AMD to Adopt DDR2 Next Year

Hack Jandy writes "According to Anandtech, AMD has already developed a new processor lineup for Athlon 64 processors with DDR2. The article states that internal AMD roadmaps indicate the processors should debut early next year and will require a new 1207 pin socket."

Comments

DDR2

[Anonymous Coward]

So we have to dance to get the damn thing to work?

Re:DDR2

[Anonymous Coward]

The Intel marketing people in colourful bunny suits has been doing their song & dances for years. It is time for AMD to catch up.

Socket 1207

[fr0dicus]

Will pincounts be the new megahurtz?

Re:Socket 1207

[Basje]

They need the extra number for the integrated PCI Express controller, integrated on the chip. Chips without that controller will have less pins.

I think they will try to keep the number of pins down: more pins is more expensive to manufacture and transport.

Re:Socket 1207

[Petersson]

Will pincounts be the new megahurtz?

My wallet hurtz... But I've discovered I don't need it, since after almost twenty years of gaming I've quitted and I use my Athlon XP system mostly as DVB PVR, video player, some video editing and dvd burning. What I ask for is better OS, not hardware.

BTW since all most of the controllers are on CPU, I expect motherboard prices will decrease since there is not much to remain on them. In extreme, why spend money on the processor pins and the socket itself? Why not solder the CPU to motherboard, like in the old times of some 386 boards? With adequate hole in mobo, we can cool the CPU from both sides.

Re:Socket 1207

[VitaminB52]

Yep - it's about time for a new CPU logo: Hedgehog inside :) .

Re:Socket 1207

[StarWreck]

An integrated PCI Express controller would solve the only down-side of AMD's current use of an onboard memory controller. That downside is added latency when using system memory for the graphics subset.

In AthlonXP and Pentium 4 architecture, data from the graphics card would only have to pass through the North Bridge to get to the memory. However, in current Athlon64 architecture it has to pass through both the North Bridge and the memory controller built into the CPU. This slows it down a bit.

Adding an integrated PCIe controller to the Athlon64 would get rid of this slow-down once and for all... however, this problem has not really been much of a problem because graphics cards have so much memory nowadays that they hardly ever have to access system memory anymore.

Aarrrrgh....

[Solder Fumes]

And here I was, thinking Socket 939 was going to be good for a LONG time, and bought a new motherboard....

Oh well, it's not like motherboards are the most expensive part of a computer.

Re:Aarrrrgh....

[NerveGas]

Personally, I never buy a motherboard with future CPU upgrades in mind. It's just not worth it, upgrading your CPU within the same general architecture rarely gives you much real-world performance.

The real performance boosts come from radical architectural changes - new memory subsystems, new processer types, new interconnect, etc. - and for any of those, you're going to need a new motherboard, period.

steve

Socket A

[tor528]

What happened to the good old days, when pin counts lasted years and years?

Re:Socket A

[orz]

They integrated the memory controller on to the CPU. Now, every time they switch memory technology (DDR -> DDR2 in this case), they have to switch CPU sockets also.

Re:Socket A

[zakezuke]

What happened to the good old days, when pin counts lasted years and years?

Ummm... we have socket 754 which is pretty much supports the older athlons as well as the Sempron and Athlon 64s. This is common on those sub $400 pcs that you find on the retail circuit.

It's sort of like the old super socket 7 of old. Nice the fact that AMD offered CPUS as fast as 450 and 550 IIRC, even a tad higher than 450mhz in the amd k6-3 mobile if you were lucky enough to find them. While not nessicarly the best upgrade choice they are not only an option for the budget minded but most importantly those last generation high end chips either hold their value or increase in value. The socket-7 run would have been limited to 233/266mhz had it not been for AMD.

But this is all accidemic as well... Socket-a has been around for a very long time... offering speeds as low as the 600mhz... as high as 3.2ghz AFAIA. More over they are still in production.

So what happened to the good old days when pin counts lasted for years and years? They are still here, and in fact improved thanks to AMD so long as you ignore Slot-A. While I would strongly reccomend going 939 if you can, 754/slot-a is still an option even for those who gotta have 64bit CPUs. Just like the end of the 21st century when you "could" go slot-1 or stay with (super) socket 7 a while longer, or hell even 72pin simms if you really wanted to.

