The comparison was against the two types of processors in the Samsung S4, one of the most recently introduced phones based on ARM processors and one of the fastest ARM based units on the market. I don't know why you would not consider that a fair comparison?
> Technical consulting firm BDTI pointed out that the compiled code for the Intel processor was not executing all instructions that were intended for the RAM test.
>To rectify the situation, AnTuTu issued revision 3.2.2 to the benchmark Wednesday evening. The revision still uses the ICC compiler, but the resulting scores are drastically different for the Intel processor. The AnTuTu CPU and overall scores dropped by approximately 20 percent, while the AnTuTu RAM score plummeted by approximately 50 percent
If these numbers are for real, then the topend smartphone devices, will move to Intel shortly .. ie almost same performance at half the current drain of ARM chip!! is there any catch? are we missing something here?
"well positioned for strong growth over the next few years." In number of shipped units.. yes .. But that's not what investors look at.
Android is Linux underneath and runs just as well on x86. I doubt if there are many software hiccups, and most aps would need a simple recompile. I would say most aps are developed on Linux workstations.
Pretty significant results from the Intel camp. I am not pro or against Intel, but this will have some impact on the naysayers out there. Unless ARM or Qualcomm et. all can make some really fast improvements in power consumption versus processor power, they could take some big hits in sales. I would be wary of my Qualcomm stock holdings.
Because I can show independent benchmarks between S4 and K900 that prove my point:
Also if you understand microarchitectures then you'd be very very sceptical of any claims that a 2-way in-order CPU outperforms a 3-way out-of-order one, especially when it is 2 vs 4 cores as well...
Yes it is a well-known fact that browsers and JIT compilers vary a lot in performance, even between versions.
Worse, it's relatively easy to use compiler tricks to make tiny JIT benchmarks (such as SunSpider) appear much faster than they really are.
So if you want to compare actual performance then these micro benchmarks don't provide a good estimate - not for native performance and not for JIT compiled code.
It is a different ISA and most instructions are treated as a single op down the pipeline. Intel does not need to be as deeply OoO. Also, OoO depth does not scale linearly. Its not a very good metric when comparing completely different ISAs. Also, the new Atom's decoders are significantly improved over old Atom.
Unfortunately most of the popular benchmarks for mobiles are rubbish - including AnTuTu.
Based on my experience with benchmarking I'd say the scores in your link are not at all an accurate representation of actual CPU performance. If the source code was available I could explain why.
Besides SPEC, reasonably good CPU benchmarks are Geekbench and Phoronix with results available for a wide range of CPUs - not just mobiles.
The geekbench scores are not that accurate in projecting performance for Atom devices. If you look up the scores for the Motorola Razor I and Razor M, they will show a large performance gap favoring the Razor M. Actual comparisons of the phones showed the Razor I was as fast if not faster while using 15 percent less energy.
The AnTuTu bench showed near equal performance
DELIBERATELY USING CAPITAL TO DRAW THE ATTENTION OF
PETER CLARKE :
Considering the significance of these test results and the controversy it has raised ( as seen from the Comments here ) will it be possible for you ask ABI Research to provide additional technical details on (a) the Benchmark programs ( were they industry standard ? ), test procedures for measuring current nto CPU etc.
Am I mistaken or what??? The benchmark results do not show Intel outperforming the ARMs. It appears only marginally better than the Exynos one and it underperforms when compared to the others (CPU, 2D, 3D and Performance Index... forget about the RAM access, which does not mean much).
...and these numbers matter why? The reason Apple and Samsung don't publish benchmarks like this is because they don't matter. Unless you plan to run SETI@home on your mobile, these numbers mean next to nothing.
No they don't. It means that the numbers jockeys got out and crafted some comparisons that are to the advantage of Intel. Battery life is a whole different ballgame where factors such as display power consumption are more dominant. I can say with high certainty that this alone will not increase your battery life by 2x.
Agree. The benchmark setup and detailed results should be published. It is very easy to chose data that makes once processor look better than the other.
The other question is which intel parts are they using? Are these a select few parts screened from hundreds for their favorable power characteristics? There is usually silicon variation and it is possible to find a few parts that work well but these numbers will be difficult to reproduce in volume production.
Looks to me like this is Intel marketing at work.
