AMD chip

[micker377]

From "simmtester.com":

AMD to launch new Socket M2 processor


Tuesday, November 08, 2005

AMD plans to launch a series of 940-pin Socket M2 CPUs – a dual-core processor codenamed Windsor and the single-core Orleans and Manila processors – in the second quarter of 2006.

As a result, AMD’s Socket-939 processors are expected to be phased out one year after the release of the Socket-M2 CPUs, indicated the makers.

Except for the Manila series, the Windsor and Orleans lines will integrate AMD’s “Pacifica” virtualization technology, the makers noted. The dual-core Windsor CPUs will be part of AMD’s Athlon 64 X2 line, whereas the Orleans will be featured in the Athlon 64 series with the Manila to be part of the Sempron processor line, according to the makers.

AMD has started sampling its Socket-M2 CPUs with its customers, said the makers, adding that the actual release will not kick off until the second quarter of next year, a little later than expected. As reported on October 17, DDR2 demand was expected to grow significantly in the first quarter of 2006, buoyed by the launch of AMD’s Socket-M2 processors during that period.

With the upcoming release of AMD’s new Socket-M2 CPUs, Socket-939 CPUs are expected to be phased out by the chipmaker in the second quarter of 2007, according to the motherboard makers.

 

[cdogg]

Oh, well that's nice! Let's see, in the past 3 years, we've seen Socket 754, Socket 939, Socket 940, and now Socket M2 (which also has 940 pins).

Did these guys forget that the everlasting Socket A platform is what sealed their rising popularity over Intel? Changing the platform this often takes away any advantage they once had over Intel which has been playing that game for too long.

If they stick with the M2 for at least 4-5 years, I might reconsider my take on this. But until then...



~cdogg
"Insanity: doing the same thing over and over again and expecting different results." - Albert Einstein
[tab][navy]For general rules and guidelines to get better answers, click here:[/navy] faq219-2884

 

[jisoo23]

I agree with cdogg. I was thinking about moving to a socket 939 after several happy years on a socket A, now I'm not so sure anymore. Bah!

 

[PRPhx]

And here's some more to add to the mix. AMD is planning on 4 core Opterons by 2007! There is a story in the Nov. 18 PC World detailing AMD's plans for the next couple of years.
Here is the link:

http://www.pcworld.com/news/article/0,aid,123559,00.asp

 

[cdogg]

Now that I've taken a couple deep breaths and read up on the issue...

If you still have Socket A, you can wait 6-9 months and go with an M2 platform, but it's not really the best option. The 939 CPU's are still going to be made up until at least Q2 2007, which is plenty of time to get a fast 939 CPU now upgrading a year and a half or so down the road with one of the last 939 models. Doing so will stretch the life out of the 939 well into 2009.

M2's main advantage is DDR2 technology along with some other minor chipset updates, but it's real attraction is that it will likely last as long as the Socket A 462 platform did.


My prediction: Socket 939 is here to stay for a couple years. Don't expect it to fade away until early 2008.

~cdogg
"Insanity: doing the same thing over and over again and expecting different results." - Albert Einstein
[tab][navy]For general rules and guidelines to get better answers, click here:[/navy] faq219-2884

 

[Sympology]

We all complain about socket upgrade, but it's a catch 22 situation.

We all want the fastest processors, but are we willing then to run this superduper processor with old memory, old chipsets, etc etc.

After all take a mobo from just a few years ago.

USB2?
DDR2?
PCI-X?
64 bit support?
Bluetooth
802.11g?

At the end of the day, if your going to perform a upgrade cycle every few years, the chances are they'll be new technology you'll want to go with it.

Unified Display Interface anyone?

http://www.channelregister.co.uk/2005/12/21/unified_display_interface/

Only the truly stupid believe they know everything.
Stu.. 2004

 

[ceh4702]

So where is the link to the article you were reading?

If you do not like my post feel free to point out your opinion or my errors.

 

[Sympology]

CEH4702...lost me...

The link is about yet another new standard to highlight how mobo's and new technogies are coming out, hence having a new socket variation is not a big an issue as made out for many people. After all if you are likely to be spending money on the latest and greatest processor, you are often likely to want the latest and greatest add ons.

I guess you could reverse this and say "If I want the new stuff, I still want to use my old processor", but at the end of the day we all no there is no such thing as "future proof"

For example I'm running an Athlon 1.8XP so if I was to upgrade, I'd, like my mobo to support new technologies that have arrived since then, so a socket upgrade is not an issue.

