AMD Ryzen AI 7 350 Benchmark Tips Cut-Back Radeon 860M GPU

[Cpt.Jank]

AMD's upcoming Ryzen AI Kraken Point APUs appear to be affordable APUs for next-generation thin-and-light laptops and potentially even some gaming handhelds. Murmurings of these new APUs have been going around for quite some time, but a PassMark benchmark was just posted, giving us a pretty comprehensive look at the hardware configuration for the upcoming Ryzen AI 7 350. While the CPU configuration in the PassMark result confirms the 4+4 configuration we reported on previously, it seems as though the iGPU portion of the new Ryzen AI 7 is getting something of a downgrade compared to previous generations.

While all previous mobile Ryzen 7 and Ryzen 9 APUs have featured the Radeon -80M or -90M series iGPUs, the Ryzen AI 7 350 steps down to the AMD Radeon 860M. Although not much is known about the new iGPU, it uses the same nomenclature as the Radeon iGPUs found in previous Ryzen 5 APUs, suggesting it is the less performant of the new 800 series iGPUs. This would be the first time, at least since the introduction of the Ryzen branding, that a Ryzen 7 CPU will use a cut-down iGPU. This, along with the 4+4 (Zen 5 and Zen 5c) heterogenous architecture, suggests that this Ryzen 7 APU will prioritize battery life and thermal performance, likely in response to Qualcomm's recent offerings. Comparing the 760M to the single 860M benchmark on PassMark reveals similar performance, with the 860M actually falling behind the average 760M by an average of 9.1%. Take this with a grain of salt, though, since there is only one benchmark result on PassMark for the 860M.


View at TechPowerUp Main Site | Source

 

[Cheeseball]

Okay, 4+4c sounds like it could fit a proper Steam Deck-like APU since the extra cores would not be siphoning power from the overall TDP. This is the main problem with the 7840U/8840U/Z1M model, and why they can't go below 15W as well as the Sephiroth/Aerith.

The 860M would be my only concern, but it should still be better than the Aerith APU's older 8 CU RDNA2 cores.

Strix Point APUs (HX 370 and 365) can do this somewhat well since its mostly 4+8c/6c and the "c" cores power down to just under 500mW each which is why it can still be more efficient than the full fat 8-cores in than Phoenix/Hawk Point. It still suffers when TDP is set to below 13W-15W, but better efficiency than last gen.

Intel recently did this surprisingly well with Lunar Lake, especially the 258V, which is expected since its literally 4+4e without HT. On this new ExpertBook that I'm testing (63Wh battery), I'm getting 10+ hours of battery life with normal usage (Salesforce, JIRA, CodeCommit, GlobalProtect VPN, some internal work-developed software, Spotify in the background, etc.) at 7W to 25W TDP, what ASUS calls "Standard" mode.

 

[AcE]

Okay, 4+4c sounds like it could fit a proper Steam Deck-like APU since the extra cores would not be siphoning power from the overall TDP.

They already made something like this for handhelds, it's called Zenith 2 Extreme afaik (not 100% sure), but what is fact, is, that it has 3 cores and 5 Zen 5C cores. And then the iGPU with 16 CUs. It's a bit odd, don't know how good it will perform in gaming.

 

[Cheeseball]

They already made something like this for handhelds, it's called Zenith 2 Extreme afaik (not 100% sure), but what is fact, is, that it has 3 cores and 5 Zen 5C cores. And then the iGPU with 16 CUs. It's a bit odd, don't know how good it will perform in gaming.

Thats an awkward code name. LOL. Zenith II Extreme was an old TRX40 motherboard by ASUS.

Z2X or Z2 Extreme perhaps. But 3 full and 5c cores does sound more gaming-oriented since even today most games really just utilize around 4 cores heavily, unless you crank up the graphics with a dGPU since the CPU needs to process more data to feed it.

If that 3+5c+16CU is in the works, I'll welcome it, either for handhelds or ultrabooks.

EDIT: Also that PassMark benchmark between the 760M and the alleged 860M lacks a lot of info. Maybe the 860M is falling behind because its being run at a lower TGP? I see no reason why a RDNA3.5/RDNA4 part with the same amount of CUs would be weaker than the RDNA3 predecessor (e.g. 680M vs 780M, both 12 CUs).

 

[wNotyarD]

Thats an awkward code name. LOL. Zenith II Extreme was an old TRX40 motherboard by ASUS.

Z2X or Z2 Extreme perhaps. But 3 full and 5c cores does sound more gaming-oriented since even today most games really just utilize around 4 cores heavily, unless you crank up the graphics with a dGPU since the CPU needs to process more data to feed it.

If that 3+5c+16CU is in the works, I'll welcome it, either for handhelds or ultrabooks.

Should be Z2 and Z2 Extreme, the predecessors inside the Rog Ally and Legion Go are the Z1 and Z1 Extreme. My guess is just that @AcE made a little mistake.

 

[AcE]

Thats an awkward code name. LOL. Zenith II Extreme was an old TRX40 motherboard by ASUS.

Yea but it's a marketing name, not code name.

If that 3+5c+16CU is in the works, I'll welcome it, either for handhelds or ultrabooks.

It could work well, yea... because it's a bit odd i'm not sure though. The GPU is strong on the other hand.

Should be Z2 and Z2 Extreme, the predecessors inside the Rog Ally and Legion Go are the Z1 and Z1 Extreme. My guess is just that @AcE made a little mistake.

Dude it's exactly what I said. Zenith 2 Extreme.

 

[wNotyarD]

Dude it's exactly what I said. Zenith 2 Extreme.

You did your post almost at the exact time as I did, so it didn't show up. As far as I know, though, the name "Zenith" was never used for these SoC's.

