Exynos 2100 vs Snapdragon 888: Samsung Galaxy S21 Ultra sustained performance and throttling test

Our in-depth look at the differences between the US and International versions of the Samsung Galaxy S21 Ultra is slowly turning into a regular editorial series.

Today, in yet another edition of the Qualcomm Snapdragon 888 vs. Exynos 2100, we are taking a closer look at sustained chipset performance and thermal throttling.

As usual – for context – Samsung has been releasing two versions of its premium smartphones – one with a Qualcomm chipset and another one with its own Exynos chips. So, as usual, we dig in to find any potential differences between the two versions. In other words, is one more awesome than the other?

Chipset Samsung Exynos 2100 Qualcomm Snapdragon 888
CPU 1x 2.9 GHz – Cortex-X1, 3x 2.8 GHz – Cortex-A78, 4x 2.2 GHz – Cortex-A55 1x 2.84 GHz – Kryo 680 Prime (Cortex-X1), 3x 2.42 GHz – Kryo 680 Gold (Cortex-A78), 4x 1.8 GHz – Kryo 680 Silver (Cortex-A55)
Manufacturing node 5nm EUV 5nm
GPU Mali-G78 MP14 (14 cores) (Vulkan 1.1, OpenCL 2.0, DirectX 12) Adreno 660 (Vulkan 1.1, OpenCL 2.0, DirectX 12)
Memory support LPDDR5, 3200 MHz, max 50 Gbit/s, max size 16GB LPDDR5, 3200 MHz, max 50 Gbit/s, max size 16GB
Storage UFS 3.1 UFS 3.0, UFS 3.1
Multimedia support Max display res: 4096 x 2160, Video capture: 8K@30fps, 4K@120fps; playbak: 8K@60fps Max display res: 3840 x 2160, Video capture: 8K@30fps, 4K@120fps; playbak: 8K@60fps
Neural processor (NPU) Triple-core NPU (up to 26 TOPS) Hexagon 780 26 TOPS
Modem LTE Cat. 24, up to 3000 Mbps down, 422 Mbps up; 5G SA/NSA/Sub6/mmWave, up to 7.35 Gbps down, 3.67 Gbps up LTE Cat. 22, up to 2500 Mbps down, 316 Mbps up; 5G SA/NSA/Sub6/mmWave, up to 7.5 Gbps down, 3 Gbps up

After initially exploring the difference in benchmark performance between the two chips inside an otherwise identical Galaxy S21 Ultra unit, we then moved on to compare overall battery endurance. In both instances, the data was pretty consistent and led to the conclusion that the two S21 Ultra variants perform very similarly.

Of course, this was never going to be the full story. There is always plenty of finer detail to analyze. Even so, most of your really passionate comments on the matter kept circling back to the topics of thermal throttling and sustained performance in the ongoing debate on which is the superior chip. So much so, in fact, that we thought it merits some additional research on our part.

Exynos vs. Snapdragon S21 Ultra - sustained performance review

Just like before, we took our US Snapdragon 888 S21 Ultra unit and the EU Exynos 2100 model that we used for our in-depth review, and we ran the exact same set of tests on them, with the exact same resolution settings – QHD and adaptive refresh rate set to Auto.

CPU Throttling test on the Exynos 2100 Galaxy S21 Ultra - Exynos vs. Snapdragon S21 Ultra - sustained performance review
CPU Throttling test on the Exynos 2100 Galaxy S21 Ultra - Exynos vs. Snapdragon S21 Ultra - sustained performance review
CPU Throttling test on the Exynos 2100 Galaxy S21 Ultra - Exynos vs. Snapdragon S21 Ultra - sustained performance review
CPU Throttling test on the Exynos 2100 Galaxy S21 Ultra - Exynos vs. Snapdragon S21 Ultra - sustained performance review
CPU Throttling test on the Exynos 2100 Galaxy S21 Ultra - Exynos vs. Snapdragon S21 Ultra - sustained performance review
CPU Throttling test on the Exynos 2100 Galaxy S21 Ultra - Exynos vs. Snapdragon S21 Ultra - sustained performance review

CPU Throttling test on the Exynos 2100 Galaxy S21 Ultra

CPU Throttling test

CPU Throttling test is a great app to kick things off with. A few important notes are in order, though. What we are trying to achieve here is put the highest possible strain on the chipset. This is best achieved by hitting it with a lengthy, synthetic multi-core load. We have looked into the matter in-depth in the past and CPU cores remain the single most significant heat source in a chipset, with the GPU hardly contributing in comparison. Since sustained performance in a passively-cooled smartphone system depends pretty-much entirely on thermal management, heat build-up and thermal headroom, we need to torture the CPU to get any usable data. It is important to note, however, that this is not a realistic real-world load. Even the most challenging, most demanding Android games around won’t be hitting all of the CPU cores all at once. Hence, the numbers on the graphs should be taken as absolute worst-case scenarios.

