Are any of these benchmarks using 64-bit code at all? A57 is quite a bit slower when executing 32bit or Thumb code.
Are any of these benchmarks using 64-bit code at all? A57 is quite a bit slower when executing 32bit or Thumb code.
do you have a source on that?
Samsung could enable 64-bit once Android 5.0 arrives, here's hoping they will
Please see the picture Sweepr posted just above your post. It shows 32 bit and 64 bit code running on A57.
In addition to what was mentioned in this picture the speed-up also comes from the streamlined and extended ISA in AArch64 mode:
- 31 instead of 14 GP registers available
- barriers with release and acquire semantics
- integer divide
- NEON/VFP no longer optional
- NEON/VFP embedded into ISA instead of using co-pro interface with separate status flags
- 64-bit AAPCS enables more efficient procedure calls (e.g. 8 parameter/return + 8 scratch)
Why? It might drain battery faster and I'm not sure how long it could sustain all 8 cores at high clocks but some apps might benefit from this capability. Also Cortex A57 @ 1.9GHz seems to beat Cortex A15 @ 2.2GHz in the single-core score.
Are you sure it's using all 8 cores in Geekbench? MT score seems to be in line with the ST score for a quad-core part - 3.15x scaling for Tegra K1 and 3.4x for Exynos 7 Octa.
No need to get defensive, as I said, many companies would do/do the same to make their products looks better (Samsung included).
I was at a local Bestbuy checking out the Note 4 (S805) and they all scored way better than the AT review tables show. (?) I took a screenshot from a display model. Not that I put much stock on mobile benchmarks (especially SunSpider) but the difference is way too big, I think?
AT uses Chrome. You were probably using the stock Android browser which is much more effective at benchmarks like SunSpider.
I always wondered why AT insisted on crippling the Android devices in these web based benchmarks...
I get why they use Chrome to compare across Android devices, but it does seem a bit unfair to compare against an optimized iOS-Safari score and not at least include Android-default browser in the actual device review. It doesn't need to be included in the long term comparison tables, but at least include it in the primary review.
They will be 10-20% higher. In any case, that's the real world benefit of 64-bit.Can't wait for 64-bit scores.
This CPU is not going to be in any large tablets for the holiday season, is it? I see the Galaxy Tab S 10.5 has what seems to be the previous iteration of this new Octa. The Tab S 10.5 is what I was planning on getting my wife this year, her TF300 just seems slow now.
There's a new Galaxy Tab S 10.5 (SM-T805S) coming with Exynos 5433 this month, but it's not clear if it will be launched outside of Korea.
http://gfxbench.com/device.jsp?benc...b+S+10.5+(Mali-T760,+SM-T805S)&testgroup=info
Another round of results.
DroidFish (Chess Engine)
Galaxy Note 4 - Exynos 5433 (4x A57 @ 1.9GHz + 4x A53 @ 1.3GHz)
2 Threads: 1205
4 Threads: 1473
8 Threads: 1520
Galaxy Note 3 - Exynos 5420 (4x A15 @ 1.9GHz + 4x A7 @ 1.3GHz) @ 4 Threads: 1033
Other DroidFish results from popular devices:
Xiaomi MiPad (4 cores Tegra K1 32 bit) 1.357.000 !!!
ASUS Transformer TF701t (4 cores Tegra 4 1.7 Ghz ): 1.054.000
IPhone 6 (2 cores Apple A8 64-bits 1,38 GHz) 1.270.000
Teclast P98HD (8 cores MTK8392 2 Ghz) 1.241.000
Ipad Air (2 cores 64 bits 1.4 Ghz) 1.218.000
iPhone 4S (Apple A5): 225.500
iPhone 5 (2 cores): 476.800
Iphone 5s (2 64bit cores 1.3 Ghz) 1.152.000
iPad Mini 2 (2 cores 64 bits 1.3 GHz) 1.128.000
iPad4: 510.000
Nexus 10 (2 cores): 510.000
Google Nexus 5 (4 cores): 508.000
Samsung Galaxy 3 (2 cores): 333.567
Note 3 (4 cores Snapdragon 800 2.3 GHz): 581.000
Galaxy S3 (4 cores Exynos 4412 1.4 Ghz): 574.000
1.9GHz Cortex A57 beats 2.2GHz Cortex A15 (32-bit Tegra K1) by 8.5% in this benchmark (1473 vs 1353) @ 4 Threads.
This seems to use bitboard, so it would definitely benefit from 64-bit. Are your figures for x86_64?Look like it use Stockfish chess engine, interesting since this allow cross comparison with X86 plateforms, as a clue at 2GHz Kaveri 4C/4T would do about 2340 Knps while an equally clocked i3 Haswell should be at 2490 Knps, a 2C/2T Haswell being at 1950 Knps, Stockfish is optimsed up to SSE4 for Intel/AMD wich bring 5% better perfs and use also BMI instructions in Haswell that bring another 2%, so the soft is perhaps favouring thoses uarchs but it put in perspective the perfs of phones SoCs, that is, the equivalent of Kaveris or i3s clocked at about 0.8-1.2Ghz...
1.9GHz Cortex A57 beats 2.2GHz Cortex A15 (32-bit Tegra K1) by 8.5% in this benchmark (1473 vs 1353) @ 4 Threads.