Intel might actually do well next year. Actually out of QC and NV, it’s Intel that has the highest chance.
Cause NVL SoCs will be good. NV would be great comp but WoA needs to improve
nvidia cannot ship x86
intel will take a miracle to compete on-time
who else?
Yep.Intel might actually do well next year
ehhhhh I mean. They've shipped GT3/4e attempts for more than you could possibly remember.Actually out of QC and NV, it’s Intel that has the highest chance.
Maybe.Cause NVL SoCs will be good
Qualcomm. Technically. Glymur is a halo-ish part.
Everybody, Print it, hang it above your bed, so that you could see it every morning and evening, and remember that APUs are going to be priced based on CPU AND GPU performance. Not just GPU.But the pricing there isn't based off of the GPU performance only. In case you forgot, those products also have CPUs attached to them as well. The pricing for these products will factor in both sides of the argument. The top end 800Gs were factored like they are what they are: 8 core CPUs with a competent - but not super powerful - iGPU attached. Trying to suggest APUs should be priced to compete against dGPUs is a bit ridiculous, it completely ignores half the product.
Indeed.I think the 120W TDP is the case when both the CPU+GPU are fully utilized (not in the Flow z13's case, but a hypothetical mini PC or large workstation laptop with sufficient cooling split in 55W+65W for the SoC) Thoughts?
Intel will compete only with 128 Bit chips, so not technically halo competition, however, they will have far superior product in that 128 bit bus category.Intel might actually do well next year. Actually out of QC and NV, it’s Intel that has the highest chance.
Cause NVL SoCs will be good. NV would be great comp but WoA needs to improve
He means NVL-AX.Intel will compete only with 128 Bit chips, so not technically halo competition, however, they will have far superior product in that 128 bit bus category.
Both. Current graphics rendering expects all graphics computing to be done in one single NUMA. To still go chiplets with computing the links between each of them would need to be able to handle the total internal bandwidth, and that's financially unfeasible in a consumer product and is the more costly the more you want to scale it up. Without that you will see dramatic slowdowns due to bottlenecks, kind of akin to Zen's CCX to CCX latency but much worse since no software is prepared at all for something like this to happen. The result would be lower framerates for two compute chiplets than for one.My biggest question is why they’re sticking to just one GPU die on consumer products? Does it cause too many software compatibility problems or is the required packaging just too expensive?
It's all feasiable and pretty easy, the hard part is making the serial parts of your gfx pipeline (lot of them) parallel.To still go chiplets with computing the links between each of them would need to be able to handle the total internal bandwidth, and that's financially unfeasible in a consumer product and is the more costly the more you want to scale it up. Without that you will see dramatic slowdowns due to bottlenecks, kind of akin to Zen's CCX to CCX latency but much worse since no software is prepared at all for something like this to happen. The result would be lower framerates for two compute chiplets than for one.
you can't.But AMD for some reason doesn't appear to want to push for the required rendering software changes
It's feasible and pretty easy except you can't. Good succinct way to describe the conundrum indeed.It's all feasiable and pretty easy, the hard part is making the serial parts of your gfx pipeline (lot of them) parallel.
you can't.
No one will ever bother with anything resembling explicit multiadapter.
Yeah they can, you can rent MI300X on a lot of places.It's feasible and pretty easy except you can't
a) dense cores suck for non-ghetto purposescould they really cram 80 CUs with everything else less and like only 8 dense cores etc? 2 GPU dies doesnt make much sense for mobile devices
They could physically fit more GPU chiplets in future Halo parts if they 3D stacked the CPU chiplets on top or something.could they really cram 80 CUs with everything else less and like only 8 dense cores etc? 2 GPU dies doesnt make much sense for mobile devices
AMD had been proposing big APUs for mobile for years, when Apple did it PC OEMs finally realised there is a market for them.AMD VP explains why the Ryzen AI Max likely wouldn't exist without Apple
At CES 2025, AMD's Joe Macri praised Apple for proving that people would buy powerful computers without discrete GPUs.www.engadget.com
Thats interesting, all the M4 Pro references make sense now.
With the success of Apple Silicon, Macri was finally able to get approval to spend a "mind boggling" amount of money developing the Ryzen AI Max. "I always knew, because we were building APUs, and I'd been pushing for this big APU forever, that I could build, a system that was smaller, faster, and I could give much higher performance at the same power," he said.
There was never enough market interest, suits are needed for final approval, but they never got that close.whoever kept denying the big APU project needs to go... AMD still has bad suits calling shots
OEMs did.whoever kept denying the big APU project needs to go... AMD still has bad suits calling shots
wow it took 10 years, kinda sad the OEMs have no visionOEMs did.
AMD's been trying to pitch a chungus since Kaveri.
They had vision.wow it took 10 years, kinda sad the OEMs have no vision
Uh no Intel spent years shipping fat GFX configs.Big APUs were always against the interest of both Intel and Nvidia, because the only vendor that could've provided them was AMD
Those were still 128 bit configs .Uh no Intel spent years shipping fat GFX configs.
Only that they sucked.
Good news, halo doesn't!