It's no different than cars. They keep sticking new bodywork and engines onto old platforms, but eventually the platform becomes so out of date, the whole thing has to be redesigned from bottom up.



Only the truly stupid believe they know everything.
Stu.. 2004

 

[cdogg]

Stu,

Point taken. However, there are some examples where that's not always the case. Nvidia's original Nforce chipset, for example, allowed you to run three generations of the Athlon XP - the Palomino, Thoroughbred "A", and early Thoroughbred "B" versions. The memory requirement for all 3 was DDR 266 (PC2100), and speeds ranged from 1333MHz all the way up to 2.2GHz+.

The Nforce2 chipset gave you all these options, plus the Barton and Sempron cores that supported up to 400MHz FSB. Of course the difference in this chipset, was the need for DDR 333 or DDR 400 in some cases depending on the CPU.


Having said all that...

I don't think the average home user will upgrade the motherboard just because there is a new feature (like USB2, PCI Express, etc in the past). Most consider this a secondary benefit. Speed is most important. The ability to upgrade from an Athlon 1500+ to an Athlon 2600+ was a stroke of genius by AMD sticking with the Socket A platform for so long. Squeezing just an extra couple years out of a PC rig with just one or two upgrades means a lot to the mainstream. Socket A had what Socket 754 never did, and what Socket 939 seems to be losing - longevity.

So if you look deeper past all the "whining" from some people, there's a good argument that the M2 seems to come at an odd time. I see no reason why they couldn't add DDR2 support for the Socket 939 chipset...


~cdogg
"Insanity: doing the same thing over and over again and expecting different results." - Albert Einstein
[tab][navy]For general rules and guidelines to get better answers, click here:[/navy] faq219-2884

 

[Sympology]

Of course it may be just a case of AMD chopping the current chipset short in order to extend the life of the next gen.
It may of course be a problem that they have with the current set up that hasn't reared it ugly head, maybe it's unstable with DDR2, it may the next gen of processors run really badly on the current hardware. Most chip makers are a couple of years ahead in development and it's these "new" setups that are causing the headaches, not the current, or near future products.

At some point they have to say bye to the old hardware...of course I was ready to cry the other day when I rememebered my laptop had no serial ports when our PBX went TU and I didn't have the drivers installed for my USB to Serial adapter...well that's progress....lol...



Only the truly stupid believe they know everything.
Stu.. 2004

 

[cdogg]

You made a reference to a chipset having an OLD platform and having to be redesigned or updated. Well, at what point does it become old? 754 didn't even last a year, and 939 had about a year and a half under its belt before its successor was named. Socket A? It lasted well over 5 years.

Also, you can change the chipset without changing the socket. Nforce to Nforce2 is an example, where new features were added like Dual-channel DDR, faster FSB, USB2, etc. But Socket A remained.

So I think you are confusing the chipset with the socket interface. They are separate beings...


~cdogg
"Insanity: doing the same thing over and over again and expecting different results." - Albert Einstein
[tab][navy]For general rules and guidelines to get better answers, click here:[/navy] faq219-2884

 

[Sympology]

cdogg I see exactly where your coming from with Socket A lasting 5 years yet 754 lasting less than a year. I'm not technical down to a component level, but it maybe just a case that Socket A was so well ahead of processor deveopment it was good platform for many years. But currently processors are going through a similar evolution that made the jump from 486 to Pentium.
Whereas in the early Athlon days effectivly all that was added was more speed and extensions, now we are getting Dual / Quad cores, 64 bit processing etc etc. It may be a simple case that AMD are just developing new technologies at such a rate, that the foundation on which these are based bring up headaches they could do without.

As for what is old? In the PC hardware world about 2 weeks isn't it?
Many say 18 months is when something is "old" in computing terms, so by that my pc belongs in a museum. I thought it was still pretty good until I set up my nephews Athlon64 laptop and then i cried....



Only the truly stupid believe they know everything.
Stu.. 2004

 

[cdogg]

I'm sure that you are right in saying that AMD scrapped the 754 and the 939 for a reason, perhaps a design flaw.

But it is not "the reason" that concerns me, because given recent history, we know that a good design that lasts several years is possible. And again, you should be able to keep the socket while changing out the chipset. Often, it's only the chipset that needs updated.

The only beef I had originally (and felt relaxed on until today) is that I think the company needs to carefully consider the effect this could have on the mainstream. One of the reasons why I made the switch to AMD a while back was the long life of the socket A platform versus what Intel had to offer. In addition to costing more, Intel had a much shorter upgrade path.

