Hacker News Reader

Hacker News

Comment by fancyfredbot | Hacker News Reader

Parent Full thread

fancyfredbot

They aren't divesting from discrete GPUs. The Battlemage launch is tomorrow.

View on HN

Gelsinger doesn't think there will be any demand at all for GPUs in the near future!

> "Gelsinger said the market will have less demand for dedicated graphics cards in the future."

From: https://www.techspot.com/news/105407-intel-not-repeat-lunar-...

He may have been talking about something like, "all GPU stuff will just be integrated into the CPU" but I highly doubt that's realistic. Nvidia's latest chip die size is enormous! Their new 5090 die is reportedly 744mm squared: https://www.tomshardware.com/pc-components/gpus/the-rtx-5090...

There's no way you're going to be getting an equivalent level of performance from a CPU with integrated GPU when the GPU part of the die needs to be that big.

He said less demand. He didn't say there would be no demand. Importantly he didn't say intel was divesting discrete GPUs. Perhaps even more importantly they demonstrably aren't divesting.

With the B770 16GB gone and the idea for the B580 to be cheaper than current 7600 XT @ 16 GB by cutting 4 GB makes Battlemage DOA. A gamer making an investment on a card for ~3 years cares less about spending ~$30 more vs being unable to run high resolution texture packs on a gimped GPU. The XMX cores are superior for AI for a month until Blackwell with smaller 2 and 4FP units, but a month is too little lead to overcome the CUDA software inertia. The next beancounter CEO gets the gift of terrible sales numbers and the excuse to drop the prices to move them out before RTX 5000 & RX 8000 competition hits.

Maybe you are right and Battlemage is DOA. Perhaps intel know that and just want to dump inventory before announcing they will get out of the GPU business.

On the other hand, maybe not.

My point is that although you might think they are going to divest the GPU business in future, we don't know that for sure and it's kind of weird to present it as having already happened.

Intel's Xe2 GPU architecture works well as an efficient iGPU and has market fit there. The problem is taking it the other way to discrete high end and data center rather than the opposite as NVidia has done with GPUs and AMD with CPUs. Using the low power consumer target to experiment on architecture and new nodes that then scales to the high end has been Intel's strategy ("efficiency" cores) is logical from a foundry perspective for monolithic chips. But Intel has failed to execute in the new AI data center investment reality. Where are the GPUs from Gaudi 3? Pat: "Putting AI into all the chips, not just ones in the cloud, might be more important in the long run" maybe, but failing to win where the money is at now is a huge barrier to surviving in that long run.

Reply on news.ycombinator.com