Snapdragon 8cx Gen 3 launched: First 5 nm Windows PC SoC with four Cortex-X1 cores at 3 GHz, 6 nm Snapdragon 7c+ Gen 3 announced too

[Redaktion]

Qualcomm today unveiled the Snapdragon 8cx Gen 3 and Snapdragon 7c+ Gen 3 Compute Platforms that will power the next wave of Windows-on-ARM Always On Always Connected PCs. The Snapdragon 8cx Gen 3 is the first 5 nm SoC for the PC and features a 4+4 CPU with Cortex-X1 and Cortex-A78 cores along with other Qualcomm connectivity and security features.

https://www.notebookcheck.net/Snapdragon-8cx-Gen-3-launched-First-5-nm-Windows-PC-SoC-with-four-Cortex-X1-cores-at-3-GHz-6-nm-Snapdragon-7c-Gen-3-announced-too.582290.0.html

[Erik]

A lot of vague claims, but it looks like Mediatek and (allegedly) Samsung/AMD having an interest in Windows on ARM might have woken Qualcomm up.

[ChrisGX]

>>This time, however, we are getting to know that the Snapdragon 8cx Gen 3 will offer four Cortex-X1 Prime cores clocked at 2.995 GHz and four Cortex-A78 cores at 2.4 GHz under load.>>

You are confused. First, Qualcomm described the CPU as having 4 Prime/performance cores and 4 efficiency cores. A78 cores are performance cores not efficiency cores. A55 cores are efficiency cores. Second, Qualcomm didn't say whether the Prime/performance cores were X1 cores or A78 cores, only that the cores would be clocked at 3.0GHz. And finally, the words 'under load' in this case are kind of meaningless. Yes, if there are  8 or more threads being handled by the CPU at once then it is possible that all cores will be doing work. But, that can happen when workloads are heavy or light - it depends on whether software has been written to take advantage of multiple simultaneous threads as well as load and other factors.

Based on the 3GHz clock of the Prime cores and the claim of a 1.4x single-threaded performance lift I'd guess that all Prime cores are Cortex-A78s. (X1 cores clocked at 3.0GHz would probably be faster than 1.4x the single threaded speed of the A76 cores in the previous generation Snapdragon 8cx.)

[Tridents]

Qualcomm today unveiled the Snapdragon 8cx Gen 3 and Snapdragon 7c+ Gen 3 Compute Platforms that will power the next wave of Windows-on-ARM Always On Always Connected PCs. The Snapdragon 8cx Gen 3 is the first 5 nm SoC for the PC and features a 4+4 CPU with Cortex-X1 and Cortex-A78 cores along with other Qualcomm connectivity and security features.

Considering what we know of the previous 8cx-Gen-2 performance, a 40% improvement in single-threaded performance puts it exactly in the X1 performance ballpark at 3GHz.
And you could not achieve up to 85% multicore improvement by sticking with A55 cores.

[8&8]

Nice! Nice chip!

Hope to see soon true benchmarks also because according last leak on single thread reached 1010 ok GB5 and more of 5000 on multiple.
This is a bit weird comparing to SD888 that get a little higher values.

[8&8]

@ChrisGX, performance and efficient cores are relative terms.
See Alder Lake cores, big cores are new, little cores instead are old skylake.

X1 comparing to A78 or others consume more and are bigger and more powerful.

When we will see next 8cx gen4 that will be competitive with M1 and probably will use 4C X2 and 4c X1, obviously X1 will be efficient core aka little cores.

[okidoki]

No ethernet port? :o GTFO! ;D

[ChrisGX]

I want to acknowledge the points of @8&8 and @Tridents in a general way and knowing now that XDA Developers is the source of the claim that the Snapdragon 8cx Gen 3 CPU is composed of X1 and A78 cores alone, I don't want to pit myself against that.

Still, here is the way I came at the issue. I considered a single threaded Geekbench score of 730 to be representative of the Snapdragon 8cx Gen 2. A 40% increase on that, as it happens, squares very nicely with the early benchmark numbers we are seeing in the Geekbench database for the new chip. It is obviously wrong to say, as you seem to be saying @Tridents, that, on current indications, an X1 core at 3.0GHz would be likely to return a single threaded GB5 score of 1025, say, when X1 cores at 2.8GHz already return better results than that. It was those benchmark results that made me think the A78 was the more likely candidate and if the scores don't shift upwards from where they are now the A78 would still look rather likely.

I note that Windows Central has offered a GB5 single threaded score for the SD8cx Gen 2 of 797 as more representative. So, if that number is right we can expect to see single threaded GB5 scores for the Gen 3 SoC that exceed 1100. The A78 at 3GHz, it should be noted, couldn't post such a high score. That said, scores in the low 1100s would still be abysmally low for the X1 SoC at 3GHz. That is in the vicinity that the X1 at 2.8GHz already scores today. Of course the 5LPE process just isn't up to it.

[ChrisGX]

BTW @8&8, whatever the shortcomings of my original comment I don't see the prospective X2/X1 performance/efficiency scenario that you present as very likely. First, as clock speed alone really defines the contrast between performance and efficiency cores in the scenario you present it would be simpler and more effective to optimise for energy efficiency in that situation with an 8 by X2 implementation - 4 set to relatively high clocks and 4 to lower clocks. Second, this is just not the likely direction of things. Remember that the custom ARM Nuvia core will be ready for a 2023 release so the 2022 Gen 3 release of an 8cx SoC based on physical IP licenced from ARM will likely be the last one that we see.