27 comments

[ 8.8 ms ] story [ 43.4 ms ] thread
I am a little confused, I thought Intel had a big push the last couple of years to create its own GPU:

https://www.intel.com/content/www/us/en/products/docs/discre...

TSMC makes Intel's GPUs
Intel has been making GPUs since the early 1980s, starting with the 82720, or the 82716 if you want to be picky and require a pure-Intel design. They announce a new GPU effort every few years, at about the time it's clear that the previous one has failed.

Again being picky, in theory their integrated graphics are a "success" in that they sell well, but that's because vendors get them for free with the CPU and so don't have to go through the expense of adding a discrete one.

I mean, they're a success in that even a weak discrete GPU is extremely overkill for the majority of people who just want to browse. You can only integrate that kind of card into another chip because the overhead of adding IO and another PCB is just too high for such a weak GPU.
The silicon is just one piece of the puzzle. CUDA and the rest of the software stack is huge advantage for NVIDIA.
Intel people would point at SYCL if you told them this
Intel focused on SyCL which not many people seem to actually care about. It looks far enough removed from CUDA you’d have to think hard about porting things as well. From what I understand ROCm looks very close to CUDA.
Good to hear. More than two players in the GPU market is a really good thing and their recent dedicated consumer GPUs are really good value in their segment. It will take a few generations until they might catch up to Nvidia, but I am hopeful. This is a good thing.
So they are just moving GPU production in-house? Weren't they already designing GPUs for contract with TSMC?
Deja Vu all over again.

2016 Nervana. Intel would lead in AI training. The "Nervana NNP" was the future.

2019 Habana Intel announced the Gaudi and Goya chips as their new official AI strategy, effectively killing the Nervana project.

2021 Xe general HPC/AI GPU (Ponte Vecchio) Intel said they will be shifting to the "AI chip" market.

2023 The "AI PC". every consumer CPU would now be an "AI Chip" with NPU (Neural Processing Unit).

2024 Intel is now "AI Systems Foundry" to focus on making AI chips for other people (like Microsoft and Amazon).

2026 Intel will start making GPUs

3 years too late to get serious about GPUs.
It's too late after Sony and Microsoft choose AMD64 SOC (CPU+GPU) as gaming platform. AMD is the only provider over 10 years.
Intel has been designing GPUs manufactured on TSMC nodes across client and datacenter for at least the past 5 years. The client chips are price competitive but not performance competitive with AMD/NVIDIA/Apple. The data center roadmap has historically been a huge mess with cancelled products left and right. But, to say "Intel will start making GPUs" seems misleading. Perhaps "Intel to try to inject sanity into its GPU roadmap" would be a better headline, though I am skeptical one hire will do anything to fix 10+ years of mismanagement.
It's a confusing article. It's strongly implies that Intel will make GPUs for data centers. It says Intel will produce GPUs without saying whether they are manufacturing them in house or not.
It is my understanding that this isn't happening in any meaningful capacity, they're simply using the kit no longer relevant to R&D.

I'm still not entirely convinced they actually did Arc themselves. It has all the hallmarks of a project that was bought or taken. Every meaningful iteration keeps getting pushed back further out towards the horizon and the only thing they've been able to offer in the meantime is "uhhhh what if we used two"

intel has been making graphics silicon since the 90s, the current discrete graphics effort has been going for at least a decade, and in areas like low power video decode and encode it could be argued intel is class-leading. the concept of the "GPU" is a quarter of a century old. this is an especially poor article, especially for a publication running as long as techcrunch.
The CEO of Intel already specifically said Intel has given up competing with Nvidia.

That’s the spirit!

The most rapid path that Intel has to selling competitive GPUs, would be to licence designs from Groq, and apply all effort to getting them working on 14a.

Hyperscalers would bite their hand off and would be a viable alternative to TSMC.

Nvidia has left the door open with the non-exclusive license in the acquisition

Wtf Intel is doing, they can't even communicate well !
I'd be more excited about Intel making HBM.
“We realized that we were too focused on the consumer market. “ - Intel
Didn't know my A770 wasn't a GPU.
I wonder what the hell is that damn thing I select sometimes on my Vulkan engine with --gpu-select=Intel :)