Intel Debuts New Smartphone Chips at MWC
Intel's dual-core SoC
The Z2580 will be a dual-core chip at 1.8GHz. Both cores will have Hyper-Threading (for a total of four threads), and the chip will use a dual-GPU configuration of the SGX544. That's the same GPU that currently powers the iPad 2, but Intel's version will be clocked at 533MHz (current implementations of the SGX543/544 tend to run at 200-300MHz). The next-generation SoC's GPU probably won't be the fastest chip on the market by the time it ships, but it'll be significantly faster than what's in the iPhone 4S right now and therefore likely still competitive.
Intel's CES reference design
The chip's CPU, on the other hand, should be extremely competitive. Hyper-Threading, combined with the low degree of quad-core optimization in current shipping smartphone products, give Intel an edge here. In situations where quad-threaded applications are available, HT will allow the chip to achieve higher performance. When it isn't -- which will likely be the majority of the time -- the Atom core won't be penalized.
This leaves the question of how Intel will squeeze a dual-core 1.8GHz chip into a smartphone where a 1.6GHz single-core is currently, without moving to a new process? Intel isn't saying yet, but we bet we know why. To answer this question, we need to broaden our perspective from considering just the CPU, and look at the entire phone.
The flat blue line shows current power consumption; the dotted blue line estimates a modest decrease in line with an improved 32nm process
A modern smartphone contains a screen, Flash memory, DRAM, a GPU, and a cellular/WiFi radio -- to name just a few components. The question isn't "How much power will the CPU draw," but "What's the power consumption of the device?" A phone with a weak GPU or an ultra power-efficient radio can use that headroom to increase the performance of another component.
By the time the Z2580 launches, Intel will have further improved its 32nm process, likely reducing power consumption below the Z2460's level. Chances are good that the next-generation XMM 7160 radio included with the new dual-core chip will be built on 28nm; we've included a non-Intel graph below showing Xilinx's estimated power savings when moving from 40nm to 28nm. The figures are strictly representative, but they show the sort of gains multiple companies have claimed from the process shift.
Intel isn't sharing any details on what innovations its baking into the next-generation dual-core SoC, but we expect the chip will modestly expand Medfield's competitive range vis-à-vis the solutions it ships against at launch. When Intel historically launched itself into PCs, data centers, and the HPC market, it succeeded in all three by offering price/performance competitive solutions. This worked particularly well in data centers and big iron implementations in the mid-90s, where the RISC vendors of old largely dismissed x86 as a consumer platform until it was too late to reverse the trend.
Phones aren't guaranteed to follow the same path, but Intel's manufacturing prowess and steady progression will make life difficult for its ARM-based competition. When we talked to the company, we were told Santa Clara is pleased with the response its gotten from Medfield thus far. According to one unnamed source, the question has gone from "Will you ever get in phones?" to "How long until you have phones that can do X, Y, or Z?"
2012 is all about building momentum around Medfield, and creating phones that can challenge incumbent ARM-based designs in a variety of markets.
Intel's CES reference design
The chip's CPU, on the other hand, should be extremely competitive. Hyper-Threading, combined with the low degree of quad-core optimization in current shipping smartphone products, give Intel an edge here. In situations where quad-threaded applications are available, HT will allow the chip to achieve higher performance. When it isn't -- which will likely be the majority of the time -- the Atom core won't be penalized.
This leaves the question of how Intel will squeeze a dual-core 1.8GHz chip into a smartphone where a 1.6GHz single-core is currently, without moving to a new process? Intel isn't saying yet, but we bet we know why. To answer this question, we need to broaden our perspective from considering just the CPU, and look at the entire phone.
The flat blue line shows current power consumption; the dotted blue line estimates a modest decrease in line with an improved 32nm process
A modern smartphone contains a screen, Flash memory, DRAM, a GPU, and a cellular/WiFi radio -- to name just a few components. The question isn't "How much power will the CPU draw," but "What's the power consumption of the device?" A phone with a weak GPU or an ultra power-efficient radio can use that headroom to increase the performance of another component.
By the time the Z2580 launches, Intel will have further improved its 32nm process, likely reducing power consumption below the Z2460's level. Chances are good that the next-generation XMM 7160 radio included with the new dual-core chip will be built on 28nm; we've included a non-Intel graph below showing Xilinx's estimated power savings when moving from 40nm to 28nm. The figures are strictly representative, but they show the sort of gains multiple companies have claimed from the process shift.
Intel isn't sharing any details on what innovations its baking into the next-generation dual-core SoC, but we expect the chip will modestly expand Medfield's competitive range vis-à-vis the solutions it ships against at launch. When Intel historically launched itself into PCs, data centers, and the HPC market, it succeeded in all three by offering price/performance competitive solutions. This worked particularly well in data centers and big iron implementations in the mid-90s, where the RISC vendors of old largely dismissed x86 as a consumer platform until it was too late to reverse the trend.
Phones aren't guaranteed to follow the same path, but Intel's manufacturing prowess and steady progression will make life difficult for its ARM-based competition. When we talked to the company, we were told Santa Clara is pleased with the response its gotten from Medfield thus far. According to one unnamed source, the question has gone from "Will you ever get in phones?" to "How long until you have phones that can do X, Y, or Z?"
2012 is all about building momentum around Medfield, and creating phones that can challenge incumbent ARM-based designs in a variety of markets.
