Samsung Exynos 5250 - production starting in Q2 2012

  • Thread starter Deleted member 13524
  • Start date
Pretty much any application, even these benchmarks, have enough variability in their workloads to have the processors of these devices ramping up and down a lot to save power at every opportunity under their actual power profiles. Simply eliminating that ramp-up time by maxing out the cores constantly will deliver a substantially higher level of performance on average (and require substantially more power, which is obviously why those power saving behaviors have to exist in the first place), without even needing to expose a higher peak frequency mode.

Finding sloppy sections within an article, even in relatively informative articles, isn't hard. Ars didn't provide the supporting info necessary to make their generalizations and summarizations about percent boosts "across the board" in the performance on other benchmarks (and, as mentioned, the variance already inherent in some of these benchmarks due to their poor design can add too much margin of error to their results to be useful when quoting a specific number), so that part of the article was easy to ignore.

Also, while acknowledging that their CPU Monitor didn't show any unusual CPU boosting going on in GfxBench 2.7, they made another unfounded claim that GPU boosting was clearly in effect because of the presence of the LCD frame rate adjust function. Sloppy again, since the section of code they had reproduced apparently showed that the function wasn't being applied to anything, yet the observation, as misinterpreted as it was by the author, highlights the specific path to boosting the GPU one way... interesting in itself.

Still, the actual code they reproduced, the specific testing they described, and the specific performance they measured was enlightening all by itself.

I wasn't aware that the device you and Anandtech were testing back in July was a Note 3.

The Ars piece was titled something like "Note 3 Benchmark Boosting Inflates Scores by Up to 20%" and was focused specifically on the Note 3. They appeared to have accurately measured and compared scores to a similarly CPU-binned S800 device in the G2. Whether or not they were aware that the G2 boosts in various benchmarks too was irrelevant to their specific point and findings, especially because they neither claimed nor implied that the G2 wasn't also guilty of boosting. It simply wasn't a topic within the focus of their Note 3 article.
 
That and they're definitely aiming for 4GB RAM to justify (to the press) the 64bit capabilities of the Cortex A57 they'll probably use.

Their first 20nm chip, if it combines 3D TSV, proper big.LITTLE implementation with the Cortex A50 combo and a Mali T760 should be really fast.
 
they neither claimed nor implied that the G2 wasn't also guilty of boosting
Of course they implied it. The G2 was compared to the Note3 throughout the review, including the benchmark section, so harping on Samsung without saying a thing about the G2 is a clear implication that Samsung is dirty and LG is clean.
 
That and they're definitely aiming for 4GB RAM to justify (to the press) the 64bit capabilities of the Cortex A57 they'll probably use.

Their first 20nm chip, if it combines 3D TSV, proper big.LITTLE implementation with the Cortex A50 combo and a Mali T760 should be really fast.
I really doubt we'll see the T760 in the upcoming Exynos line. ARM themselves stated that first products are towards the end of 2014. Either we'll see some higher-end T62X or a Rogue implementation, at this point it's anybody's guess what they'll do.

A TSV implementation would give them immense advantage that I don't think the competition could imitate any time soon.

I still giggle at the sight of them using LabView at each of their live power comparison demonstrations, definitely interested to see what's going on on that screen in more detail.

Edit: I see that they actually FINALLY released the Arndale Octa board; http://www.pyrustek.com/ And it's the 5420, not the 5410 as originally planned for release. This is the final nail in the coffin for the 5410. The Arndale boards are the official dev platforms for Linaro, and it makes sense to skip the 5410 as it's effectively useless as a dev platform. They also mentioned the ODROID-XU2 which probably also has the 5420.
 
Last edited by a moderator:
The LG G2 was "clean" (no app-detect boost) for that particular benchmark, not that a point about the Note 3's boosting couldn't still be made either way. Obviously, for the relevance of the particular numbers, comparing to a device that wasn't also boosting in Geekbench was a good move on Arstechnica's part, whether they did that through intelligence or just by luck.

Again, the only focus of that Ars article was the Note 3's boosting, from the title of the article, to its sub-header, to its content. It wasn't attempting to make a commentary on the general state of Android OEM benchmark boosting, nor how honest one manufacturer is versus another in general (no commentary on Samsung in general nor LG in general was made.)

The hostility to this little Ars piece is just overdone. It made a few unintelligent, unsupported comments about boost percentages across the board in other benchmarks and also that Samsung must be doing "something" to the GPU as well, but it can at least be credited with revealing that this particular app-detect boost was being applied only to benchmarks this time, which is contrary to Samsung's claim and leaves the benchmark results to be representative of nothing. Revealing the app-detect code and also quantifying the effect of the boost in Geekbench was just good work.
 
Some bits I found interesting:
64-bit CPU core for Smart Devices 2 step approach:
1.AP with ARM’s 64-bit core
2.AP with Samsung’s own 64-bit core

I see a mention of exynos 4quad + modap. Anyone know more about it?
 
Ahhh the hexacore. Sounds IMHO like a far better perf/mW balance to have a 2+4 config than 4+4 especially if it goes into smartphones. Isn't that one though slated for the Note 3 Neo or not?

Yes it is the one used in Neo (or lite).
It actually scores more than exynos S4 in antutu(it has a better GPU maybe that is why) and supports lte could have been a competitor of snapdragon 600 if they had released it earlier.

And the first soc from Samsung with HMP enabled.
 
Last edited by a moderator:
Back
Top