News Alleged Zen 5 'Strix Halo' Mobile APU has more GPU cores than RX 7600 XT or PS5 — features monster RDNA 3.5 GPU with 40 compute units

[Alpha_Lyrae]

This is basically all incorrect which makes the validity of the rest of your post questionable.

The folks who have replaced the 6400 in the Ally for 7200 have seen linear performance increases in anything that wasn't completely CPU bound (not to mention the DDR5 testing on the socketed APUs). 720/1080p aren't arbitrarily CPU bound just like no resolution is, because it's dependent on your hardware. 12CU is absolutely not enough graphics performance to cause a CPU bound situation in very many titles. The power bias is also not towards the GPU in AMD's APUs unless you're using third party software to force such behavior.

There are gains in iGPU performance at DDR5-7200 if you OC GPU to 3.3GHz using an 8700G. In a power-limited scenario, as most laptops are, this is NOT achievable. Therefore, the primary limiter of performance is overall power and CPU IPC and/or clocks, and iGPU clocks, if there's power to spare. Handhelds are even more power-limited. Only when you can remove power limits completely and can clock CPU and iGPU high enough will you hit a memory bandwidth limit. In laptops and handhelds, STAPM will limit performance after 2 minutes unless custom firmware or 3rd party SMU override is used. This cuts CPU and iGPU boost to fit within a power profile design (15W, 28W, 35W, 45W, etc). STAPM is only disabled in desktop parts like 8700G/8600G (after AMD fixed it in a post-release AGESA).

Vega 8 in 5800H/HS gained quite a bit of performance over 4800H/HS just by moving from Zen 2 to Zen 3 and clocking iGPU higher to keep up with increased CPU IPC. iGPU clocks were also decoupled from CPU domain.

You're making assumptions to fit your particular viewpoint, which are not correct. And yes, 1080p is CPU-limited, and why AMD even bothers to add more GPU cores to Strix Point. If they were strictly memory bandwidth limited, AMD would not increase CU count from 12 to 16CUs, as there would be no gains. APUs are limited by a confluence of factors, not solely memory bandwidth.

Strix Halo is targeting 120-170W TDP, which should provide efficient performance.

 

[thestryker]

There are gains in iGPU performance at DDR5-7200 if you OC GPU to 3.3GHz using an 8700G. In a power-limited scenario, as most laptops are, this is NOT achievable. Therefore, the primary limiter of performance is overall power and CPU IPC and/or clocks, and iGPU clocks, if there's power to spare. Handhelds are even more power-limited. Only when you can remove power limits completely and can clock CPU and iGPU high enough will you hit a memory bandwidth limit.

This is quite literally just nonsense. The first thing I cited was people who literally swapped LPDDR5 6400 with LPDDR5 7200 in the ROG Ally (maximum 35W device) and got linear performance increases in non-CPU bound workloads. There are absolutely many things that go into the performance limitations but to claim that you cannot hit memory bandwidth limit except at really high clocks is just pure willful ignorance.

And yes, 1080p is CPU-limited, and why AMD even bothers to add more GPU cores to Strix Point.

This is one of the dumbest things I've ever seen someone proclaim as fact. 1080p like every other resolution can be CPU limited, but that certainly doesn't mean that it is. If they're adding CUs it's because they've found they can get enough extra performance to justify the cost. That doesn't mean that memory bandwidth isn't limiting maximum performance. They may also be adding cache which can have a pretty big impact, especially at lower resolutions, without a big increase to memory bandwidth.

APUs are limited by a confluence of factors, not solely memory bandwidth.

Weird that you say this after claiming 1080p is CPU limited and you can't have memory bandwidth problems until overclocking. Make up your mind it's either multiple factors including memory bandwidth or it's not.

 

[usertests]

This is technically achievable, but the design is backwards, IMO. CCDs don't need fanout bandwidth, but a GPU sure does and it needs it directly connected to memory/cache. It'd make more sense if the CCDs were LPDDR5x MCDs in this scenario. There are 10.7Gbps LPDDR5x modules from Samsung, which would offer 339.2GB/s at 256b + any MALL bandwidth amplification. These should be ready by the time Strix Halo launches in H1 2025.

Here's a leak that suggests Strix Halo might max out at LPDDR5X-8000:

AMD Strix & Strix Halo APU Official Docs Allegedly Leak: Up To 16 Zen 5 Cores, 40 RDNA 3+ iGPU, 60 AI TOPs, LPDDR5x-8000, 32 MB MALL

An alleged "official" AMD document with information regarding AMD's Zen 5-powered Strix & Strix Halo APU has leaked out.

wccftech.com

 

[TheHerald]

Man, I wish/hope this is real. To match the fps/graphics of a $500 PS5 you need an RTX 4080 laptop, and its hard to find one under $2k, and the total system power draw is over 200 watts. I just want a reasonably thin 17" laptop with that PS5/Xbox chip in it, and an NPU attached.

A ps5 is nowhere near a 4080, not even the mobile one.