January 19, 2022

The Snapdragon 8 Gen 1 Efficiency Preview: Sizing Up Cortex-X2

On the current Qualcomm Snapdragon Tech Summit, the corporate introduced its new flagship smartphone processor, the Snapdragon 8 Gen 1. Changing the Snapdragon 888, this new chip is about to be in quite a few excessive efficiency flagship smartphones in 2022. The brand new chip is Qualcomm’s first to make use of Arm v9 CPU cores in addition to Samsung’s 4nm course of node expertise. Upfront of gadgets coming in Q1, we attended a benchmarking session utilizing Qualcomm’s reference design, and had a few hours to run assessments centered on the brand new efficiency core, based mostly on Arm’s Cortex-X2 core IP.

The Snapdragon 8 Gen 1

Relatively than proceed with the 800 naming scheme, Qualcomm is renaming its smartphone processor portfolio to make it simpler to grasp / market to customers. The Snapdragon 8 Gen 1 (hereafter known as S8g1 or 8g1) would be the headliner for the portfolio, and we anticipate Qualcomm to announce different processors within the household as we transfer into 2022. The S8g1 makes use of the newest vary of Arm core IP, together with up to date Adreno, Hexagon, and connectivity IP together with an built-in X65 modem able to each mmWave and Sub 6 GHz for a worldwide answer in a single chip.

Whereas Qualcomm hasn’t given any further perception into the Adreno / graphics a part of the {hardware}, not even giving us a 3-digit identifier, we’ve got been advised that it’s a new floor up design. Qualcomm has additionally advised us that the brand new GPU household is designed to look similar to earlier Adreno GPU sfrom a function/API standpoint, which signifies that for present video games and different apps, it ought to enable a easy transition with higher efficiency.  We had time to run just a few conventional gaming assessments on this piece.

On the DSP facet, Qualcomm’s headlines are that the chip can course of 3.2 Gigapixels/sec for the cameras with an 18-bit pipeline, appropriate for a single 200MP digital camera, 64MP burst seize, or 8K HDR video. The encode/decode engines enable for 8K30 or 4K120 10-bit H.265 encode, in addition to 720p960 infinite recording. There isn’t a AV1 decode engine on this chip, with Qualcomm’s VPs stating that the timing for his or her IP block didn’t synchronize with this chip.

Qualcomm’s Alex Katouzian

AI inference efficiency has additionally quadrupled – 2x from structure updates and 2x from software program. We have now a few AI assessments on this piece.

As normal with these benchmarking periods, we’re very occupied with what the CPU a part of the chip can do. The brand new S8g1 from Qualcomm includes a 1+3+4 configuration, just like the Snapdragon S888, however utilizing Arm’s latest v9 structure cores.

  1. The one huge core is a Cortex-X2, working at 3.0 GHz with 1 MiB of personal L2 cache.
  2. The center cores are Cortex-A710, working at 2.5 GHz with 512 KiB of personal L2 cache.
  3. The 4 effectivity cores are Cortex-A510, working at 1.8 GHz and an unknown quantity of L2 cache. These 4 cores are organized in pairs, with L2 cache being personal to a pair.
  4. On the highest of those cores is an extra 6 MiB of shared L3 cache and 4 MiB of system degree cache on the reminiscence controller, which is a 64-bit LPDDR5-3200 interface for 51.2 GB/s theoretical peak bandwidth.

In comparison with the Snapdragon S888, the X2 is clocked larger than the X1 by round 5% and has further architectural enhancements on prime of that. Qualcomm is claiming +20% efficiency or +30% energy effectivity for the brand new X2 core over X1, and on that final level it’s past the +16% energy effectivity quoted by Samsung shifting from 5nm to 4nm, so there are further efficiencies Qualcomm is implementing in silicon to get that quantity. Sadly Qualcomm wouldn’t go into element what these are, nor present particulars about how the voltage rails are separated, if this is similar as S888 or completely different – Arm has said that the X2 core might provide lowered energy than the X1, and if the X2 is by itself voltage rail that might present assist for Qualcomm’s claims.

The center A710 cores are additionally Arm v9, with an 80 MHz bump over the earlier technology seemingly offered by course of node enhancements. The smaller A510 effectivity cores are constructed as two complexes every of two cores, with a shared L2 cache in every advanced. This format is supposed to offer higher space effectivity, though Qualcomm didn’t clarify how a lot L2 cache is in every advanced – usually they do, however for no matter purpose on this technology it wasn’t detailed. We didn’t probe the quantity in our testing right here as a result of restricted time, however little doubt when gadgets come to market we’ll discover out.

On prime of the cores is a 6 MiB L3 cache as a part of the DSU, and a 4 MiB system cache with the reminiscence controllers. Like final 12 months, the cores should not have direct entry to this 4 MiB cache. We’ve seen Qualcomm’s principal high-end competitor for subsequent 12 months, MediaTek, showcase that L3+system cache will probably be 14 MiB, with cores gaining access to all, so it is going to be attention-grabbing to see how the 2 examine when we’ve got the MTK chip to check.

Benchmarking Session: How It Works

For our benchmarking session, we got a ‘Qualcomm Reference Gadget’ (QRD) – that is what Qualcomm builds to point out a illustration of how a flagship that includes the processor would possibly look. It seems similar to trendy smartphones, with the objective to reflect one thing that may come to market in each software program and {hardware}. The software program half is necessary, because the companion gadgets are seemingly a few months from launch, and so we acknowledge that not the whole lot is closing right here. These gadgets additionally are usually thermally just like a future retail instance, and it’s fairly apparent if there was one thing odd within the thermals as we check.

These benchmark periods often contain 20-40 press, every with a tool, for 2-4 hours as wanted. Qualcomm preloads the system with quite a few widespread benchmarking functions, in addition to a knowledge sheet of the outcomes they need to anticipate. Any member of the press that desires to sideload any new functions has to no less than ask one of many reps or engineers within the room. In our conventional workflow, we sideload energy monitoring instruments and SPEC2017, together with our different microarchitecture assessments. Qualcomm by no means has any subject with us utilizing these.

As with earlier QRD testing, there are two efficiency presets on the system – a baseline preset anticipated to showcase regular operation, and a excessive efficiency preset that opportunistically places threads onto the X2 core even when energy and thermals is sort of excessive, giving one of the best rating regardless. The talk in smartphone benchmarking of preliminary runs vs. sustained efficiency is a protracted one which we gained’t go into right here (most noticeably as a result of 4 hours is simply too quick to do any intensive sustained testing) nevertheless the efficiency mode is supposed to allow a ‘first run’ rating each time.

Leave a Reply

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