 

[AcE]

You did your post almost at the exact time as I did, so it didn't show up. As far as I know, though, the name "Zenith" was never used for these SoC's.

Yea I guess you're right. Brain playing games.

 

[Neo_Morpheus]

Steam Deck-like APU

Given the success of the current Steam Deck, I'm pretty sure that ValvE will double down on a more custom SOC, like Sony and MS did with their consoles.

And who knows if the same thing will happen with the rumored new Steam Machine.

 

[Cheeseball]

Given the success of the current Steam Deck, I'm pretty sure that ValvE will double down on a more custom SOC, like Sony and MS did with their consoles.

And who knows if the same thing will happen with the rumored new Steam Machine.

Whenever that new Steam Deck 2 comes out, I'm sure they would have another custom SoC made for it. If they did it for the LCD and OLED models (7nm and 6nm), why not for the next, especially since they technically control (or at least have direct access to) the software stack too.

I'm guessing if they would have their own in-house Steam Machine, they would most likely use Strix Halo with the 20/40+ CUs.

 

[AnarchoPrimitiv]

I wish AMD would make some APUs that have the full fat iGPU (16CUs), but only 6 CPU cores.....that would make an excellent budget gaming APU...., but knowing how OEMs are, they'd probably only put them in models paired with a dGPU completely defeating the point

 

[Neo_Morpheus]

Whenever that new Steam Deck 2 comes out, I'm sure they would have another custom SoC made for it. If they did it for the LCD and OLED models (7nm and 6nm), why not for the next, especially since they technically control (or at least have direct access to) the software stack too.

I'm guessing if they would have their own in-house Steam Machine, they would most likely use Strix Halo with the 20/40+ CUs.

Seems thats their plan for both devices.

 

[GenericUsername2001]

I wish AMD would make some APUs that have the full fat iGPU (16CUs), but only 6 CPU cores.....that would make an excellent budget gaming APU...., but knowing how OEMs are, they'd probably only put them in models paired with a dGPU completely defeating the point

On the reverse side, a chips with 8 CPU cores with 32 MB L3 cache, but only a really basic iGPU, intended for use in laptops that also have a dGPU, would be nice. AMD could save a ton die space that way, but I guess the cost of setting up the production masks & what not must be too much for a more specialized product like that.

 

[Cheeseball]

On the reverse side, a chips with 8 CPU cores with 32 MB L3 cache, but only a really basic iGPU, intended for use in laptops that also have a dGPU, would be nice. AMD could save a ton die space that way, but I guess the cost of setting up the production masks & what not must be too much for a more specialized product like that.

They did this, it's the 7740HX. Has the 610M as the iGPU (2 CUs) and 32MB L3. The bad part about it is it follows the chiplet design like the mainline desktop CPUs (it's a lower TDP 7700X), so the low and idle power consumption is quite bad (does not go below 15W to 20W minimum) due to the I/O die.

 

[GenericUsername2001]

They did this, it's the 7740HX. Has the 610M as the iGPU (2 CUs) and 32MB L3. The bad part about it is it follows the chiplet design like the mainline desktop CPUs (it's a lower TDP 7700X), so the low and idle power consumption is quite bad (does not go below 15W to 20W minimum) due to the I/O die.

Yeah, should have clarified a monolithic CPU, the 7740 HX is basically just a rebadged desktop chip. I do hope that AMD can tame the idle power usage on their chiplet based stuff, would offer them a lot more flexibility for the mobile market; will have to see how Strix Halo works out in that regard, as it will be chiplet based but designed specifically for mobile uses.

 

[Tek-Check]

While all previous mobile Ryzen 7 and Ryzen 9 APUs have featured the Radeon -80M or -90M series iGPUs, the Ryzen AI 7 350 steps down to the AMD Radeon 860M. Although not much is known about the new iGPU, it uses the same nomenclature as the Radeon iGPUs found in previous Ryzen 5 APUs, suggesting it is the less performant of the new 800 series iGPUs. This would be the first time, at least since the introduction of the Ryzen branding, that a Ryzen 7 CPU will use a cut-down iGPU.

- the narrative here is not entirely correct; it's more complicated
- '-90M' iGPU name was introduced with Strix Point, therefore occupying a new space with best graphics for 12 core APU
- naturally, all other, lower iGPU names, including 880M and 860M, are reserved for lower SKUs with 10 cores and less
- they decided that 12 and 10 core SKUs 375, 370 and 365 are R9
- the only SKU on Strix Point chip with 8 cores (3+5) and 880M that is named R7 is 360 PRO, for some Lenovo systems
- https://www.amd.com/en/products/processors/laptop/ryzen-pro/300-series/amd-ryzen-ai-7-pro-360.html
- Kraken chip for R7 350 with 8 cores (4+4) will have 860M with 8 CUs; Phoenix/Hawk Point R7 with 8 cores also have 760M with 8 CUs
- so, 890M is now what 780M used to be; 880M is now what 760M used to be, 860M will be what 740M used to be, etc.
- the iGPU has actually been upgraded across the stack with 4 extra CUs on each tier
- we should not look into face value names of iGPUs

1. Strix Point R9 375 and R9 370 with 12 cores - 890M with 16 CUs
2. Strix Point R9 365 with 10 (4+6) cores and R7 360 PRO with 8 (3+5) cores - 880M with 12 CUs
3. Kraken R7 350 with 8 (4+4) cores - 860M with 8 CUs
4. perhaps future R5 and R3 will have '840M', we don't know at the moment

- they split R7 naming between the two dies, hence confusion about iGPU names
- R7 APUs have both 880M adn 860M, depending on the die used, Strix or Kraken