, which is much better than socket 754/socket 939. Power consumption should be the same for socket m2 cpus unless the memory controller uses a lot more juice (which I doubt). The fact that AMD spec-ed the socket to work at higher currents just means that they are leaving themselves some headroom for jacking up the clockspeed if needed even if they havent done the 65nm shrink (or if the shrink just sucks), it DOES NOT mean that your 2.6GHz A64 will start burning more power for no apparent reason. I'm guessing that the K10 will also use hypertransport (and an integrated mem controller) so it could conceivably be socket-compatible with M2 K8/9s as long as AMD doesnt try to make the jump to DDR3 on the initial batch. AMD can conceivably stick 4 cores into socket 939 since the system request interface has 4 ports for them (if it can tame the power consumption, which is doubtful at 90nm), so that could happen on socket M2 as well.

All in all, I'd say M2 will remain AMD's desktop socket until DDR3 (or whatever is the next DRAM standard) reaches a price parity with DDR2.

 

[Hard Ball]

It seems that AMD has the plan to use the M2 as a testing ground for some of the newer technologies, such as DDR-II on-die controller, 65nm process, pacifica and presidia, and perhaps some new SIMD instruction sets.

I think that outlook of the near term future, one year from now, in the fall of 2006 will be very similar to that of the Winter of 2002-03. At that time, the AXP T-bred, and especially the bartons was an incremental improvement on the Palomino core; with new technologies such as 200MHz FSB (double pumped), 512k L2, etc. M2 CPUs would be in a very similar position, namely trying to make incremental improvements and trying to squeeze as much performance boost to K8 / K9(DC K8s) design, before the big switch to K10.

K10, as indications have it now, will start to appear on the server/workstation side sometime around the turn of the calendar between 06 and 07, and the desktop/notebook segments might not see it until significantly later. I think Conroe will beat K10 to the consumer segments by a few months, since it's supposed to come sometime in 2H 06. During Q4 06 and Q1 07, I would expect intel platforms to outperform the M2s for at least a few months. So I don't think it would be a very good idea to commit a lot of cash to an M2 setup, which is likely to both be a dead end investment, and not living up to the competition soon after its advent. It would be wise to either wait for the K10 on desktops, or to give a try on the Conroe in that time frame; just as it wasn't a very good idea to get a Barton CPU in spring of 03, when you know the Northwoods are beating them, and K8 was just on the horizon.

Socket1207 (F?) would be a much more likely candidate to have longevity, since it is most likely the initial socket for K10 servers. 1207 or a similar pin-count will be the final version of the K10 desktop platform. And the future of memory is not DDR-II, it will be FBDIMM and DDR-III. If anyone wants to upgrade soon, S939 would be a perfectly alright alternative, since it might have just as long life as the M2, as long as the supply of DDR memory continues to be strong. Alternatively, people can give Yonah a try in the next few months, and move on to K10 a year later.

 

[xbdestroya]

@Furen: Well, assuming we use DDR3 as the point of transition, estimates put 2007 as the time we'll see DDR-3 enter the marketplace. Now, assuming it performs at launch the way DDR-2 didn't, I think we'll see AMD as soon as possible come out with a chip that supports it, if only for the server space. Agreeing that they may not launch a desktop variant of the chip immediately due to DDR3 prices, I could see a situation in which the DDR-2 solution sticks around for awhile. Still, that gives Socket M about two years before DDR3 definitely begins penetrating the desktop; they may support Socket M2 after that, but I would be surprised if there wasn't another desktop socket introduced by then.

I guess K9/K10 is what I'm most curious about at this point though. With Intel's new push towards the M architecture, I wonder if that's causing a change in direction over with AMD's next gen development.

 

[xbdestroya]

Hard Ball I think you make a good point about the future Opterons; they may just well skip DDR-2 entirely, since their supposed arrival puts them just about when we'd be seeing DDR3 and FB around. In that sense Socket 1207, if no other, will likely be K10 ready - even if it's not the K10 that initially populates them.

But on your desktop analysis:

K10, as indications have it now, will start to appear on the server/workstation side sometime around the turn of the calendar between 06 and 07, and the desktop/notebook segments might not see it until significantly later. I think Conroe will beat K10 to the consumer segments by a few months, since it's supposed to come sometime in 2H 06.

I mean we're not talking about a few months if that scenario comes to pass, we're talking about a year or more.

