I may be shortly in the market for a new notebook/convertible, and I was browsing my options of CPUs. With the knowledge that Silvermont had been released a few months ago, I decided to look at some Z3740 benchmarks in comparison to my current APU (an AMD A6-4455M in a Samsung Series 5 535U3C) and this is what I found:
What is very weird here is that the Z3740 performs better than my A6-4455M in many tests; in some tests it is even twice as fast! However I do not understand why this is; the Silvermont core is simpler far less capable inherently than Piledriver. For example; Piledriver has a 4-wide decoder with AVX capability as opposed to a 2-wide decoder in Silvermont. Silvermont has one 64-bit LSU while Piledriver has two. Piledriver has 512-bits worth of SIMD execution units whilst Silvermont has only 256-bits worth.
There are few benchmarks where the Z3740 does worse than the A6-4455M; one of which is the Cinebench R10 single benchmark, which I assume the A6 won because I assume Cinebench was utilizing AVX instructions which Silvermont doesn't have (IIRC Silvermont has only two 128-bit SIMD EUs; and unlike Piledriver's similar setup, they CANNOT fuse to execute a 256-bit instruction...)
Now see this is shocking to me. With Piledriver at 2.1 GHz and Silvermont at 1.86 GHz, it almost makes no sense for Silvermont to be ahead. Though this seems slimly possible keeping in mind Piledriver's front-end issues.
Anyways, what the point of this thread is:
Is the Z3740 really more powerful than my A6-4455M? Will it enable me to run my games faster?
If not; would a Jaguar offering be in the cards?
(If not I'll just jump to Haswell; but I'd rather use as little power as possible.)
This is an interesting topic, to be sure. That A6 model is actually a single core CPU, despite what AMD's marketing says. The Z3740 is a quad core, 4 thread CPU, which explains the performance difference. The integrated graphics in the AMD (Radeon 7500G) isn't powerful at all, again despite AMD and manufacturer's touting the "AMD Radeon" inside, which makes us think of grand dedicated GPUs used in gaming machines. It's max clock speed is lower than the Atom in question, though in other respects it's got a bit more power.
AMD's newer A6 APUs attempt to take on Intel Bay Trail, while Piledriver is last gen-- also makes a difference given Moore's Law and how fast CPUs advance.
Another important thing to note is that power consumption is high relative to performance on the A6-4455M: it's a 17 watt part while Intel Haswell Core CPUs consume just 15 watts, and the Intel Atom Bay Trail 2.2 watts. Battery life is generally much better with the Bay Trail Atom vs. other x86 processors.
However, I wouldn't say that a Z3740 is going to run circles around your machine. The real world difference isn't all that far apart. You'd have to look to Haswell Core CPUs for a serious jump, especially for gaming.
-------------------- Lisa Gade Editor in Chief, MobileTechReview
Ah yes, what you say definitely does have some truth to it; however, (and I mean this respectuflly) your reply doesn't quite answer my question fully.
The A6-4455M isn't exactly a "single core" CPU. It's in between a dual and single core; having a single core's worth of the architecture's front-end (it only has one instruction fetch pipeline and one instruction decode pipeline), as well as a single core's SIMD execution cluster (it assigns one SIMD cluster per module), however, what it duplicates in order to allow itself be labelled as a "dual core" is the integer execution pipelines and memory pipelines, and while these are the heart of every CPU's architecture, AMD's been running into a lot of front-end challenges; as expected. (Hows that for a run-on sentence? :P)
However, this doesn't quite explain as to why, even in a number of single-threaded benchmarks, the Atom comes out on top. eg. The SuperPi benchmarks; despite only running on a single thread, run ~33% better on Silvermont than on Piledriver... This is very interesting and the prime source of my confusion.
Also, you're definitely right about the graphics marketing scheme; while the Radeon series uses the same GPU architecture per generation (the architecture of my 7500G is VLIW4; the same architecture used in the GPUs such as 6970, so they're technically "OK" to market stuff like this... Unfortunately.)
It seems that, to get a noticeable jump in performance, I will have to jump to Haswell (I'm not exactly against that, I just want to go as quiet, cool, thin, frugal, etc. as possible; and obviously Atom is right there with these ideals.); but out of curiousity, what do you think of AMD's Kabini/Temash platform using Jaguar CPU cores and GCN GPU cores? Have you ever reviewed a unit with these technologies?
Also, if its not too much to ask, Lisa; have you ever found that devices with Bay Trail ever throttle speeds much? That's part of my suspicion as to why there is such a stark difference in CPU benchmark scores (atleast it's part of my many suspicions as to why this is...) and I haven't been able to get my hands on a Bay Trail part to see how hot it'd go...