Our source claims that AMD's Socket AM2 delay is actually the by-product of "engineering change at (the) last moment". He claims his company has to redesign the PCB more than "others", since they need to support DDR-II 800 and DDR-II 1066 memory standards.
First of all, whether that was the sole reason for the delay or not, there is something wrong with the memory controller which has been independantly reported by The Inquirer, THG, and Anandtech. DDR2 800 support was probably added because the delay to fix the memory controller gave them time to, not adding DDR2 800 gave them the delay. In any case, DDR2 800 was always going to be added to AM2, it was just a matter of when.
Now in terms of whether DDR2 1066 will be supported. My personal view is that it probably won't be at launch. Going from DDR 400 to DDR2 800 is already a massive bandwidth increase and I doubt going to DDR2 1066 will make much of a difference in comparison. There is also no reason for AMD to be using up their revision headroom so early. Obviously DDR2 1066 will be added later, which is why they are telling motherboard makers to over-design now, but I'm sure most mainstream to high-end makers would have already done so anyways.
DDR2 1066 will probably be unofficially supported by the OMC much like DDR 500 is supported now.
Looking at that article again, makes me wonder how much of an improvement AM2 will really bring. It's an age old debate of course, but realistically the major advantage of AM2 is DDR2 whose main feature is more bandwidth. I just don't see K8 being bandwidth starved right now, which is of course the feature which AMD makes fun of Intel (in Netburst at least) for. If K8 was so bandwidth starved, it would have gobbled every bit of additional bandwidth going from DDR 400 to DDR 480, but in reality there was only a 3-5% performance increase on the X2 4800+.
Based on the tests that we’ve seen here today, AMD’s reluctance to move to higher bandwidth DDR2 offerings makes a lot more sense. The plain fact of the matter is that at the current clock speeds at which the Athlon 64 and X2 line are running, most desktop applications see virtually no benefit from higher bandwidth memory.
I bet the Socket M2 with DDR2-800 or -1066 are not really for the K8 line at all. I bet that AMD is looking at a new upcoming architecture that will need to use that bandwidth. It could be four cores, a 65nm series of chips that scale up to 3.5GHz- who knows. I do know that AMD likes to use its sockets a long time, so that's why they are doing this- in my opinion.
As I am sure you are aware, an increase in chip speed generally utilizes a proportional increase in memory bandwidth. You probably also know that a dual core chip requires up to twice as much bandwidth as a single core chip, at the same speed. (much more so with servers)
Since Amd will be putting out faster dual core chips than the FX57, it only makes sense to build a platform that is capable of "future development"
I seriously doubt the usfullness of high latency DDR2 1066 at this point, maybe next year.
I'm sure Amd will need every bit of umf, if conroe turns out to be as good as expectations.
I know faster chips,especially dual core, will need more bandwidth, but with a 2.4GHz dual core only seeing a 3-5% performance increase with DDR480 I just wonder what clock speed we will need to justify DDR2 800. This will be even more so if AMD is really going to introduce L3 cache on their faster processors which should reduce the need for more bandwidth. Seems a bit counter-productive unless they are so concerned about the higher latency that the L3 cache is necessary.
Servers as you noted are another story. AMD has quad core server chipps coming of course, but I wonder when their quad core for desktop will be coming to compete with Kentsfield. AMD always markets socket F as ready for quad core, but they've made no such claim for AM2 and I haven't heard much on that front.
nice find! but i still wonder (and im sure its not just me) how amd will respond to intel's conroe....
They may not need to. While I don't expect anything huge as far as a performance boost from M2 initialy beyond the new CPUs, though the OC range may be a bit higher, or gains from OC may be better with the extra DDR2 bandwidth. However, their natual progression of the chip may put them around Conroe when Conroe actualy comes out. There's 6 months to see what happens, and Conroe may not be all it appears to be. It may be better as well. But we'll find out when it's released.
Remember AMD is far more relient on low latency that low bandwidth (the opposite to netburst).
The higher the DDR2 clock rate the lower the latency, so yes Opteron will gain, but will not use the extra bandwidth (unless you go quad core and actually use the cores to their fullest)
I have been sayin gthat AMD will support 1066 for months. All of you who think that AMD is not tweaking the HELL out of AM2 to improve power and ILP is underestimating the creator of X86-64.
I know they will get 20-30% out of this move, not just through DDR2 but through process. They are being tight-lipped for a reason. They don't want to give Intel too many clues as to what they will do. K8 is not out of steam yet. I think that's one of the reasons why they aren't trying to hit 3GHz yet.
And don't be surprised if you see L3 happen also. SInce the FX and Opteron 8xx are such high margin chips they can afford to use more wafer per chip. They don't have to have EVERY chip faster than EVERY Intel chip. They only need to have the fastest chip.
DDR2-800 will give these CPU's a 10% improvement in performance. Let's not forget about the new SOI process which AMD and IBM stated that it will increases transistor performance up to 20% on 90nm and 40% on 65nm. With all these improvements (and who knows what other tweaks), AM2 will boost a total of 30% improvement which will rival Intel's Conroe "supposed" 20% advantage over current AMD processors. :wink:
People have a right to dream. Look at the THG bench,think that the mem. contr. might have been working corectly.
And silicon on insulator to increase count of transistors? Link? And if i remember correctly from a site that was supposed to happen in 2005.
939 is not bandwidth starved. I don't think faster memory is going to give a boost. DDR2 is just good for higher densities, because the latency is higher. Low latency is one of K8's biggest strength. K10 will be interesting.
If AMD can make a 20-30% and keep cost down that would be great. However, AMD is creeping up the price chart fast. The price cut in April may help get the RD guys more money but then who is going to buy the AM2 for lets say ($959-$1300) on broads that don't capture all the power it can put out. Once I see a board come out that can meet or exceed the spec AMD says I believe. Make a believer out of me AMD (Umm Numrukiao... Dr. Jones)
* The components are twice as fast as today's highest speed DDR2 memory products.
* The first computer systems equipped with the advanced DDR3 memory technology are expected to arrive in late 2006.
* The main advantages of DDR3 are the higher bandwidth and the increase in performance at low power.
* The DDR3 SDRAM devices will offer data transfer rates up to 1600 Mbps (megabits per second).
* The supply voltage for the memory technology is being reduced from 1.8 volts for DDR2 to just 1.5 volts for DDR3 targeting a work day equivalent of battery time.
* The voltage reduction limites the amount of power that is consumed and heat that is generated in connection with the increase in bandwidths.
Similar to DDR2, don’t expect DDR3 to be groundbreaking at the beginning. The memory modules will be compatible with high speeds on the Pentium architecture. Logically, they will also have relaxed timings, and operate at much higher speeds. Is DDR2 short lived? We are not sure. As much as the memory industry seems point forward, it in fact is not. Quite frankly, this has left us with many unanswered questions. We will be working with our sources throughout the industry to gather as much information as possible.