Premium Segment SoC gets Cortex-X CPU Core

After the 2021/2022 product cycle it was a little bit morenesting Than Qualcomm would have liked, 2023 has been a much clearer year for the prolific SoC and cellular modem supplier. After launching the first part of their Gen 2 family earlier this year with the flagship Snapdragon 8 Gen 2, the company is poised to iterate through the next step of its product family with the Snapdragon 7+ Gen 2. It targets what is now the “premium” market segment. Qualcomm’s traditional $400 to $600 flagship, which focuses on flagship-level features with more modest performance and costs, for the Snapdragon 7+ Gen 2, Qualcomm aims to provide a significant performance boost to the platform.

Positioned as the successor to last year’s Snapdragon 7 Gen 1, this year’s iteration of the Snapdragon 7 is, in general, focused more on improving performance than adding features. While the early part of last year added mmWave support and new CPU and GPU architectures — notably Armv9 CPU cores — this year there are only a few new features. Instead, it’s what Qualcomm is touting as one of the biggest performance boosters ever for the Snapdragon 7 family. This is enabled in large part by the much-welcome pivot from Samsung’s beleaguered 4nm process to TSMC’s 4nm process, reversing the switch Which Qualcomm made last year for its well-received mid-cycle Snapdragon 8+ Gen 1 segment.

Also new this year, Qualcomm is giving hints that this won’t be the only part of the Snapdragon 7 Gen 2 we see this year, against the decision to launch the first part of Gen 2 as version 7+ instead of 7. In short, the launch of the Snapdragon 7+ part leaves Qualcomm field to launch the vanilla Snapdragon 7 part later. Granted, Qualcomm isn’t explicitly announcing any such part now, but there’s little reason to launch the 7+ first unless they have plans for something below; Otherwise, they could have launched it as a 7-bit Snapdragon 7 Gen 1, which was always a single-chip stack.

In terms of CPU organization, the Snapdragon 7+ Gen 2 retains the same 1+3+4 CPU core configuration that we’ve seen for the past few generations of the Snapdragon 7 family. The big news here is that the better-performing Prime core gets a significant performance improvement. , where Qualcomm is switching from using a slightly higher mid-core to completely using the higher-performance CPU architecture.

So, for the first time ever for a Snapdragon 7 part, Qualcomm is exploiting one of Arm’s Cortex-X cores for the Prime core. The Cortex-X2 used here is technically Arm’s previous-generation design, so it won’t step on the toes of a Snapdragon 8 Gen 2 and Cortex-X3 core. But compared to the A710 core used in the 7th generation core core (and 7+ intermediate cores of the 2nd generation), the Cortex-X2 represents a significant improvement in both IPC and clock speed. As a result, the peak clock speed of the Prime core has moved from 2.4 GHz to 2.91 GHz, which further complicates the IPC gains of the more complex cores.

Finally, Qualcomm is calling for a CPU performance improvement of “up to” 50% for the 7+ Gen 2 compared to the 7th Gen 1; Almost all of this comes from the new Prime core.

The trade-off is that such a significant performance boost is only truly accessible for single-threaded workloads, since there is only one Cortex-X2 core. The three average (performance) cores are again based on Cortex-A710, and clocked at 2% higher than before. As such, the first generation 7+ will not see huge gains on highly multi-threaded workloads. The improved power efficiency of TSMC’s 4nm process should pay some dividends there, but some of those gains have been invested in making the power-hungry Cortex-X2 viable from a battery life perspective.

Meanwhile, the 7+ Gen 2 also includes a faster Adreno GPU. As has been the case with Qualcomm’s integrated GPUs for two generations now, the company doesn’t assign a product number to them — let alone reveal important architectural details — so there’s a limited amount of detail we can share. Based on the feature summary, this does not appear to use the newer 8 Gen 2 GPU architecture; So it looks like Qualcomm has integrated a larger version of its current GPU and that has certainly given it a nice clock boost.

Be that as it may, the GPU performance outlook for the new SoC is significant: Qualcomm boasts a massive two-fold performance improvement over the seventh generation of the first — a platform that delivered only 20% more than its predecessor. While they’re not top-tier SoCs, Qualcomm still likes to position the Snapdragon 7 series as a good match for gaming smartphones, especially in China, so it’s no surprise to see Qualcomm invest so much in GPU performance. .

Finally, Qualcomm advocates a 13% improvement in power efficiency over the 7th generation from the first, at least based on “extended daily use.” Switching to TSMC’s 4nm process should pay big dividends, as evidenced by last year’s 8+ Gen 1 segment, but at the same time it’s clear that Qualcomm has invested a significant portion of those gains into improving overall performance.

Feeding the Dragon is a 32-bit (double 16-bit) LPDDR5 memory controller. Unlike the Snapdragon 8 Gen 2, the 7+ Gen 2 doesn’t get support for faster LPDDR5X memory, which means the status quo prevails for the Snapdragon 7 family. In this case, that means support for memory speeds up to LPDDR5-6400, which works out to 25.6GB/s. seconds of memory bandwidth. In contrast to the significant increase in CPU and GPU performance, there will be a lot of pressure on Qualcomm’s system cache and submemory to keep the various processing blocks fed.

Speaking of which, it’s not just CPUs and GPUs that have seen huge performance increases. Qualcomm’s Hexagon DSP/AI engine block also received a significant performance boost, rivaling the 2x increase in GPU. Qualcomm was light on the technical details here, but in our briefing there was no mention of features like INT4 or micro-tiling – two key advantages of the next-generation Hexagon block over the 8 Gen 2 – so it seems likely that this is a significantly enhanced version of the Hexagon block used in The previous seventh generation.

One piece of Snapdragon 8 technology that makes its way to the Snapdragon 7, however, is the triple 18-bit Spectra ISP. Replacing the 14-bit module featured in previous generations of the platform, the 18-bit module in the 7+ Gen 2 will bring support for computational triple-exposure HDR video capture, as well as enhanced low-light photography, which Qualcomm calls a Mega Low Light feature. The end result is that the second generation of the 7+ can capture at a higher resolution when using the no shutter lag function, and combined with the updated GPU, it can now record 4K video at up to 60fps, doubling the 4K30 limit of the seventh generation. the first.

Finally, rounding out the package is a duplicate of Qualcomm’s Snapdragon X62 integrated modem. Like last year’s SoC, this is a 16 mmWave + Sub-6 design that can achieve a maximum theoretical download rate of 4.4Gbps. However, this year’s design comes with a twist: Dual-SIM Dual-Active (DSDA) support, a first for the Snapdragon 7 platform. Both active radios on the 7+ generation support 5G and 4G connections, allowing dual-SIM users to use any Network they basically want on either radio. This is another distinguishing feature, which until now was limited to Qualcomm’s Snapdragon 8 platform.

For non-cellular connectivity, the 7+ Gen 2 uses the FastConnect 6900 Radio System. This is a relatively modest update over the previous 6700 Radio, increasing Bluetooth support to version 5.3 of the protocol, and increasing the peak bandwidth of the Wi-Fi 6E streaming 2×2 radio to 3.6 Gbps thanks to concurrent dual band (DBS) support.

In short, Snapdragon 7+ Gen 2 will come to market very quickly. According to Qualcomm, phones using the SoC will be available later this month, with Redmi and Realme among the OEMs slated to launch around the new chip.