Re:Socket A

[Shanep]

I forgot to mention that at least slut A always comes with at least 3 ports. The others come with a maximum of 2, but usually only come with 1 unreliable port which is hard to get working. These common units tend to be expensive to run and also come with an output only port which is very noisy.

1207 and Virtualization

[StarHeart]

The blurb mentions 1207, but the article only talks about M2(940). I have read mention of 1207 in relation to chips with the PCIe controller onboard. But not signs of tha in this roadmap.

This roadmap seems to suggest at least that virtualization will only come in chips with the M2 socket. I will be disappointed if that is true. I had planned to upgrade to dual core chip with virtualization, but keep my 939 board. Maybe by then I will be looking to upgrade to PCIe and won't care. I have an AGP board now.

Re:1207 and Virtualization

[orz]

The article does mention a 1207-pin "Socket F" in a footnote at the end. But yeah, the submitter got that detail wrong.

Re:1207 and Virtualization

[cartoon]

Well, look no further:

http://www.anandtech.com/mb/showdoc.aspx?i=2471

It's a full AGP x8 implementation, not a pokey AGP-thru-PCI or something like that. In fact, at present AGP is faster than PCIe on it... But read the review for yourself.

Re:1207 and Virtualization

[kesuki]

Games right now dont even use 1/2 of the banwidth on a AGP 8x slot.

Right because we all know a $400 GPU that has 3x the transistors of the main CPU can only be used -- For Games. PCIe was designed to take care of the 'limitations' of AGP, manyly that is was a one way street, ramp up the ability to push data to the card, and ignore the ability to send data back on the return trip.

PCIe Is required for card to card intercommunication, AGP simply doesn't have ability for the cards to send enough data back down the bus for 2 cards to work in parallel on the same graphical computation.

Considering how powerful and advanced modern GPUs are a bus system that allows them to return data to the main GPU is very important, because it allows 'other' programs to use them, programs that aren't 'primarily' concerned with outputting to the 'screen' but rather to a file, such as 3d-rendering software, image manipulation programs, mpeg-4 compressors, even audio effect software can utilize the GPU, because while it's designed for calculating graphics certain audio computations can be run on them faster than on a general purpouse cpu.

PCIe is better, it makes GPUs more valuable to non-gamers, and frankly gamers while a lucrative market segment are not the 'whole' picture. Not by a long shot. Many graphic professionals pay $1,000 or more for a 3-d card that can perform the rendering work in hours that a normal CPU would take weeks to compute.. So since PCIe works for gamers too, it would be more profitable for the graphic card makers if PCIe was the prevalent and dominate technology and they didn't have to build 'high end' agp cards anymore..

PCIe too

[Max Romantschuk]

According to the Inquirer AMD plans to integrate PCI Express as well. This would be very nice indeed, but I guess it's not exactly press release grade information at this point.

Not yet.

[aaron_ds]

1207 pins, pffft!

I'll hold out for the 1337 pin AMDs.

I for one welcome our elite cpu overlords.

Transitional moment

[Anonymous Coward]

I really am a geek now.

I read that as AMD would be adopting Dance Dance Revolution 2. XD

Re:Transitional moment

[iwan-nl]

If you think ddr2 the game is more geeky than ddr2 the memory, we must have a really different definition of geekness.

Of course they changed the socket... again...

[OlivierB]

They make a living out of this!

I am an Apple kind of guy.
When I switched a couple of years ago, the thing I was most upset about was the inability to upgrade my system myself.
I was afraid that with Macs I would be locked in the hardware and would have to upgrade the whole machine when I needed an upgrade. Well that's true: if you want to upgrade the CPU on your Mac you have to change your machine (Ok you could maybe buy some "overdrive" for your Mac).
Well on x86 it's the same thing!
Theoritically you could swap out your processor for a faster one, but the average production life of a CPU socket is LESS than the average time you use a CPU before thinking about upgrading it.

