AMD Ryzen 7 8700G desktop APU with Radeon 780M pops up on Geekbench

[Redaktion]

Soon to be officially unveiled by AMD, the Ryzen 7 8700G has surfaced on Geekbench. The Hawk Point AM5 desktop APU went through ML tests on the benchmarking platform, which evaluated the AI capacities of the chipsets. According to the results, the upcoming processor did better than the  Ryzen 9 7940HS.

https://www.notebookcheck.net/AMD-Ryzen-7-8700G-desktop-APU-with-Radeon-780M-pops-up-on-Geekbench.786362.0.html

[A]

Just for context, this number is around iPhone 15 Pro Max result, M3Max score is almost double than that of AMD. M3Pro is x1.5, and base M3 is also slightly higher.

Core Ultra is worse than all of the above. "AI-ready" my *ss.

P.S. There's a kink in comparison with iPhone though, iPhone beats it twofold in some tasks and loses twofold in some, getting the same ML score overall.

[A]

P.S. NPUs are compared, of course. GPUs give higher score.

[Anonymousgg]

8700G ONNX DirectML: 5255
7940HS ONNX DirectML: 5002, 5066, 5094, 5174, 5041, 16232

Clicking on each of the 7940HS results indicates the 780M graphics was used, except for RTX 4080 Laptop for the 3x higher result. The 8700G result was also using 780M graphics. The Ultra 7 155H results used "Intel(R) Arc(TM) Graphics".

1-5% faster is within expectations for a desktop APU version with slightly relaxed power limits. I don't see any evidence of XDNA being used in any tests, whether it's at the Phoenix (10 TOPS) or Hawk Point (16 TOPS) level. Intel's NPU also wasn't used. These accelerators aren't well supported in general yet. They would also probably give significantly lower results, since according to AMD, Phoenix and Hawk Point deliver 23 TOPS from the GPU (possibly also using the CPU).

images.anandtech.com/doci/21178/Ryzen%20AI%20Press%20Deck_Under%20embargo%20until%20Dec%206%20at%2012PM%20PST-01%201%20%2826%29.png

[A]

They would also probably give significantly lower results

This, so it doesn't actually matter if NPU or GPU was used. NPU will not get a higher score, 5200 GPU is bad news for NPU.

[Neenyah]

People buy G CPUs for their budget-oriented gaming rigs, like the legendary 5600G, so this whole article is way off (and pretty irrelevant).

[RobertJasiek]

People buy G CPUs for their budget-oriented gaming rigs

Or for dGPU-load computers with iGPU as trouble shooting backup.

[Neenyah]

True Robert, that too 👍

[Enma45]

Just for context, this number is around iPhone 15 Pro Max result, M3Max score is almost double than that of AMD. M3Pro is x1.5, and base M3 is also slightly higher.

Core Ultra is worse than all of the above. "AI-ready" my *ss.

P.S. There's a kink in comparison with iPhone though, iPhone beats it twofold in some tasks and loses twofold in some, getting the same ML score overall.

All the people on the forums have the same opinion that Intel has fallen short in CPU and NPU compared to what AMD launched with Zen 4 7040 Phoenix, but you already know Intel's marketing and their briefcases have been keeping a company afloat for more than 10 years that he rested on his laurels because he had a monopoly.
Now we just have to see AMD and Hank Point have done a great job with the drivers and new Bios with significant improvements and see what functions AMD's XDNA has that we still don't see what utilities and they promised that in 2024 they would have new useful functions for Windows 12 and everything regarding artificial intelligence

[A]

People buy G CPUs for their budget-oriented gaming rigs

If Windows will start using AIs for everything running them on external GPUs will be a waste of power. This comparison is more like "current state of tech" than real competition.

Hank Point

"Hank" made my day.

[Anonymousgg]

They would also probably give significantly lower results

This, so it doesn't actually matter if NPU or GPU was used. NPU will not get a higher score, 5200 GPU is bad news for NPU.

Right.

The 45-50 TOPS "XDNA2" in Strix Point could be faster than using the iGPU, or at least comparable.

The NPUs in Phoenix/Hawk Point and Meteor Lake are there to lay the groundwork and get developers used to them. They might save some battery life in some applications if you're lucky. Otherwise they are a curiosity that can be ignored for this generation.