4.3.4 Larrabee2
Knights Corner is an optimization and upgrade of Knights Ferry. It is only natural that the Larrabee2 graphics card appears, but Intel did not announce it to the public, and Intel has done such things.
The PCB date is 3611 weeks and the GPU date is 1147 weeks.

4.4 Unknown
　At the beginning of 2018, Intel first revealed the prototype of the fourth-generation discrete graphics card at the ISSCC 2018 International Solid-State Circuits Conference. The prototype graphics card was manufactured using the Intel 14nm process and integrated 1.524 billion transistors.
　Where is Intel going, will it open a new era? Wait and see!

Very interesting article. Thank you for posting. Info on Intel GPUs isn't very plentiful online.

I'm curious, do you know if Intel Extreme Graphics is related to the i740 family or if it is a new design?

None of the pictures work... 🙁

Very nice overview of all the Intel graphics chips. Thanks!

I was at Siggraph 2008 and watched the Intel presentation for Larrabee. It failed to materialize in any meaningful way. CUDA was the de facto standard in scientific computing very soon thereafter.

PS Images work fine for me.

Larrabee became xeon phi. There was some interesting work done on early Larrabee, such as the first raytraced Quake with actual playable frame rates. Done by this guy (http://www.q4rt.de/Daniel%20Pohl/). iirc the early Quake3 raytracing that he did actually got him the job at Intel where he ended up working on Larrabee and did the Quake:wars raytraced demo for it. I believe vlask member on these boards has one, unsure if he ever got it working though?

PPS Images work fine for me too.

I do remember the Quake demos. However the Xenon Phi boards were not very popular either, as far as I can tell. Back then and now everyone I know uses Nvidia boards.

I knew of at least one of our competitors development projects with them in the industry I work, mainly used for CFD. I would have thought they would do well being x86 in'all... however there were development issues with them and they weren't just a co-processor that integrated well with current x86 execution. Similar to a GPU you had to upload programs to the Phi's own eco system so the execution was extremely sandboxed. Intel had the right idea back in 2006/07 with the way we now know GPU's to have gone (unified architectures), however it seems smaller pipelines and more of them became the preferred option. Perhaps it wasn't specialised enough for graphics, more for compute and then by the time they realised, nV had progressed beyond the horizon with their Tesla architecture. Shame really... given then lack of uptake and massive price when realeased, I'm hoping to pick one or two up on the cheap for some funzzies at some point (although not too sure how they would fare with todays CPU throughput?).