Howdy, Stranger!
It looks like you're new here. If you want to get involved, click one of these buttons!
Quick Links
Intel to make high end discrete GPUs
https://www.anandtech.com/show/12017/intel-to-develop-discrete-gpus-hires-raja-koduri-as-chief-architect
And they've hired away the head of AMD's GPU division to lead the effort.
This will be Intel's third effort at making a high end discrete GPU. The previous two were spectacular failures, not helped by Intel's efforts at getting far too cute.
In 1998, Intel released the i740 Starfighter. Intel claimed that the then-new AGP slot was so fast that the video card didn't need dedicated memory on the card. They were spectacularly wrong about that. Intel then backed away from the discrete video card market, though some of what they had done did go into the awful GMA integrated graphics products that they inflicted upon the world over the course of the next decade.
In 2009, Intel demonstrated their Larrabee GPU for the first time, after long promising that it would blow away the competition. For some inexplicable reason, Intel had the idea that you could make a GPU out of a bunch of x86 cores. Naturally, Intel showed it off not doing graphics, but a compute benchmark. Observers noticed that a heavily overclocked prototype of an Intel GPU that wasn't anywhere near launching lost outright to an AMD GPU that was already available at retail for under $400 in Intel's own cherry-picked benchmark. Larrabee got canceled not long after that, though its successors did eventually turn into Xeon Phi, which is also a ridiculous product, and the PCI Express version of it has already been discontinued.
Thus, it seems that Intel was about due to take another stab at the GPU market. I'm not sure if they'll actually try to do it right this time, or if they'll once again find some creative way to do it all wrong, such as using Optane as GPU memory.
And they've hired away the head of AMD's GPU division to lead the effort.
This will be Intel's third effort at making a high end discrete GPU. The previous two were spectacular failures, not helped by Intel's efforts at getting far too cute.
In 1998, Intel released the i740 Starfighter. Intel claimed that the then-new AGP slot was so fast that the video card didn't need dedicated memory on the card. They were spectacularly wrong about that. Intel then backed away from the discrete video card market, though some of what they had done did go into the awful GMA integrated graphics products that they inflicted upon the world over the course of the next decade.
In 2009, Intel demonstrated their Larrabee GPU for the first time, after long promising that it would blow away the competition. For some inexplicable reason, Intel had the idea that you could make a GPU out of a bunch of x86 cores. Naturally, Intel showed it off not doing graphics, but a compute benchmark. Observers noticed that a heavily overclocked prototype of an Intel GPU that wasn't anywhere near launching lost outright to an AMD GPU that was already available at retail for under $400 in Intel's own cherry-picked benchmark. Larrabee got canceled not long after that, though its successors did eventually turn into Xeon Phi, which is also a ridiculous product, and the PCI Express version of it has already been discontinued.
Thus, it seems that Intel was about due to take another stab at the GPU market. I'm not sure if they'll actually try to do it right this time, or if they'll once again find some creative way to do it all wrong, such as using Optane as GPU memory.
Comments
What I found funny is I was following this story vaguely for the last couple months. Raja Koduri goes on leave to prep for next phase of development. Many speculated he would be leaving AMD. Raja Koduri leaves AMD earlier this week. Many speculated he would join Intel... And now Raja Koduri joins Intel.
If Intel is successful in this venture it would be really bad news for nVidia.