MU_Engineer :
There is no doubt that at least some of the X3s will be no more than a "Agena" Phenom X4 die with one core turned off. This has happened before in both makers' models, such as the "Manchester"-core 939 Athlon 64 3200+ and 3500+ units as well as the Core Solo and Core 2 Solo CPUs. The question is whether AMD specifically intended to make 3-core chips as well. AMD had been touting the "modularization" of the 10h and upcoming "Fusion" when the first information on them came out:
http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=3050&p=2 It appears to me that AMD may very well have aimed at something much like the engine manufacturers did in making a modular design, where the number of cores and cache can be changed to suit a particular need.
I suppose the proof of that would be if AMD supplants the Agena-with-a-disabled-core X3s with a native 3-core CPU or if all 3-core chips are 4-core chips with a disabled die. The fact that the Toliman X3 got its own code name suggests it gets its own die mask and there will be native 3-core chips, if AMD follows its previous code-naming convention.
I disagree. Considering AMDs deceptive PR releases this past 14~15 months, I would doubt that "touting" anything implies that they are deliberately planning, designing or manufacturing that thing with specific intent. Maybe they are, maybe they arent. I certainly wouldnt take them at their word, and in the case of the tri-core, unless they actually give someone a demonstration of the manufacturing process using actual 3 core masks, then I see no reason to trust that the tri cores are anything more than defective quad cores. There
was some mention of a 3 core die late last year, or early this year, from AMD, but I suspect that was not the result of any belief that there might be a profitable market for a 3 core chip, but the result of prior planning to account for possible methods to offset the loss of manufacturing cost associated with defective quad core dies. In that specific case, and that case only I would agree that AMD 'planned' tri-cores. The one other case in which I might agree, given reliable proof, would be if AMD had considered the potential increase in wafer yields based on the reduced area required for a 3 core die vs a 4 core die and then only if they yielded better margins (more profitable) than quad cores. But this seems highly unlikely. Furthermore, considering AMD seems to have felt a need to have a product which could compete against Intels quad core MCM, I see no logical reason why they would willingly produce a product 1 core shy of a full deck. And remember, if they deliberatly mask the dies for 3 cores, with their purported yield issues, they will wind up with many dual cores from a wafer, but no possibility whatsoever of producing of a quad from that wafer. Even with quad core, the possibility exists of salvaging a dual core processor if 2 of the 4 cores are bad. I also dont agree that assigning a designation to a product implies that the product will be purposely manufactured.
From what I read in the anand article it does not appear to address the concept of modularization from the physical aspect as does the automotive industry, but rather simple seperation of process. IRT the fusion, where the automotive industry looks to modularize by reducing asemblies to their smallest possible repeatable/interchangeable components, to reduce manufacturing processes, AMD is just breaking the inclusive functions into independant modules. This is not to say that what either AMD of the automotive industry is wrong, just different.
MU_Engineer :
Actually, it is a good analogy. A specific engine block is cast in a specific mold. The general *design* can be adapted for a different number of cylinders, but only one type of block comes out of a specific mold, such as a specific die comes from a specific mask. You could make a working I-5 out of an I-6 block by removing the piston from the 6th cylinder, putting on a head with only five intake/exhaust ports and plugs, as well as remapping the firing order to not include the sixth cylinder. This is roughly what's done with making the four-core Agena CPU into the three-core Toliman CPU. You could also make a five-cylinder engine using its own specific mold, which would be analogous to a native Toliman X3 mask made upon a modular design. Of course, no engine out there has permanently deactivated cylinders such as the hypothetical I-5 I described due to balance issues as well as the fact that it's the same size and weight as the more-powerful I-6 version and thus less efficient. Plus, it's easier to recycle a dud block than to make a deactivated version, but you get what I am aiming at here.
It is true that a specific engine block is cast in a specific mold....due to the nature of casting, it can be no other way. However, in the case of the modularized engine blocks and cylinder heads, it is not the mold that that is the primary concern. Remember, manufacturers still utilize single cast molds:
-Only 1 block per mold
-Molds are made from blanks
-The mold is destroyed after the casting is complete (as part of the manufacturing process)
-In some processes (depending on the compleity of the part) the blanks are destroyed (EX Lost Wax Process)
-There is no cheap alternative because of the voids within the block and cylinder heads for coolant.
-A reusable mold is possible, but would result in a significantly weakened casting for a number of reasons
-Permanent mold (die) casting still produces too many flaws to be relied upon for consistancy in casting parts with asymetric voids, especially those that are subject to persistant cyclic loading.
As with all things, there are exceptions to rules, and the exception in this case are air cooled engines. Because aircooled engines rely on direct heat transfer to the atmosphere rather than transfer through an intemediate medium, they have cooling fins and contain no internal voids to circulate coolant. As such, these types of motors can and regularly are successfully cast using the die casting method.
But in the case of the lost mold, the function and quality of the mold itself are directly controled by the mold blanks. By using this charecteristic with the modularized process, a single cylinder blank or single bank of cyliders can be designed with uniaxial symetry thus allowing an infinate number to be identically replicated, then joined until the desired number cylinders for the casting is achieved. In the modularized process, the system revolves around the blanks, not the molds, since it is the blanks which are used to create the molds, and molds are lost for each casting. This is unlike previous generations of engine design in which a single blank was designed, replicated and in turn used to replicate mulitple molds, or single molds (lost wax). The current process (which has still yet to be universely adopted) allows the manufacturere to use the same equipment repetitively without having to rely on multiple lines for different castings
or reseting/retooling the entire process for a single product line everytime a different casting is called for. In the new process, the changes are addressed at the 'lowest' potential variable allowing greater flexability while achieveing greater consistancy. As such, any number of possible block configurations can be achieved up to the limit of the foundry equipment, without any significant changes to the process or setup, because the modularization of the smallest segment of the block allows it to be 'scaled' through the adaptation of the blanks.
Staying with modularity in automotive manufacturing, not so long ago, when electrically operated windows were a luxury option purchased only by the rich, it was the norm to find vehicles which used seperate non interchangable regulators, tracks, and integrated reduction gear boxes/electric motors for each windows. This was a blatent waste of manufacturing resources. In most of todays vehicles, by modularizing assemblies at the smallest level possible, not only are the motors seperate from the reduction gear housings, and completely interchangeable, but the reduction gear cases and regulators are completely interchangeable. Only the tracks which must conform to both the shape of the door and the window (thus eliminating the possibility of symetry in multiple planes) cannot be interchanged. But the automotive manufacturers continue to press further. In my personal vehicle, not only are the window motors interchangeable betweent the doors, but also the tailgate and even the windshield wipers. In fact, both the rage selection motor and the drive selection motor on the transfer case are the same motors used in the doors.
In order for AMD to achieve that form of modularization, they would have to go to a MCM. From what I gather in Anands article, AMD is didlling with altered die masks for a single die, and while thay can take the single die approach, and do so succesfully, they would have to have a seprate mask for every different model they choose to produce. The MCM approach on the other hand, would only require a single mask for each specific module and be much closer to how the autmotive industry has implemented modulariation. Again, niether approach is particularly right or wrong, but each does have its advantages and disadvantages. In the case of the single die approach to fusion, if one module is corrupt, unlike the quad core, the entire die is lost. This is a case where AMDs implementation of modularization provides no option to salvage the die.
So, because todays engines are designed to scale in cylinder number (through the manufacturing process, not the operation such as cadilacs 4-6-8 engines) and quad core is not designed to scale (through the manufacturing process) I assert that the analogy is inaccurate, and that TCs is accurate.
Facebook
Twitter
Instagram