Old in PC terms depends on what component you are referring to. The socket interface doesn't apply to being obsolete that quickly. Now the PC as a whole? Yeah, that's about 2 weeks!


From a technical standpoint, I understand that there are some things that are beyond your control, and you may be right about the dual/quad cores playing a role in it...that's why I'm hoping they get it right this time and that the M2 is here to stay for a while.

~cdogg
"Insanity: doing the same thing over and over again and expecting different results." - Albert Einstein
[tab][navy]For general rules and guidelines to get better answers, click here:[/navy] faq219-2884

 

[kmcferrin]

So if you look deeper past all the "whining" from some people, there's a good argument that the M2 seems to come at an odd time. I see no reason why they couldn't add DDR2 support for the Socket 939 chipset...

Actually, there are very good reasons why.

The ever-changing socket format is extremely frustrating, but it is an unfortuante side-effect of the processor designs that AMD have implemented that allow them so much performance. The inclusion of of an on-chip memory controller means that there is another piece of hardware that needs to interface with the mainboard, hence more pins. That's why there was originally the socket 754 Athlon 64s and socket 940 Opterons. The biggest difference in the Opteron was that they supported SMP and dual channel ECC memory, hence the additional pins. Then S939 came around for the desktop, offerring dual-channel but not ECC. Hence another socket. The M2 will be a different configuration yet again because the memory controller will be different to support DDR2. There are also plans for AMD CPUs with even more pins (1207 comes to mind), that again support different memory configurations.

So every time you want to switch memory technologies, you end up having to upgrade to another MB and CPU. Still, the form factors have had significant lifecycles. Socket 754 is several years old today. It started with 2500+ range Athlon 64 CPUs, and topped out at the 3700+. Now that it is the value line there are still Sempron lines that use it, so it is still somewhat viable. Socket 939 has been around for 12-15 months, and AMD intends to still produce S939 CPUs for at least another 18 months. That's almost 3 years. It started at 3000+ Athlon 64s and has so far gone to 4800+ dual-core Athlon 64s. It will probably go quite a bit farther in the next couple of years, so S939 has quite a long life.

The other thing to remember about Socket A (462) having lasted so long, is that it still wasn't a seamless transition from one Socket A CPU to the next. If you bought into Socket A when it was first released (with the Athlon 1GHz CPUs) and followed it to it's end (the Athlon XP 3200+ CPUs) you will have had to upgrade your motherboard several times to support various cores/FSB speeds/voltage requirements/etc. So while the pin count and form factor stayed the same, there was very little "compatibility" there.

If AMD had gone with the old standard northbridge chip-based memory controller, they could have probably continued to reuse a single socket form factor for all of the Athlon 64 CPUs of the past couple of years, since the changed functionality would be on the northbridge, but you still would have to upgrade your mainboard. This is the way Intel has done it. Unfortunately, the trade-off is reduced memory bandwidth/performance, as you will have an FSB in this arrangement and memory bandwidth is limited by FSB speed. Even now with the 1066 MHz FSB, Intel is only just catching up to the amount of memory bandwidth that has been available on Athlon 64/Opteron lines for quite some time. But it should be noted that even the 1066 MHz FSB doesn't scale nearly as well as the AMD solution in multi-CPU arrangements, as all CPUs share the same FSB/memory bandwidth.

 

[cdogg]

kmcferrin,

OK, first I want to mention that you probably missed my second post in this thread. It was the point at which I calmed a bit and realized that Socket 939 was going to be around for another 2 to 2.5 years. It's probably been hard to keep up with new information and arguments being brought up in almost every post, so I can understand overlooking that!

At this point, the conversation has evolved beyond the anxiety surrounding the release of the new M2 platform. I'm okay with it, as I think others should be as well (for reasons stated above). But now we've moved into the hypothetical as to "why" a change is made in the socket platform to begin with.

M2 will be a different configuration yet again because the memory controller will be different to support DDR2. There are also plans for AMD CPUs with even more pins (1207 comes to mind), that again support different memory configurations...

I'm sure there are many reasons like you opened your argument with, but memory architecture changes is not at the heart of it I'm afraid. The original Socket A went from SDRAM to DDR, then to Dual-Channel DDR. Therefore, it would seem that the memory architecture is not the MAIN factor in why the socket platform changes.

