Moving to TSMC for More Speed, Lower Power

Tonight, as the darkness of the night comes to China, Qualcomm is hosting a mobile-focused product launch event called “Snapdragon Night.” The protagonist of the event is the announcement of the company’s new flagship SoC, the Snapdragon 8 + Gen1. An intermediate generation update to the flagship smartphone SoC, Snapdragon 8 Gen 1, 8 + Gen 1 follows the tradition of releasing Qualcomm’s annual refreshment. A product that improves performance and offers partners something new to collaborate in the second half of this year. Especially this year, we are looking at the very notable changes in chips from Qualcomm.

There’s a more important story in 2022, unlike previous generations where Qualcomm only launched faster bins of existing silicon. Qualcomm has completely switched foundries and moved from Samsung to TSMC, resulting in new dies. Thanks to this, Snapdragon 8+ Gen 1 Qualcomm can enjoy one-off manufacturing benefits, improve both CPU and GPU performance, while reducing power consumption.

The new Snapdragon 8 + Gen 1 is basically the original Snapdragon 8 Gen 1 ported from Samsung’s 4nm line to one of TSMC’s 4nm lines. Under more normal circumstances, this kind of shift can be unobtrusive-or at best, an interesting exercise to look for edge cases-but for Qualcomm’s flagship SoCs, the problem is even more important.

There are few official sources and statements about the quality of Samsung’s 4nm process, but informally it has become clear that Samsung’s 4nm process is not meeting expectations. This had a chain effect on the chips created at the process node, the original Snapdragon 8 Gen 1 made it more compatible with power consumption, and Samsung’s own Exynos 2200 didn’t work anymore. Conversely, for all accounts, TSMC’s N4 process looks great, and the optically scaled node is built from TSMC’s already successful and highly high-performance 5nm technology.

As a result of this performance gap between Samsung and TSMC’s 4nm node, Qualcomm is taking the unusual step of porting (essentially) high-end SoCs to TSMC’s fabs. Qualcomm is gaining momentum and 8 Gen 1 is selling well, although it’s not strictly necessary, but it’s certainly a wise move for the company. Qualcomm faces a particularly fierce competition of this generation with MediaTek, whose flagship-level Dimensity 9000 SoC was TSMC’s flagship product for 4nm nodes. And that leaves MediaTek with a clear advantage over the original 8 Gen 1. It would be very pleased to disable Qualcomm.

Eventually, by switching fabs, Qualcomm will have the opportunity to improve the original 8 Gen 1 from both ends of the spectrum, becoming Snapdragon 8+ Gen 1. In terms of performance, TSMC nodes can easily increase CPU. GPU clock speed to improve performance. Currently, the Prime Cortex-X2 core is clocked 7% higher at 3.2GHz, but in the A710 and A510 clusters, the clock speed has been significantly improved, each increasing by about 12%. Now even the slowest A510 core can run at 2GHz. The GPU clock speed has improved as well, and Qualcomm has not disclosed a specific clock speed, but has confirmed that the 8 + Gen1 Adreno GPU block runs at a clock that is 10% higher than the original 8Gen1. increase.

However, if anything, most of Qualcomm’s profits from manufacturing node switches are invested in reducing power consumption. One of the problems with 8 Gen 1 is that TSMC’s better 4nm process means that Qualcomm has significantly reduced overall SoC power consumption at equal frequencies.

Officially, Qualcomm claims that both GPU and CPU power efficiency will be improved by 30%. As mentioned earlier, this is equal frequency and does not take into account the higher peak clock speeds of 8 + Gen 1. Therefore, the actual power savings are not that great during peak hours. Although peak-based, power savings are still important, according to Qualcomm. Overall, the company advertises a 15% reduction in SoC power usage compared to the original 8 Gen 1 under a “practical usage pattern”. This should improve the battery life of mobile phones that adopt the new SoC.

Beyond that, the official specifications for 8 + Gen 1 do not reflect significant changes to the SoC’s configuration for the original chip. Therefore, we continue to consider the same integrated X65 5G modem, the same Spectra ISP, and the same video encode / decode block (sorry, gangsters, no AV1 support yet!). So, despite assembling a new die for this year’s Mesozoic refreshment product, there are no new features to talk about in 8 + Gen1.

When it comes to SoC performance, Qualcomm officially claims that the aforementioned clock speed improvements have improved GPU and CPU performance by 10%. Although Qualcomm was unable to attend the benchmark session last week, the performance mode figures released by Qualcomm are in close agreement with these claims. Qualcomm’s Geekbench 5 results are a few percent higher than benchmarked at the December 8 Gen 1 launch event, but it’s worth noting that PCMark didn’t achieve significantly higher scores. GPU performance numbers are mixed as well, and some of Qualcomm’s official results are very close to the original 8 Gen 1 results, but the difference between peak and persistent tests is in the GPU results. I don’t want to read too much, considering how much it affects. As always, the last word has to come down to an independent third-party test, but Qualcomm is unreasonable given the thermal headroom gained by increasing clock speeds and switching to TSMC. There is nothing to claim.

Finally, for consumers, the Snapdragon 8+ Gen1 device will be available for the first time in the third quarter of this year. According to Qualcomm, many of the usual suspects, such as Asus, Motorola, OnePlus, Honor, and Xiaomi, are signing on to release new SoC-based phones.