Exynos-powered Galaxy S22 Ultra performs significantly worse than Qualcomm variant

Humza

Posts: 1,026   +171
Staff member
What just happened? Samsung was quite adamant that its Exynos 2200 premium mobile platform, based on AMD’s RDNA 2 architecture, will provide the finest experience for smartphone users. However, a reviewer based in Europe has found Samsung’s SoC to be noticeably worse in terms of GPU performance, at least when it comes to benchmarks.

Traditionally, the Exynos vs. Qualcomm debate has always concluded in favor of the latter, as phones, including Samsung’s flagship Galaxy models, have been known to provide superior performance if they’re running a Snapdragon chip inside.

It seemed, however, that Samsung’s new flagship Exynos 2200 SoC announced earlier this year would close that gap, bringing with it unheard features to the mobile space like hardware-accelerated ray tracing and variable rate shading. While we’re yet to see the latter features appear on a mobile game, it looks like Galaxy buyers still need to be aware of which SoC variant they can get their hands on.

As reported by TechPowerUp, Erdi Özüağ, a reviewer based in Turkey, tested both variants of the Galaxy S22 Ultra and found that the European-focused Exynos-powered variant (Xclipse 920 GPU) had a 42 percent performance difference over the Qualcomm variant (Adreno 730 GPU) in the 3D Mark Wild Life benchmark.

This GPU stress test and its extreme version are meant to measure device performance for short bursts of time, mirroring the experience of an intense mobile game. While casual users may not notice the difference in day-to-day tasks and usage, having to compromise gaming/graphical performance in such an expensive flagship device is undesirable nevertheless.

The reviewer also came across a similar conclusion with the GFXBench Manhattan test, where the Exynos-powered S22 Ultra was 30 percent slower than the SD 8 Gen 1 chip. Suspecting the former iGPU’s very low 555MHz clock speed, apparently rated to be 1.3GHz, Özüağ expects Samsung might roll out a software fix to narrow this performance gap.

Although a future software patch may address the issue, at least to some extent, the safer bet for now is to ensure your $1,200 Galaxy S22 Ultra has the Qualcomm SoC.

Permalink to story.

 
People are speculating it's Samsung's failed 4nm process to blame because early results showed incredible performance. They've had to step down clock speed again and again because of poor yields, apparently.
Get it together Sammy!
 
This GPU stress test and its extreme version are meant to measure device performance for short bursts of time

And thats were these benchmarks fails, since they end before the SOC is throtled down.

While casual users may not notice the difference in day-to-day tasks and usage

Which is what really matters.


having to compromise gaming/graphical performance in such an expensive flagship device is undesirable nevertheless.

Aah, another product killed by the desires of a writer...
 
I find it highly amusing that Adreno is faster than Radeon. But of course Samsung were more interested in the AMD branding than the actual technology itself. Same for all Samsung flagships, it’s just a con. They market the balls out of what is just an average Android device, put an iPhone price on it then 12-18 months later they can it’s support and the resell value drops through the floor.

 
When mobile GPU makers say nowadays the power is comparable to consoles, it is true.
They really mean that the power is there. Is crazy that PC gpu's might actually lag on the progress timeline compared to mobile advances.
This time AMD is pretty far on rasterization without silicon sliced for a portion of RT, and the power is cooler.
 
People are speculating it's Samsung's failed 4nm process to blame because early results showed incredible performance. They've had to step down clock speed again and again because of poor yields, apparently.
Get it together Sammy!
At this point, I think it is no longer a speculation. If you look at last gen's Qualcomm SN 888 and Samsung Exynos 2100, the problem with throttling is very evident on Samsung's 5nm. In fact, if we look at performance improvement from SN 865 to 888, the difference sounds great on paper, but in practice, it is not that great. This year's 4nm is basically just a rebrand and slightly refined 5nm, which is not going to have a lot of legroom for aggressive clockspeed or hardware improvements. And most reviews of Qualcomm's new chip also shows significant throttling that makes it just slightly better than the previous SN 888.

Having said that, I believe part of the issue is also due to the fact that RDNA2 is not really made to scale down to such a low power SOC. At least if we look at their desktop GPUs, all of them perform optimally at very high clockspeed, north of 2 Ghz. Other than the top end RX 6800 and 6900, every subsequent card released for mid and low end tend to try to claw back performance by focusing purely on clockspeed to make up for hefty cuts in specs. But to allow for high clockspeed means having to pay a power penalty (also resulting in higher heat output). So coupled with the less than optimal Samsung node, I am not surprised that performance is bad.
 
I knew those Qualcomm GPUs were no slouch, but amusing that it's outrunning a Radeon. And a good sign for when I get an ARM-based notebook eventually, that I will have good enough GPU performance to run some 3D games on there if I want (that won't be for a while, I just got a nice system with a Ryzen in the last 6 months.) I can tell you running Ubuntu on ARM was totally drama-free (I ran it on a Tegra K1-based ARM Chromebook a while back). All the usual software is ARM-native, and the couple items I ran that weren't native ran seamlessly with qemu (both 32-bit x86 and 64-bit x86-64 stuff worked this way.)
 
Back