AMD Zen 4 vs. Zen 4C Performance, Zen 4C Core Scaling With Ryzen 5 8500G

[phoronix]

Phoronix: AMD Zen 4 vs. Zen 4C Performance, Zen 4C Core Scaling With Ryzen 5 8500G

Besides the integrated RDNA3 graphics making the Ryzen 8000G series desktop APUs interesting, making the AMD Ryzen 5 8500G a fun benchmarking target besides its sub-$200 price tag is having a mix of Zen 4 and Zen 4C cores. Here are some benchmarks looking at the Zen 4 vs. Zen 4C performance and power efficiency when offlining various core combinations on the Ryzen 5 8500G desktop processor.

https://www.phoronix.com/review/amd-zen4-zen4c-scaling

[bezirg]

Well, Zen4C is the e-cores that we really wanted, but instead we got the monstrosity called alderlake/raptorlake/meteor lake. Let's hope for the next iteration of intel p+e cores.

[Kjell]

AMD's approach to "e-cores" makes sense unlike the monstrosity which Intel has produced. I can see it being useful for laptops or datacenters.. Desktops? No.. It's a waste of silicon. JUST MAKE PROPER CPUS

[JEBjames]

"one Zen 4 and one Zen 4C core was around 84% the speed of two Zen 4 cores"

Wow.

AMD can glue together their lowest binned Zen 4C cores, lowest yield slowest GPU dies. Glue it together. And it's still pretty decent. In fact, it's pretty good?

Kind of crazy that a top end 10900k of a few years ago is now in the same performance universe as a budget 8500g? And the 8500g is half the watts?

[Teggs]

This is interesting data. My view is that dual-core should go away entirely for desktop and laptop computers. As the bench shows, if you want efficiency you can keep two of the cores off, or at least 'off' via software. But when you need performance of any kind, you want those extra two cores.

This also indicates that at least the CPU performance of the 8300G (1Zen +3 ZenC) shouldn't be a complete disaster, unlike its PCIe connectivity. On the other hand, it shows that AMD does need to make very clear to the customer at all levels what they are buying.

[ayumu]

Very nice article. I have been looking forward to someone (anyone) looking at the 8500G in more detail for a while now.

Still, I feel it would have been nice to also look at only 4x zen4c and at only 2c zen4c (to pit directly against the 2x zen4 measurements).

[Michael]

Very nice article. I have been looking forward to someone (anyone) looking at the 8500G in more detail for a while now.

Still, I feel it would have been nice to also look at only 4x zen4c and at only 2c zen4c (to pit directly against the 2x zen4 measurements).

As mentioned in the article, it's impossible to offline CPU0 which is a Zen 4 core or otherwise disable the Zen 4 cores from the BIOS, so really can't get an accurate measurement of only 4C cores on the 8500G.

[ayumu]

As mentioned in the article, it's impossible to offline CPU0 which is a Zen 4 core or otherwise disable the Zen 4 cores from the BIOS, so really can't get an accurate measurement of only 4C cores on the 8500G.

I see. "But under Linux there is at least the ability to offline CPU cores... Though CPU0 can't be offlined". Well, that's annoying.

Look into building a kernel with CONFIG_BOOTPARAM_HOTPLUG_CPU0 and see if that helps.

Boot parameter documented here: https://www.kernel.org/doc/html/latest/core-api/cpu_hotplug.html

[NateHubbard]

So the Zen 4c is supposedly 35% smaller, and it looks like the performance is about 35% less. I guess that means it scales really well with cost / performance.
What I mean is, it should cost AMD at least 35% less to manufacture per core, probably less than that depending on yields. I have no idea what they will charge.

[X_m7]

Has there been any benchmarks of Zen4 vs Zen4c at the same clockspeed? I'd be interested in seeing how much of the 4c's efficiency advantage is just because of its lower frequency.