News Bombshell: Intel-Nvidia partnership may kill Arc GPU

[Tup3x]

So this is just going to further the quasi-monopoly that Intel and nvidia have in the laptop market? Very interesting that this development is happening right as AMD is preparing Zen6/RDNA5 APUs, which places it in a good competitive position for the first time ever in mid-range and high-end gaming/creator laptops.

That's probably pretty much what NVIDIA is after. It's pretty obvious that discrete GPUs for laptops (as in traditional way, with their own dedicated VRAM) are dead. It's easier to sell laptop with GeForce than other GPU and Intel knows that.

 

[igor_kavinski]

But I did invest on 2xRadeon Pro W7900 48GB cards for my GenAI 3D passion project. Came in at under $8000 even after tax, and definitely worth it.

Wouldn't 3xR9700 32GB be better?

 

[igor_kavinski]

It's easier to sell laptop with GeForce than other GPU and Intel knows that.

It's sad that Intel gave up without a fight. I could see 16 and 17 inch laptops with 275HX and B580 12GB desktop chip with great cooling making good financial sense against Nvidia's crappy 8GB mobile parts.

 

[marees]

So this is just going to further the quasi-monopoly that Intel and nvidia have in the laptop market? Very interesting that this development is happening right as AMD is preparing Zen6/RDNA5 APUs, which places it in a good competitive position for the first time ever in mid-range and high-end gaming/creator laptops.

What MLID was also saying.

 

[igor_kavinski]

@adroc_thurston What will be AMD's response?

What I see in my mind: Lisa Su shrugs and says to her people, "I never cared about tiny GPUs anyway. Back to work making those MIxxx monsters!".

 

[adroc_thurston]

What will be AMD's response?

Why would they care.

What I see in my mind: Lisa Su shrugs and says to her people, "I never cared about tiny GPUs anyway. Back to work making those MIxxx monsters!".

They only care about their own roadmaps.

 

[Joe NYC]

@adroc_thurston What will be AMD's response?

AMD will go back to the OEMs and will tell them:

"See, we were telling you Strix Halo is the future"

 

[dr1337]

imagine if everyone thought kabylake-G meant intel was transitioning to AMD

oh wait, some people actually thought it would happen lol

 

[Keller_TT]

Wouldn't 3xR9700 32GB be better?

I thought about it, but it didn't tally. I don't require it for LLM inferencing or quantizing very large multimodal/purpose-built models for which I have the Tenstorrent box. Also, the W9700 was still an unknown quantity.

In my case, RDNA4 would only have made sense if I needed INT4 quantization where it is vastly superior. But this build is specifically for rasterized compute on FP16, BF16 workloads and predominantly requires high bandwidth and VRAM for pose estimation, 3D Mocap, and short animated renders with high FPS. 864 GB/s with 48 GB VRAM per card means larger batch sizes, speed and accuracy without compromise/workarounds. Works great with ROCm and Pro Render. It will thump the W9700 in this case.