Stop the Intel-Atom-Bashing! I just read another magazine article about how poor the performance of…

Stop the Intel-Atom-Bashing!

I just read another magazine article about how poor the performance of...
Stop the Intel-Atom-Bashing!

I just read another magazine article about how poor the performance of the current #Intel #Atom is supposed to be, and frankly I am sick of it. I will limit my remarks on the netbook variants.

First of all I would like to remember every critic that the Atom #CPU is a full #x86 processor featuring #MMX , #SSE , #SSE2 , #SSE3 and #SSSE3 . This not only makes it a general purpose work horse for mobile computing but also provides the nowadays common #SIMD capabilities. Furthermore all ( #Netbook ) Atoms feature #Hyperthreading and part of the Atom line-up is Dual-Core. On the #GPU side the commonly belittled #GMA950 , #GMA3150 , #GMA3600 are included which are by far not comparable with the #nVidia or #AMD counterparts, yet still provide enough performance for everyday life and even low-end gaming, but most important of all Intel is the only GPU manufacturer who provides high quality open source drivers.

From the charts of one can see that the Netbook Atoms are somewhere around the mobile AMD #Sempron 3300+ on the CPU side and near the #Geforce Go 440 on the GPU side which is pretty impressive if you consider the comparatively lowest power consumption. Comparing Atoms against #ARM processors is not completely fair as these are not of x86 architecture. The Atoms main competitor, the AMD #Fusion , indeed has a faster CPU and GPU, but here it should be noted that this triplication of performance also costs at least double as much and thus targets another price segment.

One reason of the lower performance of the Atom processor is its characteristic to execute In-Order as opposed to the today common Out-of-Order execution, which not only requires a lot of silicone area but also demands its share of power. Of course, the absence of instruction-reordering and register-renaming can cause stalls, but this can be massively reduced by testing on In-Order platforms and clever programming, in example reducing unnecessary conditional computation and branching in time-critical code. Conditional computations such as selection, minimum/maximum, conditional negation/incrementation can be serialized for instance by #bittwiddeling . Branching on the other hand should be minimized on any platform as usually for each (conditional) jump the CPU caches have to be invalidated and refilled. With this in mind a lot can be accomplished on an Atom, with the positive side effect of accumulating a performance reserve on Out-of-Order CPUs. Furthermore current compilers offer optimization options specifically for Atoms such as #gcc , #clang ( march=atom ) and #icc ( -xSSE3_ATOM ). If you are #Windows user and your operating systems sluggishness is the root of your complaints I would counter that you could just recompile your Windows with something like “make localmodconfig”, oh wait you can't, as it is not #opensource ; in that case you would have to live with what you have been given or rather what you bought.

With the current iteration of the Atom series ( #Cedarview ) my only criticism, the maximum adressable amount of memory, has been raised from 2GB to 4GB. This has been in my opinion a market decision to prevent the Atom from poaching in the share of Intels low end notebook CPUs.

I myself have (positive) experiences with the #N270 , 330 and (my sole current computer) #N570 on which I run a really snappy ( #archbang ) #linux and am fully enabled to engage in my game development projects. Somewhat the downwriting of the first Celeron generation ( #Mendocino , not #Covington ) is brought back to mind which turned out to be real beasts with their then fancy MMX and on-die L2-cache. Finally I want to thank Intel for introducing this low-priced powersaving #GPCPU and close with a (+1) for the Atom.

(View on Google+)

Posted in gplus.

Post a Comment

You must be logged in to post a comment.