Advanced Micro Devices Inc (AMD)

[rupert]

Translation from Japanese of the Opteron dual-core story originally posted by petrsny. The url also provides diagrams.






*Hiroshige's Goto Weekly overseas news*
AMD throwing dual core CPU to 2005




--------------------------------------------------------------------------------

- As for AMD planning the dual core from the design beginning of K8

AMD, perhaps, has been about probably to throw dual core CPU to 2005. With dual core conversion, as for AMD there is a possibility of preceding to Intel. From various circumstantial evidence, the dual core plan of AMD started becoming clear.

AMD K8 (Athlon 64/Opteron system CPU) the occasion where architecture summary is announced, at the basic design stage of same CPU, inserting dual core conversion in range of vision in the first place 2001 October, has made that it developed clear. In other words, the present Athlon 64/Opteron family is the single core, but it is possible to expand relatively easily in the dual core.

In addition, entering into this year, it has started hinting at AMD, concerning the dual core. Japan AMD Microsoft held with "AMD & Microsoft Software Developer Conference", constitution example of dual core edition Opteron including also cash size, introduces in April. In addition, "functionally as for the dual core it is possible even at the time of Computex. If there are needs of the market, that it can offer simply ", it has explained.

The observation that increases even in computer industry same as the dual core forward fall of Intel, also AMD throws the dual core. As for a certain industry authorized personnel, as for the largest reason where Intel hurries the dual core, you say that is, because AMD throws dual core Opteron 2005 2nd quarter. In addition, as for another CPU manufacturer authorized personnel "as for dual core natural flow of industry. If 90nm from viewpoint of die/di size it can designate AMD as the dual core. That, it points out it probably will come directly introducing ".

The dual core CPU which AMD throws, is presumed that first it becomes the Opteron brand. For example, Dirk Meyer of AMD (the darkness mayor) the person (Exective Vice President and Computation Products Group), you view as follows concerning multiple core CPU.

Is "as a outline, but with short-term view, with server environment as for being the flow where the multiple core is natural it is clear. Because multiple it has threaded present server application, already naturally. When you look at future, also multiple threading environment with the client probably will become more normally. But, when (you compare with the server) it is the pace which will be been slow ",

Meyer talks as outline to the last, but the attitude of AMD is clear. With the server the multiple core being effective even at present time as AMD, the fact that with the client the multiple core becomes useful, is the case that you think that it becomes for a while first. Because of that, AMD is expected that the dual core is thrown from Opteron system.


- Constitution of dual core K8 architecture


Block diagram of dual core K8 at time of 2001 announcement
As for PDF editionthis
So, the dual core CPU of AMD becoming some constitution? First, basic architecture itself has been explained already 3 years ago. The figure K8 due to AMD (Opteron/Athlon 64) designated the explanation when announcing the architecture as the base, it is the diagram in case of dual core constitution. When you look at this, K8 from first the dual core being developed as prerequisite it is understood.

It is included in the north bridge function of K8, it has become the mechanism which is arrangemented "System Request Queue (SRQ)", with the port for two CPU cores, "Advanced Priority Interrupt Controller (APIC)" of built-in by. According to AMD, as for System Request Queue itself whose dual port conversion is possible, you say that it has been mounted already on present K8 system CPU. Because of that, with while it is the architecture of current Opteron/Athlon 64, dual core conversion can do AMD relatively easily.



The constitution example of Dual-Core Opteron which is released in 2004
Furthermore, with April this year AMD & Microsoft Software Developer Conference, also more concrete data flow figure is open. In this figure, the example where each CPU core of dual core Opteron loads L2 cash of each one 1MB is shown. With K8 architecture, because from first both cores having private cash in prerequisite, System Request Queue is installed as for this constitution itself it is not unexpected. Those where it is important are that also the concrete cash quantity, 1MB is shown. As for this, the dual core Opteron plan of AMD, the possibility of showing the fact that it is advanced concretely is high.

When we assumed if AMD, CPU of according to constitution of the dual core Opteron which is shown in April develops, as for AMD you called HyperTransport interface of 3 ports and memory interface of dual constitution, the interface constitution of present Opteron is presumed the thing which is used basically that way. When we assume, that is, as for this dual core CPU, the possibility where Socket 940 you can put similarly with present single core Opteron and the bus & pin compatibility, is high. If HyperTransport and DRAM interface the same as present condition, as for AMD it is not necessary to change the CPU socket.

AMD in the near future memory in DDR2, has the necessity to pull up interface to HyperTransport 2.0. But, 2005 first half is thought that on the delicate line, the possibility of coming placing the dual core in former memory & below of interface is higher. Because in other words, first, in the server side as for the necessity of DDR2 it is not high that much. For example, if Opteron of 2 processor constitution because it has the DRAM interface of channel 4, memory zone is guaranteed sufficiently even with DDR. In addition, in the aspect of memory capacity, low speed memory is more profitable. Furthermore, with DDR2 of the server side, "Fully Buffered DIMM (FB-DIMM)" with there is also an unsolved question which is said. With Opteron topology, as for FB-DIMM always it is not useful, but it probably is necessary still to look at the trend of DRAM industry as AMD.


- As for AMD with 90nm process dual core throwing?

When we assume that AMD throws dual core CPU to 2005, as for that it is the expectation which is produced with 90nm process. Because the 65nm process of AMD is not in time in 2005. The dual core CPU of Intel basically is 65nm process, but AMD is seen that it moves to the dual core with 90nm. Because AMD the CPU core is small, this becomes possible. Either Meyer of AMD, does not deny the dual core with 90nm as a technical possibility.

