Dom

SUMMER IS almost upon us, and in the northern US, that means spring is upon us, along with the traditional onslaught of moles that the warmer weather brings. The moles seem to know a lot about AMD, and this spring is no exception. For such little creatures, they do tend to dispel proportionately more than their share of rumors.
The first thing out of their little lips was memory. While AMD is adamant that the Opterons only support DDR 266/333, DDR 400 will work just fine. The curious bit is why they are reticent about admitting to it. The reasoning behind this is that there are no registered ECC DDR 400 modules commonly available, and it would cause more problems than it is worth to list the memory before it is available. Since the Opteron is a server chip, the more common unregistered DDR 400 isn't suitable for the large memory requirements AMD is touting.

The more thorny issue is DDR2. While DDR2 was long rumored as supported by the memory controller, our moles confirm that the standard will not work with current Opterons. It will take a new chip, presumably with a new memory controller, to support the standard. This will most likely happen with the .09 chips due early next year. The current controller is said to have most of the DDR2 functionality on board right now, but not all of it. The update should be a relatively minor one, and it will also support DDR, but not on the same motherboard. This chip should be available just as DDR2 becomes available at a reasonable cost.

There are more secrets hiding in the current Opterons. One of the most deeply hidden is power management. Unlike the early, highly combustible Athlons, the Opterons do not burn up without a heatsink. The power management is leaps and bounds above the older chips. While the "official" power management is good, full Pentium 4 style thermal throttling is on the chip, but not activated yet. What is less clear is whether this is a management decision, the Opteron is seen as a server chip, so it shouldn't be needed, or because it was delayed because the functionality was not complete. Look for it in the Athlon 64 this September.

The little critters were also hot and bothered about was yields. The Opteron launched at an unexpectedly low frequency. The original plan was to seed "pre-release" chips on the market early this year at 1.4 and 1.6 GHz, and release the chips at 1.6, 1.8 and 2.0 GHz as the x42, x44 and x46 respectively. This plan was scuttled as unexpectedly low yields on the higher speed parts forced a 1.4 GHz to market, and one look at the price structure tells us that there are a lot of them to get rid of. The chip is capable of quite high speeds, and those chips are up and running now, but don't look for them real soon. What is less clear is where the problem lies. AMD has historically had excellent production facilities, and the Dresden fab has been an excellent example of this in years past. The scaling problems with the Palomino and Thoroughbred chips were due to design, not process/production issues, and the Opterons look to be hindered similarly.

The Athlon 64 is said to have the same problems as the Opterons, and is currently set to debut at 2.0 GHz with a 3400+ PR rating. AMD is most likely wishing this was due to market forces, not production problems. Whatever the debut speed is, it will most likely be competitive with the highest speed Pentium 4 available at that time. If AMD cannot scale the chip like they have listed on the roadmaps, they will quickly be left in the dust by Intel's Prescott chip.

The one saving grace AMD has is power. The power consumption for the K8 models is extremely low. Exactly how low is open to debate however, because the publicly available data sheets for the K8 chips are curiously incomplete. There are, however, quite complete data sheets if you are willing to sign an NDA. Unfortunately, we are not. Whatever the exact numbers are, as Aces Hardware shows, the actual consumption is well below that of a comparable Xeon. Why AMD wants to hide this is beyond us, we would be shouting an advantage like this from every rooftop we could climb to. Even empirical testing shows that a 244 running at full steam, with only an aluminum heatsink (OK, it had a copper slug), is cool to the touch.

One heat related change is the recent switch from a thermal pad to thermal grease, a material not recommended for the K7 generation of processors. The change was due to the heat spreader. This rather hefty chunk of copper presents a much larger surface area for the thermal pad to adhere to. When the thermal pad cooled, it formed such a strong bond with the heatsink that people were pulling the CPUs out of the socket when trying to remove the heatsink. Thermal grease solved this little snafu with minimal fuss, but much more mess.

AMD took another seemingly backwards step with the Opteron launch which was wrongly attributed to heat, they launched the chip in a ceramic package. With rumors of nuclear reactor level hotspots, and problems galore swirling, the moles again came through with the real reason for the change, money. Ceramic is a better packaging material in all respects, but is also a much more expensive material, the current ones we are told cost in the $15 range. Since the Opterons are meant as a high ASP product, it was deemed to be the safe choice to go with a ceramic package on launch, and move to an organic package later. The low yields of the higher speed Opterons, which depress ASPs may force a change to this sooner than anticipated. All of the Athlon 64s will be in organic packages, even the early mechanical samples carried on the moles little backs were organic.