But I don't argue with the logic, which leads me to believe that Socket M2 will be ready to perform if need be, probably in line with the higher power ratings spec'd on the motherboards. Remember, there are going to have to be some additional changes to Pentium M (in Crusoe) above and beyond hyperthreading and 64-bit to make the architecture competetive in the areas in which K8 presently holds it's leads. Obviously though on the power/silent PC appeal front I expect it to rule the roost.

 

[Furen]

I would hardly count FB dimms as the future of ram. FB dimms are made for platforms that use huge amounts of ram, which is where the buffering helps by improving signal integrity and allowing lower pincounts while scaling to the same maximum bandwidth as systems with less modules. You could say FB is registered ram taken to the extreme. Remember, though, that it leads to higher latency because the ram chips have to go through the buffer before hitting the memory controller. Also (as I understand), multiple ram modules are daisy-chained through their AMBs (advanced memory buffers... what an unoriginal name) so that raises the latency hit even more if the modules are farther away from the controller. DDR3 looks to be the future but I wouldnt expect it to be in a price parity with DDR2 in less than 3 years, which is about what it took DDR2 to reach striking distance to DDR. Also, there seems to be quite a bit of life left in DDR2, so we may even have to wait longer for DDR3.

About socket F... well the only thing I know about it is that it will have 1207 pins... that's it. I would be hardpressed to consider it AMD's next desktop socket since, for all I know, the extra pins could be more FB ram channels, a PCI-X host or (as the inquirer speculated but some people seem to be convinced of) a PCI-E bus. If it is a PCI-E bus, I wouldnt really expect to see much of a performance benefit from it. After all, graphics cards arent much faster when they are puggled directly onto the northbridge (ala NF4) or on the southbridge (like the ULi1695 or whatever it's called), your harddrive is not going to spin faster because it's now plugged directly onto the cpu and 20-30 nanoseconds less latency will hardly make the 11ms seek times any better.

 

[valkator]

well the way I see it if we go by how long each socket's lifespan and AMDs pattern for setting that... hmm socket A, a good 5 years, socket 754 like not even a year, socket 940 aka the fx-51 by example for desktop comps which lasted like 6 months, then 939 which lasted about the same. It seem to me if AMD keeps this up and if we here at AT are persistant with upgrading to the latest platforms this will happen:

1. We will be all poor from these huge changes in platforms.
2. Socket M2 will last less than 6 months
3. We will be shooting ourselves in the foot for making the upgrade to M2.

Now hopefully this next Socket will actually last longer than a year. And I am hoping it will go as long as our beloved socket A. Only time will tell.

 

[xbdestroya]

I don't see Socket 1207 on the desktop either, but I *do* see it as being able to accomodate K10, if and when it comes out. And in that vein, I expect that Socket 1207 will use one of the newer RAM generations. As for price parity, I've read that DDR3 is expected to reach 65% market penetration by 2009, so if it launches in 2007 (commercially), they're clearly looking at an aggressive ramp. D

If we can determine that Socket M2 will indeed be the desktop mainstay for AMD for the next couple of years, then I'll be going that route. Otherwise if it seems it will be simply a 'holdover,' then it's 939 skipping M2 or 754 skipping both. If it's the 'real deal' though, then it's 754 to M2.

 

[xbdestroya]

Originally posted by: valkator
well the way I see it if we go by how long each socket's lifespan and AMDs pattern for setting that... hmm socket A, a good 5 years, socket 754 like not even a year, socket 940 aka the fx-51 by example for desktop comps which lasted like 6 months, then 939 which lasted about the same. It seem to me if AMD keeps this up and if we here at AT are persistant with upgrading to the latest platforms this will happen:

1. We will be all poor from these huge changes in platforms.
2. Socket M2 will last less than 6 months
3. We will be shooting ourselves in the foot for making the upgrade to M2.

Now hopefully this next Socket will actually last longer than a year. And I am hoping it will go as long as our beloved socket A. Only time will tell.

LOL, you know you're not compelled to purchase every new socket generation that comes out though. That's kind of why I started this thread, to have a sort of long term upgrade plan in mind that smooths out the bumps, gives good distances, and reduces expenditure.

By the way, I could argue that Socket 940 has lasted longer than any of them since it supports Opterons as well - and 939 will definitely last at least a little longer before it's EOL. It'll probably be going strong for a bit into M2's life, as well as being the new target platform of lower-end Opterons.