CPU Throttling test on the Snapdragon 888 Galaxy S21 Ultra - Exynos vs. Snapdragon S21 Ultra - sustained performance review
CPU Throttling test on the Snapdragon 888 Galaxy S21 Ultra - Exynos vs. Snapdragon S21 Ultra - sustained performance review
CPU Throttling test on the Snapdragon 888 Galaxy S21 Ultra - Exynos vs. Snapdragon S21 Ultra - sustained performance review
CPU Throttling test on the Snapdragon 888 Galaxy S21 Ultra - Exynos vs. Snapdragon S21 Ultra - sustained performance review
CPU Throttling test on the Snapdragon 888 Galaxy S21 Ultra - Exynos vs. Snapdragon S21 Ultra - sustained performance review
CPU Throttling test on the Snapdragon 888 Galaxy S21 Ultra - Exynos vs. Snapdragon S21 Ultra - sustained performance review

CPU Throttling test on the Snapdragon 888 Galaxy S21 Ultra

Let’s sum-up the numbers seen during the duration of the test, as seen in the screenshots above.

For the Snapdragon 888 unit, we got max, average and min performance figures of 250,353 / 210,296 / 192,271 respectively, which works out to 77% retained performance over the test duration of one hour.

CPU Throttling test scores MAX MIN AVERAGE RETAINED PERFORMANCE AFTER 60 MIN
Galaxy S21 Ultra (Snapdragon 888) 250,353 210,296 192,271 77%
Galaxy S21 Ultra (Exynos 2100) 244,339 195,471 137,807 56%

Conversely, the Exynos 2100 managed: 244,339 / 195,471 / 137,807, which works out to 56% retained performance. That’s a substantial difference.

The Snapdragon 888 inside the Galaxy S21 Ultra maintains a higher and more stable sustained CPU performance under prolonged synthetic max load by the CPU Throttling test, compared to the Exynos 2100 model.

Looking at the graphs themselves, we can’t help but notice that the two variants seem to take a slightly different approach to their performance curve, likely down to some CPU governor tuning. The Snapdragon 888 unit appears to be trying to sustain higher average performance for as long as possible, followed by a noticeable, steep drop-off point. In comparison, the Exynos 2100 seems to ramp-down much more-gracefully and gradually. Typically, we would prefer the latter since it is less likely to cause sudden stutters. However, there is also another way of looking at this – the Exynos chip can sustain its peak performance for far shorter periods and is forced to start ramping-down due to thermal constraints earlier. Also, that sudden dip, marked with red on the chart, followed by a sudden ramp back up is the worst-case scenario for stutters.

Overall, as far as the CPU Throttling test app is concerned, it seems like the Exynos is struggling quite a bit more than its Snapdragon counterpart while also maintaining slightly lower scores overall. We would not read too much into the specific performance numbers in this specific case as both chips are comparable in short “bursty” loads. Those kinds of loads are far more likely to occur in real-life as well.

Consecutive Antutu runs

For our next set of tests, we wanted something more compound, so we turned to AnTuTu and elected to do 5 consecutive runs on each unit. We found that to be a good representation of the sustained behavior of each chip. Again, we wouldn’t fixate too much on the particular numbers since we know that the Snapdragon 888 and Exynos 2100 can score quite close to each other in AnTuTu.

A more important aspect when analyzing today’s test is the amount of performance the Exynos loses over time, compared to the much better-sustained curve on the Snapdragon. Qualcomm’s chip even started to settle in and regain some ground for the final run. Fascinating stuff.

Finally, we also wanted to do our best to stress the phones with a more GPU-centric workload, both out of curiosity and as a partial sanity check, to see if our ongoing theory that throttling is mostly CPU-bound still checks out.

Indeed, we ended up with two much-tighter lines, nearly identical in behavior. Just like in our earlier benchmark comparisons, we see that, at least initially, 3DMark’s Vulkan test appears to favor the Mali-G78 MP14 GPU inside the Exynos 2100 over the Adreno 660 inside the Snapdragon 888.

The tides then quickly turn in Qualcomm’s favor as heat builds up and thermal throttling settles in. In a theoretical parallel setup, where the Exynos has better CPU core thermal management, we wouldn’t be surprised to see the Mali-G78 MP14 with a notable, or at least measurable lead instead.

As things currently stand, though, in a more GPU-heavy workload situation, both the Snapdragon 888 and the Exynos 2100 versions of the Galaxy S21 Ultra remain very, very comparable in terms of performance.

Final thoughts

What’s the takeaway then? Well, at the end of the day, this entire debate is still a purely academic one, held out of curiosity, since in practical terms, we still maintain that the two chips, as found in the two variants of the Samsung Galaxy S21 Ultra, deliver a very comparable experience. Definitely much more so than we’ve seen with some of Samsung’s earlier flagship generations and particularly its own chipsets. If nothing else, we can probably all safely agree that the Exynos 2100 constitutes a massive and marked improvement over last year’s Exynos 990.

Exynos vs. Snapdragon S21 Ultra - sustained performance review

As for declaring one chip or the other the ultimate victor here, well, the score seems to be in Qualcomm’s favor overall. Especially if you plan on having extended gaming sessions on your Galaxy or just feel like calculating Pi in your spare time. Though once again, the practical gap in performance isn’t all that big, and you can confidently grab whichever variant is available in your country without any fear of missing out.

Leave a Reply

Your email address will not be published. Required fields are marked *