While spring may have sprung, the anticipated wave of Opteron motherboards has not. Every source contacted has said this will change soon, very soon. With the launch of AGP motherboards in about a month, along with the 1xx and 8xx chips, expect a plethora of boards. The drought will soon be over, and there will be several interesting features. While the initial crop of boards are all based on AMD's reference design, the ones due soon are not. Expect all the goodies that are currently lacking to come out in full force. The most eagerly awaited chipset is the NVidia Nforce3 based boards, which will bring the Opteron up to par with some of the best Intel Xeon based boards and surpass them in many respects. All of the ancillary features that the Athlon MP based boards lack, like TCP/IP Offload Engines (TOE), and other workstation and server features should be here by the end of June.

One feature that we have not seen is more flexible use of the Opterons memory controller. All motherboards due out use the controller as 1x128 bit, not as 2x64 bit, which the controller is quite capable of. It is possible that it was deemed to be unnecessary in the current server oriented boards, and may surface in a more memory flexible workstation setup later on. Lets hope some motherboard makers try to differentiate their products in interesting ways, we all can only benefit.

The current motherboards are all scarce, feature free and expensive. This is mainly because they are new. By the time the Athlon 64 comes out, expect the price of motherboards to plummet. One of the main benefits of the K8 series of chips is the integration of most the north bridge in the chip itself. This, along with the lessened pin count allowed by HyperTransport, will allow for 4 layer motherboards to make a return. All of the currently planned Athlon 64 boards are 4 layer, which should allow for some very inexpensive boards. 6 layer, multi-CPU Opteron boards are also said to be arriving in the not to distant future.

Early in 2004, there will be integrated graphics chipsets for the Athlon 64s. Because of the internal crossbar architecture of the Opterons, UMA chipsets should not have the same adverse impact on performance that the current crop of Athlon and Pentium 4 UMA chipsets have. The performance, while far from spectacular, should not make an enthusiast cringe.

The moles then turned to the future. Much has recently been made of AMD's partnership with IBM. Rumors have been flying thick and fast about everything from IBM buying AMD to AMD paying IBM to design their chips. Most of these rumors are based on half-truths or less, but there are some major changes at AMD R&D.

The foremost change is that AMD no longer really has a chip R&D facility. They have closed their R&D facility and split up the engineers. The flash and memory people have been split off and moved to parts unknown. The CPU team has been moved to IBM's facility in New York, and will work jointly with IBM. How much of IBM's future direction is dictated by AMD is still unknown. This is definitely a net gain for AMD, and probably a gain for IBM as well. All future chip fabrication technologies are basically the same for both companies. At the Opteron launch Hector Ruiz hinted at this, but he was vastly understating the depth of the ties.

As far as the moles know, this was not a technology purchase as much as a pooling of the minds for mutual benefit. There was an 11-digit fund transfer from AMD to IBM several months ago. The rumor mill has again been swirling with FUD, but the wrong FUD. AMD hired IBM to help them design Opterons, but the work was not on the current .13SOI chip, that was done by the time IBM got involved. The 50 or so million dollars was spent on the transition to the .09SOI process. With IBM's help, this should be a smoother than usual launch. All signs say it was money well spent.

The last open question is who will sell all these 64 bit chips? Carly Fiorina was spotted going to the private AMD suite after the Opteron launch, but there is no word from HP yet. People laughed out loud when we said Sun would use AMD chips last year. While Sun did not commit to the Opteron when it announced Athlon products a few months ago, the same sources who told us about Sun initially said that Sun is definitely going to make an Opteron box, they simply were not sure when. The same doubters fell over with abdominal cramping they were laughing so loud when we stated AMD and Dell were talking. Milk spewed from noses worldwide at the thought as people read the article over breakfast. While nothing has been announced yet, the moles outside of Dell HQ say AMD sales people are going in and out of Dell in higher numbers than last year. They also point to the lack of public rumors as a sign against this being a PR stunt. Time will tell, and that time should be around September.

---------------------------
Source: TheInquirer