AMD: Homogeneous Multicore Strategy May Be Wrong, Like the Ghz Race, So We Choose Another Way
AMD CTO Phil Hester
(Phil Hester) voiced some of the company’s doubts about how to proceed
increasing the performance of microprocessors. Instead of integrating
in one processor many identical cores a decision is considered to develop
APU (Accelerated Processing Units).
Benefit from using multiple cores in one processor,
definitely is, only for this you need the appropriate software, with support
multithreading and tasks amenable to parallelization will be fine on such
processors run. In addition, the power consumption of the new 2- and 4-core
processors turned out to be at or even lower than what was typical for
known solutions in the last few years. However, at the annual meeting with
financial analysis in the company expressed the opinion that homogeneous multicore
– an approach that may not be optimal and therefore short-lived.
Hester noted that after 2000, the industry began
the race for processor frequency, which could not be eternal and therefore was
misguided tactics. The same is happening now with the multi-core concept.
Homogeneous multicore may become increasingly inadequate over time
time and therefore AMD is betting on a new type of
processors – APU.
APU is a processor based on the idea of building a block structure (heterogeneous system) of integrated elements that has arisen since the first integrated memory controller in the core. T.e. in the future AMD intends to use ATI developments for integration into the core of the CPU (central processing unit). These processors are expected to have better performance, remain multi-core, use built-in GPUs to speed up graphics, and be more efficient per watt of power.
All this is very closely related to what was already voiced when describing the concept of Fusion by AMD. According to reports, the company plans to introduce mobile CPUs “Fusion”, combining CPU and GPU in one core in 2021. According to Mr. Hester, these processors will penetrate not only into mobile devices (PDA and others), but will also be widely used in very powerful solutions. T.e. the key point is that chips with integrated graphics will be massive – both in budget entry-level solutions and in productive workstations and “petaflop” computers, thanks to the wide capabilities of the integrated graphics chip (kernel). Recall that the GPU can be used both for its intended purpose – to process graphic data and output to the display, or use it as a stream processor – a powerful “number crusher”, i.e.to. the capabilities of its FPUs are vastly superior to those of almost any CPU.
It turns out that starting in 2021, thanks to AMD, we are moving from universal, but inefficient general-purpose processors to new chips with highly specialized hardware – with faster computations thanks to their block scheme. Fusion should be the first such solution in which AMD will implement its Torrenza concept, with which the company will launch the “era of accelerated computing”.
In addition, AMD has provided some information on some future products. In particular, in 2021 we are expecting the release of the 4-core processor “Shanghai”, which is a further development of the Barcelona core, which is scheduled for release in mid-2021. As a reminder, this will be the first processor to use a faster newer bus – HyperTransport 3 (HT3), although it will continue to use DDR2 memory. Thermal package of these solutions – 68-120 W.
As for desktop processors, in 2021 we will see the transition to socket AM3, the announcement of 1-, 2- and 4-core solutions,
working with DDR3 memory. Next year – the announcement of the HT3 bus, further development of the two-socket QuadFX platform, as well as announcements of new chipsets and graphics solutions, for example, the R600 DirectX10 chip, which is positioned as a direct competitor to the GeForce 8800.
The mobile world should also introduce new
interesting solutions. AMD is preparing new optimized in terms of
power consumption graphics chips for notebooks. Dynamic
graphics mode “means the laptop uses discrete graphics, being
connected to the network and automatic transition to an integrated graphics solution
when powered by batteries.
Based on materials