AMD's Crazy Apple M4 Killer

[Kenrou]

"The AMD Ryzen Ai Max series might be the only answer to the Apple M4, at least for Windows PCs. Let's talk about some of the claimed performance along with some benchmarks from AMD and ASUS on the new ROG Flow Z13."

0:00 - Something DIFFERENT from AMD
0:46 - Ryzen AI Max vs Apple M4 Pro Claims
1:26 - CPU TDP vs GPU TDP vs Temperatures
2:15 - Flow Z13 Hands On
2:42 - Ryzen AI Max Benchmarks (ASUS Claims)
4:17 - Ryzen AI Max Lineup Explained
5:26 - "Faster" than an RTX 4090?
5:53 - Flow Z13 Overview
9:53 - Crazy Specs
10:20 - FINALLY an Apple Killer?

 

[The Coolest]

I would pay a lot to see this in an AM5 220-250W TDP package.

 

[mackerel]

So they talked about other's claims? Don't think there was any major new info there. Most of it AMD covered faster in their keynote announcing it.

Be interesting to see it in wider games but my estimated best case was that of a mobile 4070, and time spy is about there. Said it before (not necessarily on this forum?) the main advantages are unified memory for if you want big vram stuff like LLMs, or possibly a space saving argument. Otherwise there isn't really anything new. If you want high gaming performance, it doesn't really bring any significant advantage over dGPU.

I would pay a lot to see this in an AM5 220-250W TDP package.

It gets it performance characteristics from the ram it comes with. If you remove that and replace it with socketed ram, performance will fall off a cliff since AM5 has half the bus width. I don't think there will be space to include the ram in socket.

 

[]-[itman]

Framework has announced a mini-desktop using this SiP.

 

[mackerel]

The top model could be big for those doing big AI stuff. For the same unified ram, Apple would charge you about double, and Nvidia DIGITS is in between. This doesn't consider the actual performance or software environment so there can still be more value than going AMD.

For gaming, it doesn't feel special although I don't have a system pricing comparison to go off. The main selling point there is if you have to have a tiny box which would be difficult (but not necessarily impossible) to match from a regular build.

 

[don256us]

I just watched the LTT video on the Framework desktop. Apparently this chip has an issue with RAM propagation in a Laptop form or was it an issue to have the RAM in a DIMM package. I can't recall. The point was that you have to get the system with the RAM you think you will need going foraward because whatever you buy is what you're stuck with. For now.

Why is AI such a big deal for end users? AI chips, AI OS' in Windows, Android and Mac. What does one get by using AI at home?

 

[mackerel]

I just watched the LTT video on the Framework desktop. Apparently this chip has an issue with RAM propagation in a Laptop form or was it an issue to have the RAM in a DIMM package. I can't recall. The point was that you have to get the system with the RAM you think you will need going foraward because whatever you buy is what you're stuck with. For now.

I haven't watched that video, but AMD intended for it to be used with fixed ram. It doesn't necessarily rule out removable ram totally, but I guess they couldn't find a way to make it work. It is a high bandwidth solution from having 2x the width of typical systems so imagine you need 4 SODIMMs to feed it. I'm not sure how that translates to CAMM2.

Why is AI such a big deal for end users? AI chips, AI OS' in Windows, Android and Mac. What does one get by using AI at home?

The top model is more intended for those working with big models, not your average home user. Before now, that niche was pretty much limited to Apple systems costing double.