Intel reveals Ponte Vecchio GPU for HPC/AI acceleration

Quote:

---

GPU arch will be supported by OneAPI model, and feature in the Aurora Supercomputer.

---

Read more.

Who got a house on their marketing buzzword bingo?

What's interesting is that they're saying 7nm rather than 10nm. I thought Xe was on 10nm?

I'm still very puzzled how they can create a gpu without accidentally infringing IP of the very experienced and well established competition.

Then when you look at the speed they have thrown it together, even if it's just terrible, I'd be very suspicious they intentionally stole ip

Quote:

---

Originally Posted by Yoyoyo69

I'm still very puzzled how they can create a gpu without accidentally infringing IP of the very experienced and well established competition.

Then when you look at the speed they have thrown it together, even if it's just terrible, I'd be very suspicious they intentionally stole ip

---

competition like to wait for it to be released before the accusations, then they can say about loss of earnings and all that other stuff, more coffers

Quote:

---

Originally Posted by Yoyoyo69

I'm still very puzzled how they can create a gpu without accidentally infringing IP of the very experienced and well established competition.

Then when you look at the speed they have thrown it together, even if it's just terrible, I'd be very suspicious they intentionally stole ip

---

Depends if they're utilising any "special sauce".

A GPU core is relatively unlicenseable/patentable, it's the funky cool stuff to make it do weird and wonderful better things that are.

Quote:

---

Originally Posted by Yoyoyo69

I'm still very puzzled how they can create a gpu without accidentally infringing IP of the very experienced and well established competition.

---

Easy to overlook that Intel already offer GPUs, albeit very weak integrated ones... so this is more an evolution (admittedly a pretty major evolution) than a cold start for them.

They're already pumping triangle-based geometry, textures and shaders, etc through a (underpowered) pixel pipeline to a screen buffer via their Iris/Iris Pro chipsets - so the principles are well trodden by Intel, we've just not been accustomed to them playing at the mid-to-high end of this particular field.

Fingers crossed a three-horse race in the GPU market will help drive better pricing for consumers... Intel doing this on a 7nm process is perhaps a sign they'll hit the ground running.

Oh great 'another' custom code for programs to support....

While oneapi seems like a good idea it seems like it's only designed for intel hardware....

https://i0.wp.com/www.strangerviews....5188710713.jpg

No Child/Developer Left Behind

I'm assuming the new Intel GPU will be parallel in its architecture........

Quote:

---

Originally Posted by Tabbykatze

What's interesting is that they're saying 7nm rather than 10nm. I thought Xe was on 10nm?

---

I just took that as an indicator of how far off this vapourware is from shipping.

Quote:

---

Originally Posted by Yoyoyo69

I'm still very puzzled how they can create a gpu without accidentally infringing IP of the very experienced and well established competition.

---

They already ship integrated GPUs as well as the old HPC cards that started life as Larrabee, and so already have cross licence agreements to suit.

Quote:

---

Originally Posted by DanceswithUnix

I just took that as an indicator of how far off this vapourware is from shipping.

---

I initially typed something sarcastic and bitchy about this whole Xe stuff as a paper launch akin to their 5G modems.

Patrick over at STH is losing his mind over it and so many others when really we have nothing concrete, some fancy slide and Raja Koduri doing a similar mentality speech as he would have done for Vega.