At this 12 months’s Tech Summit from Hawaii, it’s time once more for Qualcomm to unveil and element the corporate’s most vital launch of the 12 months, showcasing the latest Snapdragon flagship SoC that will probably be powering our upcoming 2022 units. Right now, as the primary of some bulletins on the occasion, Qualcomm is asserting the brand new Snapdragon 8 Gen 1, the direct follow-up to final 12 months’s Snapdragon 888.
The Snapdragon 8 Gen 1 follows up its predecessors with a really apparent change in advertising and product naming, as the corporate is making an attempt to simplify its product naming and line-up. Nonetheless a part of the “8 sequence”, that means the best finish phase for units, the 8 Gen 1 resets the earlier three-digit naming scheme in favor of only a phase and technology quantity. For Qualcomm’s flagship half that is fairly easy, nevertheless it stays to be seen what this implies for the 7 and 6 sequence, each of which have upwards of a number of elements for every technology.
As for the Snapdragon 8 Gen 1, the brand new chip comes with lots of new IP: We’re seeing the brand new trio of Armv9 Cortex CPU cores from Arm, an entire new next-generation Adreno GPU, a massively improved imaging pipeline with plenty of new options, an upgraded Hexagon NPU/DSP, built-in X65 5G modem, and all manufactured on a more moderen Samsung 4nm course of node.
The brand new chip guarantees giant will increase in efficiency and effectivity in lots of the processing parts, in addition to new options enabling new person experiences. Let’s begin over the fundamental specs and drill down the small print that now we have on the chip:
|Qualcomm Snapdragon Flagship SoCs 2020-2021|
Snapdragon 8 Gen 1
@ 3.0GHz 1x1024KB pL2
@ 2.84GHz 1x1024KB pL2
|GPU||Adreno next-gen||Adreno 660 @ 840MHz|
|DSP / NPU||Hexagon||Hexagon 780
26 TOPS AI
|4x 16-bit CH
@ 3200MHz LPDDR5 / 51.2GB/s
4MB system stage cache
|ISP/Digital camera||Triple 18-bit Spectra ISP
1x 200MP or 108MP with ZSL
8K HDR video & 64MP burst seize
|Triple 14-bit Spectra 580 ISP
1x 200MP or 84MP with ZSL
4K video & 64MP burst seize
|8K30 / 4K120 10-bit H.265
Dolby Imaginative and prescient, HDR10+, HDR10, HLG
720p960 infinite recording
|Built-in Modem||X65 built-in
(5G NR Sub-6 + mmWave)
(5G NR Sub-6 + mmWave)
|Mfc. Course of||Samsung
CPUs: Cortex-X2 and Armv9 siblings
Beginning off with the CPUs of the brand new Snapdragon 8 Gen 1 (I’ll shorthand it as S8g1 right here and there): That is Qualcomm’s first chip that includes the brand new Armv9 technology of CPU IPs from Arm, which incorporates the Cortex-X2, Cortex-A710, and Cortex-A510 in an enormous, center, and little setup. Qualcomm continues to make use of a 1+3+4 core rely, a setup that’s been comparatively profitable for the designers over the previous few years and iterations ever for the reason that Snapdragon 855.
The Cortex-X2 core of the brand new chip clocks in at 3.0GHz, which is a tad greater than the two.84GHz clock of the X1 core on the Snapdragon 888. This was truly a bit stunning to me, as I hadn’t anticipated a lot in the best way of clock will increase this technology, nevertheless it’s good to see Arm distributors now routinely attaining this. For context, MediaTek’s lately introduced Dimensity 9000 achieves 3.05GHz on its X2 core, nevertheless that’s on a TSMC N4 node. In distinction, Qualcomm manufactures the Snapdragon 8 Gen 1 on a Samsung 4nm node. The corporate wouldn’t verify if it’s a 4LPE variant or one thing extra customized, therefore why we’re leaving it as a “4nm” node description within the specification desk.
What’s most stunning concerning the X2 core is that Qualcomm is claiming 20% sooner efficiency or 30% energy financial savings, the latter determine being particularly intriguing. Samsung Foundry solely describe a 16% discount in energy in going from a 5nm to 4nm node, and clearly 30% is considerably higher than what the method node guarantees. We requested Qualcomm what sort of enhancements result in such a big energy lower; nevertheless, the corporate wouldn’t specify any particulars. I notably requested if the brand new X2 cores have their very own voltage area (Earlier Snapdragon 1+3 large+center implementations shared the identical voltage rail), however the firm wouldn’t even verify if this was the case or not. Arm had famous that the X2 can have quite lower power at the same peak performance point of the X1, if Qualcomm’s advertising supplies discuss with such a comparability, then the numbers may make sense.
The X2 core is configured with 1MB of L2 cache, whereas the three Cortex-X710 cores have 512KB every. The center cores listed below are clocked barely greater at 2.5GHz this 12 months, somewhat 80MHz soar over the earlier technology. Normally, the center cores pay extra consideration to the ability funds, so perhaps this barely enhance does symbolize extra precisely the method node enhancements.