So on x86 when you think about upgrading that 2 year old CPU to something new, well the pin layout has changed and you need to buy a new motherboard, with new type of Ram, and now new components (SATA, PCI-X etc...)
Although you could change all these components idividually, you must admit just changing the whole machine is often a better deal.

I highly suspect intel has a built-in incentive to do so as they produce chipsets for the motherboards, and most of the chips in the new parts involved when "upgrading".

Upgrading no longer exists, it should be called "changing-my-machine".

Re:Of course they changed the socket... again...

[eddy]

So on x86 when you think about upgrading that 2 year old CPU to something new

Stop blathering. On Socket-A I went "Duron 800", "Athlon 1333" and then "AthlonXP 2400+". That's three processor upgrades on one platform and the VERY SAME MOTHERBOARD.

Just recently I bought a new Socket-A MB (they're dirt cheap) and 2GB of DDR (which is similarly dirt cheap), so the last CPU has seen a MB upgrade too.

And do you know what? I play modern games on that sucker.

(I'll go S939 soon with a nice Venice and get a real use for all that memory)

Re:Of course they changed the socket... again...

[richlv]

this is happening through whole industry (and all other industries as well) - artificially reduced life span for everything.

it is possible to design compatible, long lasting technologies - but that would reduce profits. there is no long term thinking for consumer goods - of course, that's bad for environment, but who cares if only short term profit is ok ?

i've been thinking about a new computer - but when i try to select some fundamental parts to build around, i start to doubt. what disks shoulod i choose ? sata ? how long will this standard last ? what type of memory ? what motherboard factor ? what type of extension cards (ethernet, videi, sound etc) ?

for me also a big problem is interoperability - previously it was easy to test components by swapping them - change video, network, psu and check that mb again. also getting some system on feet fast was a lot easier because almost all parts were identical.

now there's a problem of finding the correct keyboard (because when i find an at, i need ps/2 and vice versa), finding correct ram is almost impossible - soon video and other cards will be incompatible with existing sets.

what about dvd* ? how long will which standard last (as can be seen, soon there will be another...) ? which of them will be the best bet for backwards compatibility ?

i plan to buy a new computer after a couple of months, but i am heavily puzzled wether that is a good idea (i'm already sure that's a bad investment ;) ).

Re:Of course they changed the socket... again...

[iPod is UNIX]

I learned early on never to take advice from an Apple user, they just make arguments for Apples current product line adjusting them as Apple changes directions. I'm going to present some of the most mindbaffling arguments from the Apple community that you may check with other sources and find out they are pretty much right on.

Apple products and Apple users arguments:

• The Newton, try to convince the Apple user this never was a very good PDA and by todays standard is totaly "out there", 8" x 5" x 1" inches and about a pound without batteries, for reference a palm is about 5" x 3" x 0.3" and about 0.3 pounds. The newton is still by many Apple users the PDA to have. Now ask the same Apple user why the iPod is much better then a Creative Zen. The Zen is to heavy, by 0.1 pounds.
• The time around 2000 when Apple users where still making arguments for cooperative multitasking which to the rest of the industry was pathetic and laughable. As Mac OS late but surely got preemptive multitasking by copying it from FreeBSD the Apple users finaly could lay the cooperative multitasking argument to rest. Not because preemptive was better (which it of course are) but because they had it too.
• The early stages of OS X (which really where an open beta), slow kernel, slow UI and not even easy to use. To the Apple users was of course the best thing. In reality it was so bad Apple don't even offer security patches for those machines even though they are just a few years old.
• The G4 cube. A bastardised computer, impossible to use. You needed to stand up to load a cd in the tray (top loaded). You had to turn the computer upside down to connect peripherals (all connectors was at the bottom of the case?!?). It had heat troubles taking down most of them. Of course by the Apple user touted as a marvelous piece of equipment and even today by many Apple users seen as the height of Apple design and innovation.
• The Mac Mini, we haven't seen the last of this yet I'm afraid. Of course by the Mac users seen as the future of Macs. Reality: Apple are in 2005 selling computers with 1.25ghz CPU and 4200RPM drive for $499, this excludes keyboard mouse and monitor and includes not even enough RAM to run the included operating system. If you could buy a similar spec PC (which you can't because there are no that slow) you would get at least keyboard, mouse and monitor. It will probably not take long before a hoard of not very happy Mac mini users put these to rest when they find out you can't even run todays software reasonably on a new computer, and tommorows will be next to impossible. The argument from the Mac crowd is that if you buy a Mac mini to play games you are stupid. Is there any other software for the Mac mini I must be stupid to try running?
• Unix, first let me explain that OS X is not a certified Unix. Unix is a trademark hold by Open Group and Apple is using the trademark without permission. Certified Unixes includes Solaris, True 64 HP-UX and other Big leage names. To an Apple user Unix has always been something weird and strange and generaly bad, the usual "not invented by Apple syndrome". Now the Apple user tells you he has a Unix too and Unix by now is the greatest thing thing sliced bread. A real life story was the Apple user who told me "All modern science is based on Unix", that tells you how much the typical Mac user knows what is under the hood of their computer. They tell you Apple is the largest supplier of Unix world wide. Of course OS X doesn't even remotely classifies as Unix and recent test has shown it is at least 10 times slower then Solaris on simple database serving. This of course gives Unix a bad reputation so you can imagine Open Group being more than upset (they have of course sued Apple over infringement). Real Unixes also has 8-

