noob2222 :
juanrga :
noob2222 :
@juan
answer one question with proof honestly if thats even possible. What is the TDP of an ARM A7, A9, or A15 cpu?
I just want to see how horrible x86 power consumption is compared to ARM in actual terms.
Try if you can, to put some proof into your theories other than marketing gimmicks or your own personal opinion.
If you were really interested in this topic, you would take a look to all the TDP and power consumption figures mentioned before, instead asking me for the same data.
You would see power consumption and efficiency for older A9 and A15, you could see how dual-core phone-TDP-rated Apple A7 SoC competes with quad-core tablet-TDP-rated x86 Silvermont or with a quad-core laptop-TDP-rated x86 Jaguar.
You would see how the estimated TDP for Seattle was quoted and how its efficiency is superior to best-class Xeons.
You would see the plans to build a ARM supercomputer about 1000 more powerful than current x86 top supercomputers, but using less power than today.
And so on, links included.
But asking as if this data was not given before... and then pretending that all this is the "personal opinion" of someone clearly shown what is the real goal. :lol:
Since you can't do a simple task, ill do it for you.
http://www.phoronix.com/scan.php?page=article&item=phoronix_effimass_cluster&num=1
Last week I shared my plans to build a low-cost, 12-core, 30-watt ARMv7 cluster running Ubuntu Linux. The ARM cluster that is built around the PandaBoard ES development boards is now online and producing results... Quite surprising results actually for a low-power Cortex-A9 compute cluster. Results include performance-per-Watt comparisons to Intel Atom and Ivy Bridge processors along with AMD's Fusion APU.
page 11:
The average power consumption for the Ivy Bridge system was 107 Watts under this load, which worked out to 2.58 Mop/s per Watt. The Ivy Bridge system for this workload was even more efficient than the PandaBoard ES at 1.78 Mop/s per Watt. The Ivy Bridge system was even with an SSD and other attached components requiring additional power than the ARM setup.
page 12
The efficiency was at 85 Mop/s per Watt compared to the Effimaß cluster at 30.79 Mop/s per Watt.
so an Ivy bridge DT cpu is 44% to 180% MORE efficient than a 12core ARM cortex a9. This isn't a high efficiency server cpu. The AMD comparison was kinda old. 40nm bobcat used and not a temash/kabini.
"oh what about the A15"
well, at times the speed of an A15 is only 27% of the I3 330M and at best 67% on scimark ... remind me again what gen the 330M is? and the power test ... not actually done as usual, just assumptions.
The Core i3 330M has a 35 Watt TDP while the Exynos 5 Dual operates within a few Watt envelope. Unfortunately due to the varying displays and other hardware differences, an easy power consumption / performance-per-Watt comparison couldn't be done for this article.
this is of course assuming the 330M runs at 35W all the time (if it even reaches 35W) and not the 3380M wich is also 35w @ 2.9ghz instead of 2.13, wich would have blown the cortex A15 away in terms of raw performance.
Therefore you decided to ignore the data and benchmarks given to you, and instead looked for the entire internet to find something that you believe is supporting your anti-AMD/anti-ARM view:
so an Ivy bridge DT cpu is 44% to 180% MORE efficient than a 12core ARM cortex a9.
LOL
Before starting, some basic keywords: low-cost, cluster, ARMv7, development boards, Cortex-A9...
Then some key parts from the review that you omit.
First, you omit how the cluster compared to the Intel Atom and the AMD Zacate.
The PandaBoard ES did better than the Intel Atom hardware that was tested in terms of performance and energy efficiency
Comparing the Effimaß cluster to the AMD Fusion E-350, the Zacate APU had better raw performance but the ARM cluster was the performance-per-Watt leader.
Therefore the old A9 _phone_ chip was able to compete with two desktop/laptops _x86_.
Second, you believe that ARM was outperformed by a top i7-3770k, but you ignore that this review was _not_ about comparing a chip to another chip. It was about comparing an i7 chip against a cheap _cluster_ made with ARM _development boards_.
While this do-it-yourself ARM cluster configuration is not the most effective setup right now, it will be interesting to see how the cluster performance works out for the next-generation ARMv8 hardware as well as the many ARM core servers coming out, such as the upcoming products from Calxeda.
