The Snapdragon 888 vs The Exynos 2100: Cortex-X1 and 5nm

The new series of Galaxy S21 devices has been on the market for a week now and we have now been able to get our hands on two Galaxy S21 Ultras – one with Qualcomm’s new Snapdragon 888 SoC and the other with Samsung’s new Exynos 2100 SoC. Both chipsets this year are more similar than ever, both now displaying similar CPU configurations, and both being produced on a new Samsung 5nm (5LPE) process node.

Before our full review of the Galaxy S21 Ultra (and the smaller Galaxy S21), today we’re focusing on the first test results of the new generation of SoCs, putting them to the test and pitting them against each other in the new competitive landscape for 2021.

The Snapdragon 888

Qualcomm Snapdragon Flagship SoCs 2020-2021
SoC Snapdragon 865

Snapdragon 888

CPU 1x Cortex-A77
@ 2.84 GHz 1×512 KB pL2

3x Cortex-A77
@ 2.42 GHz 3×256 KB pL2

4x Cortex-A55
@ 1.80 GHz 4×128 KB pL2

4 MB sL3

1x Cortex-X1
@ 2.84 GHz 1×1024 KB pL2

3x Cortex-A78
@ 2.42 GHz 3×512 KB pL2

4x Cortex-A55
@ 1.80 GHz 4×128 KB pL2

4 MB sL3

GPU Adreno 650 @ 587 MHz Adreno 660 @ 840MHz
DSP / NPU Hexagon 698

15 AI TOPS
(Total CPU + GPU + HVX + Tensor)

Hexagon 780

26 AI TOPS
(Total CPU + GPU + HVX + Tensor)

Memory
Controller
4x 16-bit CH

@ 2133 MHz LPDDR4X / 33.4 GB / s
or
@ 2750 MHz LPDDR5 / 44.0 GB / s

3 MB system-level cache

4x 16-bit CH

@ 3200 MHz LPDDR5 / 51.2 GB / s

3 MB system-level cache

ISP / Camera Spectra 480 dual 14-bit ISP

1x 200 MP or 64 MP with ZSL
or
2x 25MP with ZSL

4K video and 64 MP burst capture

Triple 14-bit Spectra 580 ISP

1x 200 MP or 84 MP with ZSL
or
64 + 25MP with ZSL
or
3x 28MP with ZSL

4K and 64MP video burst capture

Code/
Decode
8K30 / 4K120 10 bit H.265

Dolby Vision, HDR10 +, HDR10, HLG

720p960 endless recording

8K30 / 4K120 10 bit H.265

Dolby Vision, HDR10 +, HDR10, HLG

720p960 endless recording

Integrated Modem none
(Paired with external X55 only)

(LTE Category 24/22)
DL = 2500 Mbps
7x20MHz AC, 1024-QAM
UL = 316 Mbps
3×20 MHz AC, 256-QAM

(5G NR Sub-6 + mmWave)
DL = 7000 Mbps
UL = 3000 Mbps

Integrated X60

(LTE Category 24/22)
DL = 2500 Mbps
7x20MHz AC, 1024-QAM
UL = 316 Mbps
3×20 MHz AC, 256-QAM

(5G NR Sub-6 + mmWave)
DL = 7500 Mbps
UL = 3000 Mbps

Mfc. Process TSMC
7nm (N7P)
Samsung
5nm (5LPE)

Starting with the new Snapdragon 888 SoC, Qualcomm’s new flagship model takes iterative steps in this generation, with the biggest changes in the new design actually being in the form of the new Hexagon 780 accelerator, which merges traditional scalar and vector DSP operations with tensor execution mechanisms in a single IP block.

Of course, we’re also seeing updates in other parts of the architecture, with the Snapdragon 888 being one of the first SoCs to use Arm’s new Cortex-X1 CPU IP, promising big performance gains over the latest generation Cortex-A77 cores. The unique X1 cores in the Snapdragon 888 are clocked at 2.84 GHz – the same as the previous Cortex-A77 core cores from the previous generation Snapdragon 865 and less than the 3.1 GHz and 3.2 GHz Snapdragon 865+ and recently Snapdragon 870 SoCs announced.

Along with the X1, we find three Cortex-A78 cores at 2.42 GHz, again the same clocks as the previous generation of 865 SoCs, but this time with twice as many L2 caches at 512 KB.

The small cores of the Cortex-A55 remain identical in this generation, clocked at 1.8GHz.

While we were expecting the top 8 MB L3 cache SoCs this year, it seems that Qualcomm has chosen to stay with 4 MB for this generation – but at least the company uses the X1 core with the L2 cache configuration of a maximum of 1 MB.

On the GPU side, Qualcomm’s new Adreno 660 GPU now peaks at 840 MHz – a frequency 43% higher than the Snapdragon 865 GPU. The company’s performance statements here are also surprising, promising an increase of + 35 % in performance. We’ll have to see how it all ends in terms of energy consumption and long-term performance in the dedicated GPU rear section.