 

[valkator]

Originally posted by: xbdestroya

Originally posted by: valkator
well the way I see it if we go by how long each socket's lifespan and AMDs pattern for setting that... hmm socket A, a good 5 years, socket 754 like not even a year, socket 940 aka the fx-51 by example for desktop comps which lasted like 6 months, then 939 which lasted about the same. It seem to me if AMD keeps this up and if we here at AT are persistant with upgrading to the latest platforms this will happen:

1. We will be all poor from these huge changes in platforms.
2. Socket M2 will last less than 6 months
3. We will be shooting ourselves in the foot for making the upgrade to M2.

Now hopefully this next Socket will actually last longer than a year. And I am hoping it will go as long as our beloved socket A. Only time will tell.

LOL, you know you're not compelled to purchase every new socket generation that comes out though. That's kind of why I started this thread, to have a sort of long term upgrade plan in mind that smooths out the bumps, gives good distances, and reduces expenditure.

By the way, I could argue that Socket 940 has lasted longer than any of them since it supports Opterons as well - and 939 will definitely last at least a little longer before it's EOL. It'll probably be going strong for a bit into M2's life, as well as being the new target platform of lower-end Opterons.

Yea i meant the socket 940 as a desktop platform as I CLEARLY mentioned in my previous post heh.

 

[Hard Ball]

I agree with most of what everyone said. And I should have thought through more about FB-DIMM before posting. Come to think of it, FB-DIMM may stay entirely in the server/workstation market, and leave the desktop realm to DDR3.

Here are the facts that we are somewhat certain about.
--DDR-II will have a maximum of 24 months of life as the top-dog, before the introduction of DDR-III; so while I can see that DDR-II may be there for a long time on the mainstream-budget segment for a long time, I don't see how the performance-gaming sector would need DDR-II for much longer. The DDR to DDR2 jump in performance and features was not nearly as dramatic as the SDR to DDR (also RDRAM at the same time) transition; and DDR-III right now promises a much larger increase this next time around; so I think the performance desktops will migrate to it rather quickly. Perhaps there will be some K10 chips that might be able to use DDR2 and 3 or both DDR3 and FBDIMM during some transition period, that remains to be seen.

--AMD is seemingly eager to integrate PCIE control on die, which will further reduce latency of graphics traffic, and other I/O functions. And intel is also working on their CSI (or whatever it's called), so the entire trend in the industry is the continuing integration of former NB function onto the CPU die; K10 will almost certainly having something to that effect. With M2's 940 pin configuration, I just don't see it being able to handle that change, so I don't see M2 being used for K10 at all.

--K10 on Opteron 2xx and 8xx CPUs are projected to arrive Q406 or Q107; seeing how the transition to K8 went last time, there will likely be a time gap of about 6 months or more between the server/workstation and the desktop parts. I think that this is the reason why M2 was absolutely necessary, to compete with the increasing memory bandwidth of the Prescott and Yonah derivatives. So the desktop K10 parts are projected to arrive mid to late 07. You are right, it may be a full 10 months after the advent of the Conroe do we see K10s on consumer machines; during this time, I just don't see M2 being able to hold the fort and the performance crown; and Conroe should have a free hand on the consumer market for any where between 4-12 months. And AMD would want to move on as soon as possible, so they are not likely to have vigorous M2 support for long periods of time.

--K10 will initially have as endiannet said, 2/4 core versions, and "huge" caches, which no one really knows how much, plus higher clock speeds; and of course a number of architectural redesigns, probably the addition of some execution units, making it a wider core; although details are scanty at the moment. The improvement from S939 to M2 will be minimal, with a significant raise in TDP to well over 100W; while the improvement from M2 to K10 CPUs will be rather large, with an expected drop in TDP per core. So I think Conroe, and especially the later Whitefield will have an easy time man-handling any Windsor/Orlean derivative in both performance and power. So M2 ramping of speeds will likely be very limited, perhaps to 3.0 or 3.2 GHz; but beyond that, it will turn into another Tejas; so the need for K10 on desktops at a certain point in 07 will be imperative; otherwise, AMD will risk losing billions in sales during that year.

 

[n7]

Nothing is as funny as those who believe the next socket of a CPU will be the one to wait for.

It takes only a tiny bit of knowledge to know that waiting for a next gen socket is just foolish, since it too will be promptly replaced by another.

How can people not see this?!?

Let me point some AMD changes:
s462
s754
s939
M2
next one
next one
next freakin' one


How hard is it to see that waiting to upgrade is just silliness?