Indeed, the _cluster_ was not effective. Michael built it merely for development/testing purposes. His goal is to port the benchmark suite to ARM, not to build an efficient or fast cluster. In fact, the scaling in the NPB LU benchmark that you pretend to use for evaluating the efficiency of ARM was rather bad:
The last NPB test for looking at the scaling is LU.A. The LU pseudo-application is a Lower-Upper Gauss-Seidel solver. This workload did not scale as well across the cluster with going from one to twelve cores just resulting in a 4.8x performance improvement. However, this was not a failure of MPI or the PandaBoards with the scaling when going from one to two cores on a single PandaBoard ES just yielding a 29% improvement.
The _cluster_ offered only a part of the maximum performance of all the ARM chips togheter. The efficiency loses are related to the cluster setup, such as the 10/100 Ethernet and a shared NFS mount from an SDHC card. Therefore, that review was _not_ comparing ARM to x86 efficiency.
Third, the x86 chips used optimized software, the ARM cluster used buggy software:
Ubuntu 12.10 offers some remarkable ARM performance gains on the OMAP4 hardware due to the newer Linux kernel (version 3.4 at present, compared to Linux 3.2 on Ubuntu 12.04) and the major compiler upgrade (GCC 4.7 vs. GCC 4.6), but due to some early configuration problems with the post-alpha-one snapshot, the installations were reverted to Ubuntu 12.04 LTS. Ubuntu 12.10 will be loaded up on this compute cluster in the coming weeks and should result in double-digit gains.
Fourth, those Fortran-based benchmarks were testing floating point (FP). As I said before, but some people here don't read, FP has been the weakest aspect of the ARM arch. It is not a problem with RISC or ARM, simply you don't need a strong FPU in a _phone_. Therefore, ARM _phone_ chips as this tested A9 have a weak FPU. This changes dramatically with the new A-57 based in the new ARM64. FP has been improved a lot of in the new arch. In fact, the FP performance of the new A57 is about _4x_ the performance of the A9.
You mention how old is Zacate, but Jaguar is not 4x faster than Zacate.
In any case, heavy FP computations will be made in GPGPUs/accelerators. No top x86 supercomputer uses CPUs for FP. The ARM supercomputer will use ARM + CUDA GPGPU. Also, ARM is founding member of the HSA foundation.
Finally, good work of trying to compare the raw performance of a dual-core @1.7GHz _phone_ A15 chip constrained to run in the single digit TDP against a dual-core + HT @2.13GHz i3 allowed to consume ~10x more energy. LOL
Moreover, you gave only a part of the quote, cutting-off the most interesting parts. Whole quotation:
Overall the performance out of the Samsung Exynos 5 Dual on the new Chromebook is very attractive. While Ubuntu on the Chromebook isn't perfect (the broken touchpad and sound support, etc), for those looking towards the ARM Cortex-A15 for development purposes or as a test bed for experimenting with Linux on ARM, the Samsung Chromebook is a very attractive bargain priced at $250 USD.
It was surprising to see the wide performance margin the dual-core 1.7GHz A15 had over the quad-core 1.4GHz A9 in the Tegra 3. In a majority of the cases, the Samsung Exynos 5 Dual also easily beat out all of the tested Intel Atom processors. And then there was the Intel Core i3 330M, which was faster, but on the performance-per-Watt this would be a very different story. The Core i3 330M has a 35 Watt TDP while the Exynos 5 Dual operates within a few Watt envelope. Unfortunately due to the varying displays and other hardware differences, an easy power consumption / performance-per-Watt comparison couldn't be done for this article.
Aka a dual core phone-TDP-rated ARM chip beating a dual core + HT x86 laptop/desktop-TPD rated with higher clocks.
8350rocks :
juanrga :
The data about the A9, A15, Apple SoC... is _not_ based in marketing slides _neither_ in theoretical numbers.
The estimation of the TDP of Seattle was partially confirmed by Feldman (AMD) during an interview. AMD statements about the expected performance of A57-based products match very well with info and benchmarks that I have from ARM.
