You've just described the ARM server business case. Do a google for:
calxeda ECX-1000 Overview
And click on the first link.
It's too early to tell if Intel "wins" it. Even with the Intel's latest micro server chips, those ECX-1000 SoC's have a higher embedded bandwidth (yes, they have less CPU power, but that will change in a couple of weeks).
If the micro server is 'valid' market segment (debateable) then the ARM guys will have some of it, it's just a question of how much (or how little).
[And, um, exactly how would that happen?]
If you look at the smart phone app process market, it's QCOM (at the high end) that Intel will take share from. ARMH earns cents per SoC shipped.
So, if ARM isn't moving up on fundamentals, what does that leave?
Well, manipulation. If I'm missing anything here, please point it out.
Huawei has licensed the ARMv8 architecture (means they will be designing their own core to the ARM ISA).
With respect, you are being naive.
No one knows what the size of the micro server market will be. It could be zero, or as some are reporting, double figures of the server market as a whole. Intel's share could be 100% or it might not, it really is too early to call.
If Smith actually said "Intel is going to do it better, cheaper and faster", then, yes I do think he is being naive. Even Intel dont know where the market (even if there *is* a market) is heading.
FYI - the ARM guys do have the money and technology, and they are shipping today (with the main ARMy arriving next year).
Perhaps you don't understand what the news means? Regardless, I really don't understand why you have this constant commentary about what ARM is or isn't doing? You seem to think ARM are in some kind of direct competition with Intel?
Bay Trail looks good, for a closer 'apples to apples' comparison:
Sweetheart, as Alexander pointed out, this Bay Trail score is for a tablet. Only your Tegra 4 (Shield) and Apples A6X scores are tablet form factors. Hence 'Apples to Apples'.
BTW, I used a lower, more realistic score for T4 (the one you posted had baytrail losing to T4 in the Multicore Score).
No, I am not being disingenuous. There are no Tegra 4 or 5410/5420 in a tablet on Geekbench 3 (or if they are I can not find them?).
The bulky Shield as you put it, has a smaller battery than many 10inch tablets and can outlast many of them in battery tests. Yes, she has a fan so has more control over thermal throttling but I don't know how long GB3 runs for to know if this is an issue or not.
Yes, the screen will have an effect on battery (LTE isn't part of the tests). Even in heavy CPU bias tests, she did better than most tablets.
If we look at the 5410 (1.6 clock, phone) scores V T4 (shield), the T4 is typically 20% faster in the single-core score, while 35% faster in the multi threaded scores. The clock differential is 18.75%. The better multi threaded scores is likely due thermal throttling (or a better memory subsystem on the T4). The T4 is the latest A15 implementation on GB3, so you would expect it to have the highest scores (clock v clock). The 5420 should arrive soon, and it could have similar scores to the T4 (1.8 clock and faster memory sub system).
There is always the Nexus 10 (dual A15):
- it appears that geekbench3 solved the 32-bit versus 64-bit performance difference. 32-bit even seems a little faster which is what I would expect for benchmark code.
FYI: On GB2 on Android for ARM and x86, -Os was used. For GB3:
-O3 -ffast-math -fvectorize -march=armv7-a -mfloat-abi=softfp -mfpu=neon
-O3 -ffast-math -ftree-vectorize -mtune=atom -maes -msse2
both on GCC 4.8.
GB2 used GCC 4.6 (with -os which I think spat out 16bit code for ARM). Basically, the ARMy needs modern compilers to spit out decent code. BT also has an advantage with the AES tests which I'm sure the ARMy guys will start moaning about:)
The -maes switch will spit out optimized code for AES, which should use the AES instructions I believe, but someone would have to do some disassembly to confirm. ARMv8 also has support for AES.
To search on Geekbench 3 use:
This isn't strictly true.
The bulk of Android 'apps' are Java, so Intel and the ARMy are on a level playing field.
The rest of the Android apps are native code (such as games) and where Bay Trail has to emulate, they yes, performance will suffer. That said, Intel are making it much easier to deploy native apps for x86 (using ICC).
bypassing the power management?
Probably, but you can bet that the BT tablet does as well.
The interesting thing about the comparison you made is that the S800 (in a phone) keeps up with BT (in a tablet). Infact, if you remove the AES scores, S800 and BT are almost identical (for single threaded integer).
Geekbench is run by end users and as such you'll get a wide range scores for the same device (due to a wide range of factors). When choosing a score, you'll need to make sure it's somewhere in the mid range.
The scores you posted are certainly within the 'normal' range of variation, but if a whole range of scores has been boosted, then yes, GB3 would have been added to the 'go fast as you can' in the power management settings.
This power managment 'cheat' is very different from the Antutu 'cheat' that Intel did:
Use a different compiler
Use different compile switches (Intel, optimize for speed), while the ARM settings optimized for size)
Optimize out a section of the benchmark.
Any one of the above invalidate the point of the benchmark.
1/5th of the size of Atom is still huge by ARM micro controller standards, could be larger than the A7.
FYI: The economies of scale at the micro-controller size is different. You don't need leading edge silicon to produce those $1 chips. ARM is safe at the bottom.
They've tried this before (albeit with Atom at TSMC) and had few takers.