What is quite different for the Snapdragon 888 this year is that Qualcomm has moved from a TSMC N7P process node to Samsung’s new 5LPE node – the general wildcard in this whole situation, as we have had no previous experience with this new 5nm.

The Exynos 2100

Samsung Exynos SoCs specifications
SoC

Exynos 990

Exynos 2100

CPU 2x Exynos M5
@ 2.73 GHz 2 MB sL2
3 MB sL3

2x Cortex-A76
@ 2.50 GHz 2×256 KB pL2

4x Cortex-A55
@ 2.00 GHz 4×64 KB pL2

1 MB sL3

1x Cortex-X1

@ 2.91 GHz 1×512 KB pL2

3x Cortex-A78

@ 2.81 GHz 3×512 KB pL2

4x Cortex-A55

@ 2.20 GHz 4×64 KB pL2

4 MB sL3

GPU Mali G77MP11 @ 800 MHz Mali G78MP14 @ 854 MHz
Memory
Controller
4x 16-bit CH

@ 2750 MHz LPDDR5 / 44.0 GB / s

2 MB system cache

4x 16-bit CH

@ 3200 MHz LPDDR5 / 51.2 GB / s

6 MB of system cache

ISP Single: 108 MP
Dual: 24.8 MP + 24.8 MP
Single: 200 MP
Dual: 32 MP + 32 MP

(Up to simultaneous quad camera)

NPU NPU + DSP + CPU + Dual GPU
15 TOPs
NPU + DSP + CPU + Triple GPU
26 TOPs
media 8K30 and 4K120 encode and decode
H.265 / HEVC, H.264, VP9
8K30 and 4K120 coding and
8K60 decoding

H.265 / HEVC, H.264, VP9
AV1 Decode

Modem Exynos Modem External

(LTE Category 24/22)
DL = 3000 Mbps
8x20MHz AC, 1024-QAM
UL = 422 Mbps
? x20MHz AC, 256-QAM

(5G NR-6 NR)
DL = 5100 Mbps

(5G NR mmWave)
DL = 7350 Mbps

Exynos Modem Integrated

(LTE Category 24/18)
DL = 3000 Mbps
8x20MHz AC, 1024-QAM
UL = 422 Mbps
4×20 MHz AC, 256-QAM

(5G NR-6 NR)
DL = 5100 Mbps
UL = 1920Mbps

(5G NR mmWave)
DL = 7350 Mbps
UL = 3670 Mbps

Mfc. Process Samsung
7nm (7LPP)
Samsung
5nm (5LPE)

On the Samsung LSI side of things, we find the new Exynos 2100. Unlike the more incremental changes of the Snapdragon 888 in terms of SoC design, the new Exynos is a major change for Samsung’s SoC division, as this is the first major design in many years ago that no longer uses Samsung’s own CPU microarchitecture, but has returned to using Arm Cortex cores, which in this case are also the new Cortex-X1 and Cortex-A78 cores.

At a high level, the Exynos 2100’s CPU configuration looks almost identical to that of the Snapdragon 888, as both are 1 + 3 + 4 designs with X1, A78 and A55 cores. The differences are in the details:

The X1 cores in the Exynos 2100 have a slightly higher frequency at up to 2.91 GHz, while the Cortex-A78 has a significantly higher frequency than the Snapdragon, as they reach 2.81 GHz. The Cortex-A55 cores are also quite aggressive in terms of frequency, as they now reach 2.20 GHz – therefore, in general, higher clocks than the Snapdragon variant.

But Exynos is not so aggressive with cache settings. Most importantly, the X1 cores here have only 512 KB of L2 cache, which is a bit odd given the new CPU’s overall performance philosophy. Cortex-A78s also see the use of 512KB L2 caches, while small A55 cores feature 64KB L2 – less than Snapdragon counterparts.

Much like the Snapdragon, the L3 cache also drops to 4 MB, instead of the 8 MB we expected for this generation, however, Samsung surprises us with the use of an estimated system level cache of 6 to 8 MB, compared to the 2 MB design on the Exynos 990.

On the GPU side of things, we see a Mali-G78MP14 at up to 854 MHz. That’s 27% more cores and 6.7% more frequency, and the company is also showing massive performance gains by announcing a 40% generation improvement. .

Let them fight

In today’s article, we’ll focus mainly on CPU and GPU performance, as an especially interesting comparison will be to see how the two projects compare, given that both now use Arm’s newest Cortex-X1 cores and both are sporty the same manufacturing node.

GPU comparisons will also be interesting – and perhaps quite controversial, as the results will not be what many people expected.

Although we would have liked to show the AI ​​performance of the two SoCs – unfortunately, the software situation on the Galaxy S21 now means that none of the SoCs are taking full advantage of their new accelerators, so this is a topic to be reviewed in a few months. the appropriate structures have been updated by Samsung.

Source