News Intel 2Q24 Financial Results

[NTMBK]

Intel’s Falcon Shores Future Looks Bleak as It Concedes AI Training to GPU Rivals

Intel’s Falcon Shores future looks bleak as it concedes AI training to GPU rivals On Monday, Intel sent a letter to employees detailing its comeback plan after an abysmal second-quarter […]

www.hpcwire.com

Ooooh crikey, they surrendered the GPGPU race. That's bad. Like real bad.

Intel and giving up on HPC GPUs, name a more iconic duo

 

[cebri1]

I don't think can compete with NVIDIA and AMD in the space anyway. If I were them I'd try to released consumer and pro products with a lot of RAM so that is appealing for AI enthuast at home. And try to build a community that may help you when you try to move to higher level. Ofc right now the focus cannot be on a market where you are at such disadvantage and with close to zero market share.

 

[adroc_thurston]

Intel and giving up on HPC GPUs, name a more iconic duo

This is a bit extra hilarious, since

I don't think can compete with NVIDIA and AMD in the space anyway

They've been trying to enter and corner ML training market since 2016 and their acquisition of Nervana.
A bunch of Nervana products (very dead), then GPGPU stuff (also very dead!), then Habana, then back to GPGPU stuff and now again, all dead.
Delightfully schizophrenic ability of not being able to focus and commit to a single roadmap (granted, PVC being an utter and absolute trainwreck didn't help either).

AI enthuast at home

Who

And try to build a community that may help you when you try to move to higher level

They do not have the money for vanity s/w efforts anymore.

 

[NTMBK]

This is a bit extra hilarious, since

They've been trying to enter and corner ML training market since 2016 and their acquisition of Nervana.
A bunch of Nervana products (very dead), then GPGPU stuff (also very dead!), then Habana, then back to GPGPU stuff and now again, all dead.
Delightfully schizophrenic ability of not being able to focus and commit to a single roadmap (granted, PVC being an utter and absolute trainwreck didn't help either).

And that's skipping the decade of Larrabee Xeon Phi failure

 

[turtile]

And that's skipping the decade of Larrabee Xeon Phi failure

It sounds like Pat is still obsessed with Larrabee and wants to see the idea return. I don't see this turning out well for Intel...

 

[gdansk]

It sounds like Pat is still obsessed with Larrabee and wants to see the idea return. I don't see this turning out well for Intel...

It's where all the growth is. If only the CEOs before had been as obsessed maybe they could have done something other than PVC. Xeon Phi had added neural net instructions a decade ago and had a large memory capacity, it was genuinely on the right track.

 

[jpiniero]

It's where all the growth is. If only the CEOs before had been as obsessed maybe they could have done something other than PVC.

That was why they bought Nervana and then Habana. Nobody cared.

TBF, it's not entirely clear that people buying AMD are even using them for anything beyond selling AI hype to "Investors".

 

[gdansk]

That was why they bought Nervana and then Habana. Nobody cared.

TBF, it's not entirely clear that people buying AMD are even using them for anything beyond selling AI hype to "Investors".

Nobody cared at the time. But had they continued developing Xeon Phi they would have had a product, like MI, that could have been iterated into a ML accelerator by now. And it would have had some pre-existing soft support.

Leaving ML now is prudent but it is forced move. Leaving ML ten years ago was also prudent at the time but in retrospect left them unable to pivot to compete with measly AMD nor even approach Nvidia.

 

[DrMrLordX]

Nobody cared at the time. But had they continued developing Xeon Phi they would have had a product, like MI, that could have been iterated into a ML accelerator by now. And it would have had some pre-existing soft support.

Oddly enough, products like Sierra Forest etc. seem like semi-successors to Phi. All they're missing are AVX512 and quad SMT.

 

[Doug S]

Intel’s Falcon Shores Future Looks Bleak as It Concedes AI Training to GPU Rivals

Intel’s Falcon Shores future looks bleak as it concedes AI training to GPU rivals On Monday, Intel sent a letter to employees detailing its comeback plan after an abysmal second-quarter […]

www.hpcwire.com

Ooooh crikey, they surrendered the GPGPU race. That's bad. Like real bad.

Yeah Intel "surrendered" the GPGPU race in the same way I've surrendered in the NYC marathon.

Intel's GPU efforts have been plagued by software/driver issues since the early days. They have a (not great but) credible iGPU, and every time they've tried to move beyond that into big boy territory it has been a joke. Giving up in a race where the leaders were so far ahead you can't even see them is not "surrender", its admitting reality.

Intel no longer has the luxury of pouring billions into projects with zero certainty of commercial success. They have to make the hard decisions, and as hard decisions go this is an easy one.

You want to see some other hard decisions? They need to divest Altera and Mobileye. Intel needs to focus on what's important, not the forever niche idea of FPGAs in servers and a long term play in autonomous vehicles. They need the capital those would free up more than they need to own either.

 

[Doug S]

It's where all the growth is. If only the CEOs before had been as obsessed maybe they could have done something other than PVC. Xeon Phi had added neural net instructions a decade ago and had a large memory capacity, it was genuinely on the right track.

Yes that's where all the growth is today, but holding onto a product you want to believe will maybe possibly be competitive a few years from now doesn't get you a piece of that growth. It maybe possibly gets you a piece of it a few years from now - but only if the world hasn't moved on by then!

Honestly if Intel was interested in that market instead of trying to build something themselves in the same old way as Nvidia they would have been better off to have acquired a startup doing things differently like Cerebras or Groq. Either/both are the only companies with a chance of hell of catching Nvidia, but they have been under resourced especially from an ecosystem standpoint. Who knows what either could have done if they had Intel's full backing (I don't know either, this is just a hypothetical, but I think it is safe to say they would have done a helluva lot more than whatever Intel's latest GPGPU product name of the year is)

 

[jpiniero]

Oh it just dawned on me what's going to happen.

Intel will fully spin off the fabs, and sell most of it to Private Equity and then IPO it. PE will do what they do, and maybe Intel can sucker some "Investors" with the IPO.

The Foundry will implode as it will be saddled with all the debt and no customers (since Intel will eventually move all their products to TSMC and/or Samsung).

Maybe the US Gov will bail it out if there is any point.