It is well-known at this point that Samsung (NASDAQOTH: SSNLF) is planning to use its own Exynos applications processor inside of its upcoming Galaxy S6 flagship smartphone. Many believe that this processor will be built on Samsung's new 14-nanometer manufacturing technology, which should give Samsung a leg up on its competition which will be using Qualcomm (QCOM 1.41%) chips built on a less advanced 20-nanometer manufacturing technology.
Interestingly enough, I'm starting to have some doubts on whether the Exynos chip that is expected to power the Galaxy S6 will be built on Samsung's 14-nanometer process. Here are a couple of reasons why.
Reading more closely into Samsung's comments on the earnings call
At an investor conference some months back, Samsung claimed that it had begun processing the initial 14-nanometer wafers for mass production. However, on the company's Jan. 29 earnings call, Samsung's Ji Ho Pak had the following to say with respect to the company's 14-nanometer ramp:
For the first quarter, we will start 14-nanometer mass production in earnest and begin supply for customer's new product.
The phrase "we will start 14-nanometer mass production in earnest" suggests that the company's prior comments of having begun "mass production" may not have meant what everyone (including yours truly) thought it meant.
Indeed, if Samsung is beginning mass production on its 14-nanometer technology in the late January/early February timeframe, then this doesn't seem like a particularly long lead time for the launch of the very important Galaxy S6 in early March. For a roll-out in March, Samsung needs to be building these phones in earnest now, which might be difficult to do if there isn't adequate supply of processors.
On the other hand, Samsung has been ramping its 20-nanometer production for a while now, so it wouldn't be terribly difficult for the company to mass produce a bunch of chips for the S6 on its 20-nanometer technology.
The performance improvement over the Exynos 5433 isn't that huge
AnandTech's Andrei Frumusanu pointed out on the website's forums that an Exynos 7420 had appeared in the Geekbench 3 database. Geekbench 3 is a popular CPU performance test frequently used to compare the performance of various chips.
Here's the highest performing Exynos 7420 in the Geekbench database compared with one of the higher performing Exynos 5433 chips (found in the Galaxy Note 4):
Source: Primate Labs
Notice that the Exynos 7420 is quite a bit faster than the Exynos 5433 -- about 20% faster. Now, one could easily chalk this up to the increased frequency headroom that the 14-nanometer technology allows over the 20-nanometer technology. That said, it's worth taking a closer look at where the 7420 is faster than the 5433.
Look at the following sub-scores:
|
Exynos 5433 |
Exynos 7420 | |
|---|---|---|
|
Integer (single core) |
1561 |
1726 |
|
Integer (multicore) |
5964 |
7562 |
|
Floating Point (single core) |
995 |
1279 |
|
Floating Point (multicore) |
3798 |
5200 |
Source: Primate Labs
Note that the Exynos 5433 is running in 32-bit mode on Android 4.4.4, while the Exynos 7420 is running in 64-bit mode running Android 5.0. Now, you'll note the relatively modest gains in going from the 5433 to the 7420 in single-core integer performance (10.5%), but there are huge gains in going from the 5433 to 7420 in floating point performance per core (28.5%).
This is where things get interesting. In the table below, I show Apple A7 scores in Geekbench running in both 32-bit mode and in 64-bit mode (32-bit results courtesy of the AnandTech review of the iPhone 5s):
|
A7 32-bit |
A7 64-bit | |
|---|---|---|
|
Integer (single core) |
1065 |
1465 |
|
Integer (multicore) |
2095 |
2870 |
|
Floating Point (single core) |
983 |
1346 |
|
Floating Point (multicore) |
1955 |
2645 |
Source: AnandTech, Primate Labs
Notice how there is a significant uplift just from the move from 32-bit to 64-bit? In fact, given the improvement that the A7 sees in moving from 32-bit to 64-bit is larger than the difference between the Exynos 5433 and the Exynos 7420, I am now pretty sure that the Exynos 7420 may essentially be an Exynos 5433 with full 64-bit support and LPDDR4 memory support (since Samsung executives stated on the company's earnings call that LPDDR4 would essentially be required for flagship phones this year).
If the 7420 were truly a 14-nanometer chip, then I would expect a much greater performace uplift, if only due to significantly higher clock speeds.
If these Exynos 7420 numbers are representative of the final shipping performance of the chip inside of the Galaxy S6, then I now suspect that the Galaxy S6 will feature a 20-nanometer chip rather than a 14-nanometer FinFET chip as widely rumored.
