What Socs com­prise?

Electronics For You - - Chips -

A sys­tem-on-chip con­sists of many sub-com­po­nents such as one or more CPUS, graph­ics pro­cess­ing units (GPUS), dig­i­tal, ana­logue, mixed-sig­nal and ra­dio fre­quency func­tions, ran­dom-ac­cess mem­ory (RAM), read-only mem­ory (ROM), Flash mem­ory and EEPROM, os­cil­la­tors and phase-locked loops (PLLS), real-time timers, ana­logue-to-dig­i­tal con­vert­ers (ADCS), dig­i­tal-to-ana­logue con­vert­ers (DACS), power man­age­ment cir­cuits and ex­ter­nal in­ter­faces such as USB, Firewire and Eth­er­net. All this is placed on a sin­gle sub­strate.

Sim­ply put, the video en­cod­ing and de­cod­ing hard­ware pow­ers the ‘cam­corder’ func­tion­al­ity. The im­age pro­ces­sor en­sures that pho­tos are pro­cessed prop­erly and saved quickly, and the au­dio pro­ces­sor frees the CPUS from hav­ing to work on au­dio sig­nals thus al­low­ing these to work on other tasks. To­gether, all these com­po­nents and their as­so­ci­ated driv­ers de­fine the over­all per­for­mance of a sys­tem. Fig. 2). It fo­cuses on the CPU’S pro­cess­ing power and does not utilise GPU’S power for cal­cu­la­tions. The test re­sults are re­ported in num­bers. The scores are sus­cep­ti­ble to vari­a­tion among de­vices with even the same Socs.

Browser­mark ap­pears to have a bug where one stage of their bench­mark suite of tests could be in­ten­tion­ally skipped by us­ing cer­tain browsers. Ap­ple A5 CPU bench­marks are hence higher than they should be.

Gl­bench­mark 2.1 Off­screen. This is an Opengl ES bench­mark with graphic scenes rep­re­sent­ing high-end gam­ing con­tent (re­fer Fig. 3). The off­screen mea­sure­ments are where all high­level scenes are ren­dered off­screen at 1280×720 pixel res­o­lu­tion. This method is claimed to pro­vide ap­ples-to-ap­ples per­for­mance com­par­i­son for all GPUS as it utilises the GPU’S power as well for cal­cu­la­tions.

Un­like the in­flated Browser­mark score for the Ap­ple A5, Gl­bench­mark score is ac­cu­rate be­cause the A5 does fea­ture a high­end dual­core GPU.

Bench­mark trends

If you look at the top three scor­ing Socs (re­fer Fig. 2), you would find that clock speed does not de­fine per­for­mance. Snap­dragon MSM8260 de­spite claim­ing the high­est clock speeds of 1.5 GHZ still rates less than Ap­ple A5 SOC run­ning at 800 MHZ (though there is a slight ad­van­tage due to the Sa­far­iBrowser­mark bug).

The rea­son why MSM8260 lost even with the spruced-up clock speed is that this SOC is based on ARM’S Cor­tex-a8 de­sign. The other two Socs that beat it are based on the Cor­texA9, which is known to per­form 20 per cent bet­ter than A8. This re­sult also shows why you should not com­pare per­for­mance of pro­ces­sors hav­ing dif­fer­ent ar­chi­tec­ture based on clock speed alone. A bet­ter way to do so is to com­pare the per­for­mance per clock of the chips.

In the graph­ics bench­mark (re­fer Fig. 3), Ap­ple’s A5 wins hands down with its Pow­ervr SGX543MP2 graph­ics so­lu­tion. Ob­vi­ously it’s be­cause this is a dual­core GPU. More­over, SGX540 (be­longs to Se­ries 5) used in OMAP4460 was re­leased around two years be­fore the SGX543 (be­longs to Se­ries 5XT). As the chips used were

Fig. 1: Block di­a­gram of an SOC (blocks high­lighted in red are the crit­i­cal blocks that de­fine per­for­mance)

Newspapers in English

Newspapers from India

© PressReader. All rights reserved.