Whether or not "(the dual core possibility) logic (process) is not, the core it depends some size. That also the business decision which you said whether the (process technology) geometry (with producing) it does good financially includes ", (Meyer)

In other words, if it is settled in the tip/chip size to which the core sufficiently is small, the to dual core converting, does good financially, it is the case that with respect to technology & economy, can be dual core conversion with any process technology. So, die/di size of the dual core Opteron of 90nm becoming which?

First, the die/di size of ClawHammer of present 1MB L2 cash (the area of the semiconductor itself) they are 193 squares mm. Only 512KB L2 cash does not load "Newcastle (the new castle)" Athlon 64 FX of the core is presumed that the die/di going around is small, but as for Opteron and Athlon 64 system of 1MB edition, neither changes as for die/di size. Comparing the CPU die/di photograph of the Web sight of AMD, the die/di itself is the same.

193 squares mm relatively are the big die/di for PC as CPU. But, the market share of AMD at the presently it has fallen, enlarging the die/di of the server type CPU where, the quantity of shipment is small relatively probably will not trouble AMD, from at the time of peak. Fab of AMD to have fixed production capacity, AMD must bury that capacity, because. So unless it does, the enormous investment to Fab becomes wasteful. In other words, sarcasm especially, AMD because market share does not extend, maintains the die/di of CPU in principle and largely it is possible to strengthen performance.

- The to dual core converting, die/di size almost the same

ClawHammer system is the die/di of 193 squares mm, but to tell the truth the CPU core itself among those is not that much large.

AMD, has revealed the size of each block of the ClawHammer core, in the presentation with October last year Microprocessor Forum and VTF of last week. According to that, the area which the CPU core itself occupies is no more than 28%. As for remainder, L2 cash SRAM of 1MB 42%, north bridge function and I/O approximately occupies 30%.

CPU core 28% 54 squares mm
L2 cash 42% 81 squares mm
Bridge north I/O& 30% 58 squares mm





1MB edition K8 (ClawHammer) die/di
As for enlarged viewthis Similarly the diving lock

When we assume that the CPU core of dual core Opteron, sort of the presentation with of AMD & Microsoft Software Developer Conference has L2 cash of each at a time 1MB, the CPU core and the L2 cash SRAM part reach 2 times. But, if 30% I/O and the north bridge functional part do not expand interface and the like, it is not necessary to increase. In other words, the die/di approximately 70% just increases.

On the one hand, in order with movement to 90nm process, approximately 70% to reduce from present 130nm process 50%, in regard to calculation the area of block of the same number of transistors approximately becomes. Actually, first because that compared to reduction percentage becomes low, approximately 60% you estimate. So when it does like under, in regard to calculation each block of the K8 architecture with 90nm becomes.

CPU core 28% 32 squares mm
L2 cash 42% 49 squares mm
Bridge north I/O& 30% 35 squares mm


When with this presumption, the CPU core and L2 cash respectively become dual, size of each area becomes like under.

CPU core 33% 65 squares mm
L2 cash 49% 97 squares mm
Bridge north I/O& 18% 35 squares mm


When these blocks are totaled, the die/di becomes approximately 197 squares mm.


DualCoreK8 calculation figure
As for PDF editionthis
In other words, in regard to calculation, AMD when it moves to 90nm, same as current Opteron, perhaps, at die/di size below that, means to be able to actualize the dual core processor. As for this, the fact that it can make the dual core without causing the increase of cost of the die/di is meant.


- The electric power consumption and heat of AMD CPU of dual core

Another problem with dual core conversion is electric power consumption and heat. The transistor of amount and CPU which increase the CPU core increases. Because of that, active electric power consumption and leak (a leak) electric power increases. With refining the process, to somewhere TDP (Thermal Design Power: Whether thermal design electric power consumption) you can hold down rise, it becomes an important point. Meyer of AMD answers to the question as follows.

"It is outline, but the customer has understood, if how it does, whether it is possible to bring Advanced Capability to the market easily. Many customers securely, desire the fact that Advanced Capability is brought with the inside of the frame of existing electric power consumption ", that Meyer talks.

TDP of Opteron of 940 pins presently 89W. AMD can push in dual core edition Opteron into that framework, if, introduction of the dual core as Meyer says, probably becomes easy. But the て TDP of CPU of rival Intel already exceeds 100W so not to be. The still has shown the fact that it is room in AMD. Because Intel pulls up the upper limit of TDP, in the same technical based basket body and around power source, more powerful CPU of TDP whose also AMD is higher can be introduced.

There is no sign where at present, AMD has the dual core in the desktop. The fact that it is audible from industrial source to the last is just the dual core plan with Opteron. But, if, die/di size of the dual core CPU of AMD TDP and it is settled same differing from present 1MB edition K8, as for AMD a possibility of throwing to the desktop market for example that dual core CPU, as a Athlon 64 FX succession there is.

However, this is influenced by the capacity and yield rate and market share of the 90nm process of AMD. The yield rate of 90nm high capacity abundance being, in spite, when market share does not extend, AMD can make the die/di of CPU large. When it is the opposite, as for AMD because for PC it is necessary to maintain the die/di of CPU small, the desktop of the dual core probably becomes slow.

Though, in addition as for the strategy with respect to marketing it is another. When Intel starts praising the dual core extensively in 2005, there is a possibility also AMD opposing to that. Also it is little, as for AMD as for CPU itself which opposes it means to have.






http://babelfish.altavista.com/babelfish/trurl_pagecontent?url=http%3A%2F%2Fpc.watch.impress.co.jp%2....