Lastly, the brand new chip additionally makes use of 4 Cortex-A510 cores at 1.8GHz. In contrast to the Dimensity 9000 from a few weeks again, Qualcomm does make use of Arm’s new “merged-core” method of the brand new microarchitecture, that means that the chip truly has two Cortex-A510 complexes with two cores every, sharing a typical NEON/SIMD pipeline and L2 cache. The merged core method is supposed to realize higher space effectivity. Qualcomm rationalized the method by saying that in on a regular basis use instances with fewer threads energetic and general low exercise, having a single core in a position to entry a bigger L2 cache shared by two cores can lead to higher efficiency and effectivity. Sadly even whereas making this remark, the corporate wouldn’t truly element what the L2 measurement was, whether or not it’s 512KB or 256KB – if it’s the latter, then the configuration positively isn’t as aggressive because the Dimensity 9000.
The brand new Armv9 CPU IPs from Arm additionally got here with a brand new technology DSU (DynamiQ Shared Unit, the cluster IP) which the brand new Snapdragon makes use of. Qualcomm right here opted for a 6MB L3 cache measurement, noting that this was a call in balancing out system efficiency throughout goal workloads.
As for system caches, Qualcomm talked about that the chip stays unchanged with a 4MB cache, and the reminiscence controllers are nonetheless 3200MHz LPDDR5 (4x 16bit channels). It’s to be famous that, as with final 12 months’s Snapdragon 888, the CPUs not have entry to the system cache, with the intention to enhance DRAM latency. We will’t assist however make comparisons to MediaTek’s Dimensity 9000, which doubtless may have worse DRAM latency, but additionally supply as much as 14MB of shared caches to the CPUs versus simply 6MB on the Snapdragon 8 Gen 1. How the 2 chips will evaluate to one another stays to be seen in precise industrial units.
GPU: New Adreno structure with no title
Again within the day, Qualcomm’s Adreno GPU architectures had been simple to establish by way of their household in addition to efficiency ranges. Notably on the structure aspect, the Adreno 600 sequence began off with the Adreno 630 within the Snapdragon 845 a couple of years in the past, however not like in earlier iterations from the 400- and 500 sequence, we remained with that high-level description up till the Snapdragon 888 sequence.
The Snapdragon 8 Gen 1 right here adjustments issues, and albeit, Qualcomm did a fairly horrible job at advertising what they’ve this time round. The brand new GPU title utterly drops any mannequin quantity, and as such doesn’t instantly reveal that it’s half of a bigger microarchitecture shift that previously would have been marketed as a brand new Adreno sequence.
Qualcomm notes that from a particularly high-level perspective, the brand new GPU may look just like the earlier generations, nevertheless there are giant architectural adjustments included that are supposed to enhance efficiency and effectivity. Qualcomm gave examples comparable to concurrent processing optimizations that are supposed to give giant boosts in efficiency to real-world workloads which may circuitously present up in benchmarks. One other instance was that the GPU’s “GMEM” noticed giant adjustments this technology, comparable to a rise of 33% of the cache (to 4MB), and now being each a learn & write cache relatively than only a writeback cache for DRAM visitors optimizations.
The high-level efficiency claims are 30% sooner peak efficiency, or 25% energy discount on the identical efficiency because the Snapdragon 888. Qualcomm additionally uncharacteristically commented on the scenario of peak energy figures and the present scenario available in the market. Final 12 months, Qualcomm rationalized the Snapdragon 888’s excessive peak GPU energy figures by noting that that is what distributors had demanded in response to what we noticed from different gamers, notably Apple, and that distributors would be capable to obtain higher thermal envelopes of their units. Arguably, this technique ended up as being fairly disastrous and adverse by way of notion for Qualcomm, and I really feel that on this 12 months’s briefing we noticed Quaclomm try and distance themselves extra from the scenario, largely by outright saying that the one level of such peak efficiency and energy figures is for distributors to realize greater first-run benchmarking numbers.
Sadly, not like Apple, who truly use their GPU’s peak efficiency figures in transient compute workloads comparable to digicam processing, at present the Android ecosystem simply doesn’t make any superior use of GPU compute. This admission was truly a breath of recent air and perception into the scenario, because it’s been one thing I’ve particularly famous in our Kirin 9000, Snapdragon 888 and Exynos 2100 and Tensor deep-dives in criticizing all the brand new chips. It’s an extremely silly scenario that, so long as the media continues to place weight on peak efficiency figures, received’t be resolved any time quickly, because the chip distributors may have a tough time saying no to their buyer’s requests to function the silicon on this approach.
Qualcomm states that one option to attempt to alleviate this new deal with peak efficiency is to alter the best way the GPU efficiency and energy curve behaves. The workforce said that they’ve gone in to alter the structure to attempt to flatten the curve, to not solely obtain these arguably mindless peak figures, however truly deal with making bigger enhancements within the 3-5W energy vary, a spread the place the Snapdragon 888 final 12 months didn’t considerably enhance upon the Snapdragon 865.
That being mentioned, even with a 25% lower in energy at comparable Snapdragon 888 efficiency, the brand new Snapdragon 8 Gen 1 doubtless nonetheless received’t be capable to compete in opposition to Apple’s A14 or A15 chips. MediaTek’s Dimensity 9000 additionally must also be notably extra environment friendly than the brand new Snapdragon at equal efficiency ranges given the claimed effectivity figures, so it nonetheless appears to be like like Qualcomm’s alternative of going with a Samsung course of node, even this new 4nm one, received’t shut the hole to the TSMC rivals.