Re:Of course they changed the socket... again...

[Rich0]

Although you could change all these components idividually, you must admit just changing the whole machine is often a better deal.

However, keep in mind that a computer is only as fast as its weakest link. If the CPU can go up to 4GHz, but the memory bus is so slow that the CPU sits around all day waiting for data, then you might as well save money on that CPU.

Hence, motherboard technology and CPU technology tend to improve hand-in-hand. Ditto for memory speed.

Even if you could double your CPU performance on the same motherboard, would you want to? The only exception would be if you started out with a well-below-cutting-edge CPU to begin with - in which case you could upgrade to what was top-of-the-line at the time and your motherboard would probably still be fine.

In any case, I still find it cheaper to upgrade my own PCs. Why?

1. Keep the optical drives, case, floppy drive, video card (if you're a non-gamer), etc. (Maybe even the power supply if you have a decent one.) That right there is a few hundred dollars saved over anything Dell will sell you.

2. Get a decent motherboard. Performance tests have shown that spending an extra $10 on a motherboard can often improve performance SUBSTANTIALLY (moreso that spending it on CPU/RAM). Big brands know that consumers don't ask about motherboards, and they don't look at benchmarks, so they save the $10. Don't make the same mistake. Likewise, a decent motherboard will often give you technology like firewire/extra USB/etc for literally $10 more. That saves quite a bit when you don't have to run out to walmart for a USB hub.

3. Be able to overclock if you want to - in general a retail motherboard/BIOS gives you far more control over the computer. Don't be a ricer, but feel free to experiment with timings that improve performance on your hardware without wearing things out.

4. Get the components you want. Thinking about dabling with linux? Well, check the compatibility lists and you don't end up with that motherboard which was popular with Dell or Compaq or whoever which has some random glitch with linux. ACPI is nototious for being broken, and if you are using features like powernow you want a system that isn't glitchy. Check out IRC for people groaning...

Honestly, I was surprised that I actually saved money building a PC myself - Dell has access to all kinds of deals that I don't have access to. On the other hand, Dell only sells "genuine" Intel, which handicaps them. Plus, Dell forces you to toss your perfectly good case, floppy, opticals, etc. Oh, you also get penalized for not buying a monitor, and I hope you like your growing collection of keyboards/mice/etc... You may not end up being far ahead on price, but often your performance ends up being better.

Just some random thoughts...

Re:Of course they changed the socket... again...

[tomstdenis]

Just add a bit...

The PPC [G3] is about as efficient for bignum math as the AMD Athlon-XP and the P4 ALU. See this chart for instance.

I think the G4 maybe be slightly better [the instruction set remains the same w.r.t. bignum math] but still same ballpark.