Regardless of questions about whether these were the best benchmarks for mobile, or whether the source was Java or whatever, the point remains that it looks like Intel has made some real progress on reducing current drain while providing a level of performance suitable for a smartphone. Better, equal or just slightly worse than competing architectures is less critical right now (unless you're the Intel sales guy) than the fact that they appear to finally have a processor worthy of running on a smartphone's modest battery capacity. We will all get a much clearer picture once they are designed into a few more phones with wider deployment.
The benchmark used and methodology does really matter a lot. If you load your car with heavy stuff, do you expect to achieve the same top speed or petrol consumption?
Note that current drain is not relevant, what you really want to measure is task energy (which requires voltage and time to complete a task).
Also it's not news that Intel can make a phone CPU with reasonable battery life, eg. Lava XOLO and Razr I early 2012. The Z2580 in K900 is simply the dual core variant.
Likely not a leap to expect the current drain was measured at the battery connection likely with very similar battery voltages. It would be difficult at best to isolate. Other power traces in a phone / tablet.
wow... seems like the ARM guys are up in ARMS about this benchmark... the tenor of their alarm disguised as s(k)epeticism (yes we spell it with a K here) is hinting at their allegiance in this fight.
I really don't care, except made in the USA is always good for me.
I assume Intel with its new CEO might have pushed the Sales guys and the Dev teams as well.
Hope it turns a new page for intel beyond this +ve sounding article.
just curious what happened to FD-SOI / FinFet debates?
Atleast in paper, Intel's strategy is right. Intel can win the war or speed performance with ARM on 30$ mobile CPU business. But by winning that war, it will lose its main revenue source ie the 300$ desktop/laptop CPU business. So they dont want to make better performing(in speed) than ARM) lest people would switch over entirely to the mobile platform ditching the desktop/notebook platform. But to keep ARM on its toes, they are doubling down their efforts on power. By making a mobile CPU with half the power than that of ARM, Intel can keep ARM focus more on power reduction, than performance(speed) improvements. Which will keep the performance gap between mobile CPUs and desktop CPUs at a safe distance, that people will feel the need for buying an extra notebook/desktop apart from their mobile device. ie keep its main revenue source safe. :)
Maybe. Assuming mobile CPUs gain enough performance to replace say Core I7 (a tall order), there is still the i/o problem. In spite of the hype, there is more to replacing desktops than having a CPU with enough raw processing power. We need it to drive (multiple?) large screen monitors, external hard/SSD drives, keyboards, mice, Ethernet, etc., and with a single small cable and i/o controller that uses miniscule amounts of power. It will be a while.
I see it another way that has nothing to do with desktop workstations. Smartphones are following the PC evolutionary path. With increased performance comes more capable aps, which in turn drives the need for more performance. Human wants and desires are boundless. Once the more capable aps are in use, there will be minimum system requirements, just like we saw as the PC evolved. There is no static "good enough" level. The software is what will drive up performance minimums. Nobody will want a device that isn't able to run the software du jour. Since software developers will certainly make use of any performance gains, a mobile processor performance war is inevitable.
Intel is positioning well. If they can deliver the same performance with less power, then a "tock" release can increase clock speed and still stay within the power and thermal limitations. That puts the pressure on ARM to increase performance, else risk their chips not being able to run the latest software. Basically, this is what Intel did to AMD for years with desktop processors.
One can argue about benchmarks all day long, but with new Haswell in Macbook Air the battery life goes from 7 to 12 hrs. Thus it is reasonable to assume that with the new Bay Trail on 22nm tri-gate Intel will have a competitive advantage over ARM.
Intel and Microsoft are teaming up on the energy issue, at least starting with Haswell. I presume Atom will follow the same recipe going forward, if not already.
The real question is whether Windows 8 is compelling enough to take significant market share in tablet and phone.
Performance aside, I am disappointed in the lack of professionalism in some of the comments. This is a professional site, please treat as such.
On the comments wrt power and energy, if you have the same performance at half the power then the total task energy will be half as well.
Ah, now you say, 'if time is equal'. That is the one, solitary case, where the statement is true.
But why would time be equal? Especially on different architectures, with different instruction sets and different processes. Also compilers and a whole lot other issues.
