LightBlog

dimanche 4 septembre 2016

GeekBench 4: How The Processor Ranking Changed Under the New, More Accurate Benchmark

Geekbench 4 has been released, and we've seen a bit of a rebalancing of scores. Some chips have seen their scores rise, while others have fallen drastically. With Geekbench 4, Primate Labs have attempted to create the most accurate version of Geekbench yet, and they seem to have done a fantastic job.

All across the tech world, from journalists to reviewers to software maintainers, the response to Geekbench 4 has been absolutely glowing. So, without further ado, we've prepared a bit of a look into how things have changed since Geekbench 3; how the pack has been reshuffled.

Now, keep in mind, scores are not directly comparable between Geekbench 3 and Geekbench 4 (Geekbench 3 is normalized around an Intel Core i5-2520M having a score of 2500, while Geekbench 4 is normalized around an Intel Core i7-6600U having a score of 4000), so we cannot directly compare how chip scores have increased or decreased, however a look at the positioning of the chips relative to each other can give some insights into how they perform.

Geekbench 4 Comparison Single Core Average Geekbench 4 Comparison Single Core Average Full Chart Geekbench 4 Comparison Single Core Best Geekbench 4 Comparison Single Core Worst

First up is single core performance. Almost every chip improved relative to the Samsung Exynos 8890's average score, except for the Qualcomm Snapdragon 820. The Nvidia Tegra K1 with Denver cores even surpasses the S820 in single core performance after the changes, despite the S820 having a substantial lead in Geekbench 3.

The A72 cores in the Kirin 950 (from the Huawei Mate 8) in particular saw massive improvements, bringing it from behind the E8890, the S820, and the K1, to being almost level with the E8890.

This would indicate that both Samsung's M1 cores and Qualcomm's Kryo cores were previously being overrated compared to their competition, rather than just Qualcomm's Kryo cores. That being said, Qualcomm's Kryo cores do see a significantly larger drop than Samsung's M1 cores.

One of the biggest takeaways from this is that it indicates that ARM really seems to know what they're doing when it comes to single core performance, and it calls into question whether the whole "real men use custom cores" mentality actually has any merit, or if it is purely for marketing. Keep in mind, that while Qualcomm's initial rushed A57 implementation in the S810 did run into issues at 20 nm, the Samsung implementation of the A57 cores on their 14 nm process performed substantially better, and saw their position improve on Geekbench 4.

Geekbench 4 Comparison Multi Core Average Geekbench 4 Comparison Multi Core Average Full Chart Geekbench 4 Comparison Multi Core Best Geekbench 4 Comparison Multi Core Worst

In the multi-core test, thermal limits become apparent. While in the single core test every chip saw improvement relative to the E8890 except for the S820, in the multi-core test the Kirin 920, S600, S810, and older Exynos chips (except for the E7420 found in the Galaxy S6, the Note 5, and the Meizu Pro 5) dropped relative to the E8890 as well.

Processor Core Geekbench 3 Single Geekbench 3 Multi Geekbench 4 Single Geekbench 4 Multi
Exynos 5433 A57 1145 4033 948 3053
Exynos 7420 A57 1267 4290 1272 3915
Exynos 8890 M1 2161 6480 1761 5199
Kirin 920 A15 850 3042 740 1730
Kirin 950 A72 1691 6294 1703 5346
Snapdragon 600 Krait 630 2273.5 703.5 1670.5
Snapdragon 801 Krait 920 2599 960 2344
Snapdragon 805 Krait 1021 2881 1004 2508
Snapdragon 810 A57 1013 3451 1155 2450
Snapdragon820 Kryo 2357 5339 1573 3520
Tegra K1 Denver 1880 3195 1643 2616

One thing that is interesting to note is that there is only one phone that scores worse in the multi-core test than in the single core test. The OnePlus X on certain builds appears to have a scheduler issue causing it to score only around 880 on the multi-core section, while putting up around 960 on the single core section. On builds without this issue, the OnePlus X scores around 960 on the single core test, and around 2400 on the multi-core test (which is in line with how an S801 based phone would be expected to perform). The Kirin 950 in particular sees the most interesting jump, outperforming even the Exynos 8890. This may be due to the slightly lower clock speed of the A53 LITTLE cores on the Exynos 8890 compared to the same cores on the Kirin 950.

Primate Labs has been commended for the improved accuracy of Geekbench 4, and I have to agree. These results line up very well with what SPEC was showing previously, except with a fraction of the time and effort needed to obtain results. Making it easier for more people to report accurate data brings further questions about the changing SoC market. While Qualcomm is still very strong in terms of GPU and cellular radio performance, they seem to be losing the CPU edge that we saw back in the days of  Scorpion and Krait cores. We'll be taking a closer look at those questions later this week.

Also, keep an eye out for our in-depth interview with the CEO of Primate Labs, John Poole, which will be coming up shortly.



from xda-developers http://ift.tt/2cr3aJ7
via IFTTT

Aucun commentaire:

Enregistrer un commentaire