The predicted efficiency of the ARM supercomputer is based in theoretical numbers extrapolated from the _prototype_. That they are theoretical numbers is _evident_ because the supercomputer has not been still built :lol: However, those numbers are made by experts in the topic (e.g. the same supercomputer center that helped IBM to develop their supercomputer CPUs) and have been reviewed for _funding_ by a scientific European program panel of experts. Even if everyone involved is wrong by 100%, the final efficiency will be much higher than current x86 supercomputers.
As I already mentioned you at least _twice before_, there are ARM CPUs running "DT grade OS". There is no essential difficulty with coding desktop OSes and desktop applications for ARM.
You don't have any argument, benchmark, theoretical argument, link or anything rocks; you only have (i) elementary misconceptions, (ii) trivial statements, and (iii) continuous negation of the data and links given to _you_ again and again.
What I find very interesting is that _you_ very often buy
rumours about Steamroller, Excavator, Radeon, MANTLE... and _you_ are often over-enthusiast about those rumours, even when AMD doesn't confirm them. However, now, suddenly, anything said by AMD about ARM either is wrong or has to be put under 90-days quarantine. This double standard is very funny.
Now feel free to reply with some rant or insult 
First, I have never insulted you, nor have I ranted.
Second, sure, A7/A9/A15...we have those numbers, and DT CPUs blow them away. It's not even close, when I played baseball in college, we called that a "boat race".
I am talking about your theoretical performance projections about "the A57 will be <insert crazy claim with no proof here>"...so is noob and several others.
By the way, if the A57 is anything like the aforementioned ARM CPUs in performance and power consumption, then I hope you have a good recipe for crow. Because you will surely be eating a heaping helping of it when the dust has settled if the A57 isn't something like 200%+ greater performance at similar tasks...
That was my point all along...though you seem to gloss over it and resort to name calling. Stop assuming that insults will fly because you talk in such a condescending manner. You talk to everyone who doesn't feel that ARM will rule the world in the next 6 months like they're of such low intelligence...it's borderline insulting in many instances.
EDIT: Just a note, but if you actually read the phoronix review of 12 ARM cores against Intel and AMD architectures. A dual core
E-350 embedded AMD, is about on par with 5 ARM cores.
The ARM cores were clock at 1.2 GHz and the E-350 at 1.5 GHz.
Sure, the ARM cores took less power...however, the E-350 still beat 4 of them in performance.
The 3770k even beats 12 ARM cores
in the parallel benchmarks no less, and there were 12 ARM cores!
As I said earlier, ARM will not beat x86, it's just a nice diversion into mobile territory. You ranting about ARM is no different than "he who shall not be named"* ranting about PPW.
I dare not mention it, as he would surely show up...
After being corrected about it in innumerable occasions you continue with your pretension of comparing _phone_ ARM chips to _DT_ x86 chips. LOL
Let us use your 'logic', at least during a crazy instant. We take a Temash chip sub-10W TDP and compare the raw performance to a sub-100W TPD Haswell chip. Using your 'logic' we conclude that Intel "blows" AMD away. Of course, this is a silly comparison and the conclusion is invalid, but this is exactly what you _do_ in your anti-ARM crusade.
You forget again that 'my' performance projections are based in data disclosed by both ARM and AMD. It is funny that you accept marketing slides from AMD when they are about Kaveri/Steamroller, but you call them "crazy" when are about Seattle A57. Double standard.
You continue grossly misinterpreting my position and just ignoring what I said.
Who said you that "that ARM will rule the world in the next 6 months"? Offer us the link to the message saying that.
You cannot because this is _your_ claim and only from you. The fact that you continue attributing this kind of crazy statements to others, after being corrected about it, and after being warned about stopping this silly miss-attribution clearly indicates which is your real goal here. I predicted it above.
I wrote above some basic thoughts about the phoronix review of 12 ARM old phone cores _cluster_. Keywords here are _cluster_, _phone_, and _old_.
I find funny that you mention "he who shall not be named" because both of you are using the same tactics. He also ignores what is said and repeat the same mistakes/nonsense despite being corrected again and again.
). i assume core licensees are less capable of implementing benchmark optimizations. if true, it might make them (core licensees) appear more honest regardless of their intentions, albeit helpless against experienced cheaters. arm ecosystem is evolving like x86, but at a faster rate, since intel can't get in the way. that means bad things will also happen much earlier and on a much bigger and deeper scale. 
Facebook
Twitter
Instagram