Under many circumstances the chip that dissipates the most power during the task could have the lesser energy consumption completing a task.
It ain't for no reason that batteries are specified in energy capacity, not power.
I did not have to say time is equal, though I probably should have clarified that by performance, I meant processing speed, though I had thought that was understood.
As we were discussing the article and it showed both processing speed and current consumption (power) then we can easily calculate task energy which is what this article was really talking about .... equal performance (processing speed) at often significantly lower power leading to lower task energy.
I detected a sense of irritation at the discussion. You complained of a lack of professionalism and then you dissed the distinction between power and energy.
Yes, it could be assumed that the times were equal, I later stated that was the only case were your general statement about performance, power and energy, would be true.
Now you bring up processing speed! That brings with it all the same problems of your abstract 'performance' you brought up originally.
Les, my irritation if any is because I believe you are arguing semantics when the details really quite simple and this discussion relates to the article. Arguing semantics is not professional.
- Task time is inversely proportional to processing speed. Note that processing speed is one of the factors listed in this article.
- Power is directly proportional to average current (again, covered in this article).
Task Energy is proportional to (power)/(processing speed).
There are no other complexities. As this article stated, processing speed for the Intel chips were on part with the ARM chips, while average (and peak) current draw was significantly less. The ONLY interpretation that can be made is that task energy will be less. Any other conclusion would be to argue semantics.
The benchmark shows what it was trying to show, the contributuon of the processor on power draw. An attempt to isolate the processor makes complete sense. Battery life of the various phones would be meaningless as there are so many other parts of the phone that play into this.
This is not a phone benchmark per se but a processor benchmark (plus likely power conversion and memory subsystem).
The article was about comparing the processing power and power drain of two competing processors. Why would it discuss other areas of the phone design? We are engineers, I think we can figure out or research those areas separately.
The reality is that processor power still does come into play w.r.t. battery life. How many people hobble their phones as communications devices by limiting how often they check mail? That has nothing to do with the display, and everything to do with energy drain of RF and processor/baseband.
As people expect more out of phones/tablets, processor power becomes more, not less important.
Intel still treats the phone as a potential home for its processor, with the view that this home is focused on the processor, just like a PC or workstation or server. But a phone is a SYSTEM dedicated to displaying or broadcasting information, so other components besides CPU, GPU, memory are also important. So they haven't really left their old mentality. Which will be a problem for them.
I really don't understand the nature of your comment.
Intel does not make displays, or touch screens, so they don't touch those pieces of a phone ... nor connectors, etc.
Last time I checked, Intel is trying to get involved in everything they can viably be involved in a phone with ... processor, memory, baseband, even an OS initiative with Samsung.
Short of building a phone, how are you expecting them to be involved?
At the end of the day, a phone (just like a computer), is nothing more than user interface, processor, memory and communications ... and something to power it.
Please explain how they have not left their old mentality and what they should be doing different?
In terms of a large battery, that means added cost, added expensive part that can fail, more weight, larger packaging, etc. .... all things consumers do not want.
Don't see how the argument is valid. If one drives a car with engine at redline, the MPG would not be anywhere near the advertized. Do you use laptop or cell running non-stop math and graphics?
On the other hand, the relative performance is meaningful, unless you happen to be on the losing side, that is.
I'm not arguing the relative comparison which shows less power consumption at comparable performance by a processor. I'm arguing this is only a small portion of the battery life consideration. Other considerations like camera use may matter more to users, in the final purchase decision.
If one was to run any of these benchmark repeatedly, the phones battery would die quite quickly. However, we know that is not the case. However, processor power&performance does come into play in terms of total battery life and hence is important.
Sure but we know battery life even under normal conditions exceeds 3 hrs, such as in the laptop battery test, regardless of processor.
My Mom the Radio Star Max MaxfieldPost a comment I've said it before and I'll say it again -- it's a funny old world when you come to think about it. Last Friday lunchtime, for example, I received an email from Tim Levell, the editor for ...
A Book For All Reasons Bernard Cole1 Comment Robert Oshana's recent book "Software Engineering for Embedded Systems (Newnes/Elsevier)," written and edited with Mark Kraeling, is a 'book for all reasons.' At almost 1,200 pages, it ...