 

[valkator]

Originally posted by: n7
Nothing is as funny as those who believe the next socket of a CPU will be the one to wait for.

It takes only a tiny bit of knowledge to know that waiting for a next gen socket is just foolish, since it too will be promptly replaced by another.

How can people not see this?!?

Let me point some AMD changes:
s462
s754
s939
M2
next one
next one
next freakin' one


How hard is it to see that waiting to upgrade is just silliness?

I couldn't agree more. The speed that the PC industry changes at there really is NO perfect time to upgrade because something is just around the corner. Just pick a time to upgrade and prey that you made the right decision.

 

[BlingBlingArsch]

some of us dont have the money to jump on each train that comes around

 

[n7]

Originally posted by: BlingBlingArsch
some of us dont have the money to jump on each train that comes around

Then you ride your train till you have the money to hop on another train.

It's very simple.

 

[Unkno]

Originally posted by: valkator

Originally posted by: n7
Nothing is as funny as those who believe the next socket of a CPU will be the one to wait for.

It takes only a tiny bit of knowledge to know that waiting for a next gen socket is just foolish, since it too will be promptly replaced by another.

How can people not see this?!?

Let me point some AMD changes:
s462
s754
s939
M2
next one
next one
next freakin' one


How hard is it to see that waiting to upgrade is just silliness?

I couldn't agree more. The speed that the PC industry changes at there really is NO perfect time to upgrade because something is just around the corner. Just pick a time to upgrade and prey that you made the right decision.

I would also agree...IMO, the "best" time to upgrade is when your computer is too slow to handle new applications (or when you fried your cpu, either way) and you have some cash to spend.

 

[BlingBlingArsch]

Originally posted by: n7

Originally posted by: BlingBlingArsch
some of us dont have the money to jump on each train that comes around

Then you ride your train till you have the money to hop on another train.

It's very simple.

so whats the news then...

 

[bunnyfubbles]

Some people just don't get it I think. M2's DDR2 doesn't necessarily mean nothing good will happen. I've seen 2GB DDR2 kits cheaper than 2GB DDR1 kits. We won't get faster memory (although current A64s show that they don't exactly need it, not even the X2s), the good is that we'll get support for larger capacity - and cheaper per GB too. 2GB will easily be the "standard" amongst performance users when M2 hits, which is good for dualcore and multitasking. It should also help to make capcities beyond 2GB far more of a reality for those without very deep pockets.

I'm not flipping out for M2, but I'm definately not dissapointed in it, I can see some good things coming out of it and I'll most likely own an M2 system unless there's another socket change close on M2's horizon as s939 was for s754.

 

[xbdestroya]

Yea i meant the socket 940 as a desktop platform as I CLEARLY mentioned in my previous post heh.

Well, what I meant though is that you can use said Opterons on the 'desktop' 940 platform. Same registered memory, same everything. The boards don't support anything past the 100 or 200 series I think, but you can buy these Opterons off of Newegg if you want an upgrade/replacement chip.

 

[xbdestroya]

Originally posted by: n7
Nothing is as funny as those who believe the next socket of a CPU will be the one to wait for.

It takes only a tiny bit of knowledge to know that waiting for a next gen socket is just foolish, since it too will be promptly replaced by another.

How can people not see this?!?

Let me point some AMD changes:
s462
s754
s939
M2
next one
next one
next freakin' one


How hard is it to see that waiting to upgrade is just silliness?

Right... well, obviously. But that's really not what this is about. It's about maximizing price/performance. Spending however many hundred every couple of months for small percentage gains in performance is my idea of a waste; but at the same time you don't want to get left behind. I went socket 754 when I did because it was:

a) cheap compared to socket 939

b) performance-wise only lacked dual-channel (~5%)

c) AGP 6800 class cards cost less than PCI equivelents, and would last me a while

I *will* skip one of these sockets, be it 939 or M2 - or both - I just have to see which way the wind is blowing.

 

[Maximilian]

Im sticking with my socket 754. I dont need 939 or M2 as this comp will run bf2, itll run vista when it comes and itll run the latest games for a while to come yet, albeit not at the highest settings. Im just gonna keep upgrading this thing until ive maxed out my upgrade path or its no longer financially feasible to do so.

So ive already skipped 939, ill never use it, and i plan on skipping M2. In fact my next platform might be Intel, as their new architecture will be awsome and mop the floor with the current A64's. (speculation, i dont actually know if it will, i just believe it will)