You do have a good point about the history of Socket A. When the memory architecture or FSB changes with the CPU (even if the same socket is used), you still have to upgrade your board in most cases. That is true. However, if you look at my example above (in post 21 Dec 05 11:08), I gave an example of where it was possible to jump as much as 700MHz (half the current CPU speed) on an Nforce board just by upgrading the CPU and applying a BIOS update. But I realize that chances like this are rare, and it's hopeful at best that we'll get a stretch like that again!


I don't design or build motherboards. I don't design or build CPU's. So I think I will end it here as to "why" the platform changes when it does. Common sense tells me that there are many factors involved - protocols, voltages, etc. - especially now that multiple cores are being used.

~cdogg
"Insanity: doing the same thing over and over again and expecting different results." - Albert Einstein
[tab][navy]For general rules and guidelines to get better answers, click here:[/navy] faq219-2884

 

[ceh4702]

Well AMD states that the multi core processors will work in the 939 and 940 socket with a BIOS Update.

If you do not like my post feel free to point out your opinion or my errors.

 

[kmcferrin]

I'm sure there are many reasons like you opened your argument with, but memory architecture changes is not at the heart of it I'm afraid. The original Socket A went from SDRAM to DDR, then to Dual-Channel DDR. Therefore, it would seem that the memory architecture is not the MAIN factor in why the socket platform changes.

It actually is one of the main factors in this case. With socket A, the memory controller was not on the CPU, it was on the northbridge chip. You had a front-side bus that connected the CPU to the memory controller on the northbridge. The same bus connected the memory modules to the memory controller on the northbridge. If you wanted to go from SDR to DDR memory on Socket A, you put in a new northbridge chip (with a new DDR memory controller), but the CPU and memory still use the FSB to communicate with the northbridge.

Socket A CPUs were memory agnostic. They didn't care what kind of memory you connected to it because all the CPU ever interfaced with was the northbridge memory controller. So as long as the FSB interface to the northbridge stayed the same, you could pair it with SDRAM, DDR SDRAM, single or dual channel, DDR2, RDDRAM, or whatever you wanted, so long as you could find a northbridge chipset with a memory controller that supported it.

Athlon 64-based CPUs do not have a northbridge, so the old model that applied to Socket A is gone. The CPU is no longer memory agnostic, it is memory specific because the memory controller is built into the CPU. Therefore to get a new memory interface technology (DDR2, dual channel, etc) will require a CPU upgrade. Also, since there is no northbridge or FSB, the memory modules will have to communicate directly to the memory controller on the CPU die itself.

So on the original Athlon 64 the original interface was 754 pins, using 184-pin memory on a single memory channel. They went to a dual-channel memory controller on the S939 CPUs (or S940 Opterons, which always had dual-channel), which requires pins for another physical interface to the second bank of the memory. So how many additional pins are needed? At least 184, because that's how many pins the memory use to interface (i.e., 184-pin DDR SDRAM). 754 + 184 = 938. So there are a couple extra pins for additional data (for example, ECC on the Opteron).

The new socket M2 CPUs will use DDR2, which requires a new memory controller and a new physical interface. DDR2 uses 240 pins to connect instead of 184. So you would expect the pin count to increase. They apparently did some other re-engineering of their CPU so that they could keep the pin count the same (which allows them to take advantage of the economy of scale on certain physical parts/processes (i.e., the plastic socket design, the actual CPU mounting an pinout process, etc). But even so, the actual functions of specific pins will have changed, meaning that though the pin count is the same they are not compatible.

 

[micker377]

Gee, I'm almost sorry I started this thread! I was thinking of my next upgrade, and was looking at the 939 chip. I thought the news of the new chip might make me wait for a while. I too am using the xp2200. I was thinking of what it will take to go to 64 bit, I guess I'll wait until they have 64 bit all straitened out, and see what CPU they are using at that time. Plenty of information here so far - I learned a lot! (as usual)

 

[cdogg]

kmcferrin,

Yes, I admit that I overlooked the integration of the memory controller into the CPU. That alone would account for why the socket change is needed for any memory controller upgrade.

However with DDR3 on the horizon (available as soon as mid-2006), it would seem that the jump to socket M2 isn't very significant until benchmarks show otherwise. Also, DDR2 has only performed marginally better than DDR 400 (which may or may not be due to limitations within the P4 bus architecture). Still, the Hypertransport bus is full duplex. So it might stand a better chance to take advantage of DDR2's extra bandwidth. It's hard to really tell at this point.

Thanks for all the info...definitely some food for thought in here!

~cdogg
"Insanity: doing the same thing over and over again and expecting different results." - Albert Einstein
[tab][navy]For general rules and guidelines to get better answers, click here:[/navy] faq219-2884