The problem with that design really is that while it has a good RISC ISA and lots of registers it's simply not meant for math. You have to execute two 4-6 cycle multiplies to get one 32x32=>64 product whereas other cpus can get the full product in 6 cycles AT A HIGHER FREQUENCY.

Where the PPC is nice though... is in embedded work. It's not quite as efficient as the ARM but can be clocked higher. So it's a good tradeoff between something like an ARM which gets enough MIPS and way low power and an Athlon which gets very high MIPS but takes a lot of power [by comparison].

All in all though I think the PPC can hold it's own against the x86 offerings. It just doesn't scale as nicely.

I don't know about the G5 though [I don't have access to one]...

Tom

First Prime Factorization Post

[2*2*3*75011]

1207 = 17*71. I wonder why this beautiful factorization isn't mentioned in the article.

Re:First Prime Factorization Post

[Anonymous Coward]

By your reasoning, the AMD geode LX800 takes the crown:

481 pins = 13*37

OMG, 1t are t3h l33t CPU!!1!!11one11

http://www.linuxdevices.com/files/misc/amd-lx800.j pg

It doesn't matter that they change the socket.

[Ihlosi]

Really, it doesn't. Since the memory controller is integrated in the CPU, there is no way to make one of these things run in todays DDR1 mainboards, regardless of the pin count. And since DDR2 has a different pin count than DDR1, of course the pin count of the memory contoller has to change, hence the pin count of the CPU changes.

Anyone complaining about "yet another socket" apparently hasn't understood this.

great

[axonal]

Great, more fat kids falling off the dance floors.

Dominoe effect

[eddy]

I forgot the best part; my old Socket A MB (KT7A-RAID) ran on SDRAM of course, so when I upgraded that MB, I moved 512MB of SDRAM to my old linux server (K6-2 based FIC-503+), which up till that point was running on 80MB of EDO.

Now that's what I call an [cheap ass] upgrade.

DDR2 ?

[Arthur B.]

That is great news for Reiser4

(SELECT /.ers WHERE CLUE > 0)

Re:Offtopic

[TheRaven64]

This comment is informative[1]. If you have moderator points, please apply the appropriate moderation to it.

[1]in a self-referential kind of way.

Re:Is DDR2 worth waiting for?

[Anonymous Coward]

If you want to use your AGP-card then 939 is the way to go since there will be almost 100% PCI-express graphics on the new sockets, and since you can reuse your ram with good performance (DDR2 in itself won't give you much extra) go for broke right now

Re:Oh, so now we're getting into an oldest...

[TobyWong]

I have a rusty abacus at home that I use for my banking. My wife wanted one too after she saw me flying through my interest calculations but I told her "we don't waste money on newfangled technology" and bought her a pile of shiny beads to do her counting with instead.

Re:Oh, so now we're getting into an oldest...

[carl0ski]

have you tried the muliplier of 6 on that k6-3?


some k6-2+ and k6-3

let you set the muliplier to 2x and

it assumes it is 6x

very handy i once had a k6-2 at 6x83 on a p1 board maxed out :)

it retired 2 years ago

AMD Tech Tour Q and A with engineer

[valhallaprime]

Well, I am freshly back from the AMD tech tour event in East Brunswick last night, and this specific question came during the Q and A with "TEH EXPERT5" - The question of DDR2 support.

The actual engineer on staff at the event answered it, and stated flat out that there was no performance gain until at least DDR2-667, and that alone "was only about 5% or so faster than DDR400 running in dual channel mode". He even went so far as to say that "DDR2-533, with it's increased latency over DDR400, has a negative impact of OVER 5%", and makes no sense to jump to. This was because of the efficiency already inherent in the HyperTransport bus, according to him.

He talked for about 5 minutes on the issue, and the gist of it was that until DDR2-667 specifically started to become more affordable, the incremental speed boost didn't make any sense for anyone, including and users and AMD Proc Support.

Incidentally, he also mentioned that DDR2 would (of course) require significant redesign in the built-in memory controller of the 939 chips, unless registered memory was used. This sorta implies in a friday morning-drove-all-night-from-NJ way that the current 939's would not support DDR2 if there were to be 939 mobo's with DDR2 support.