News Raja Koduri Explains Why Intel's Outsourcing GPU Manufacturing to TSMC

[Admin]

Intel's Raja Koduri explains the reasoning behind the choice to have TSMC manufacture Intel's ARC graphics chips. It's about production capacity.

Raja Koduri Explains Why Intel's Outsourcing GPU Manufacturing to TSMC : Read more

 

[hotaru.hino]

Plus if ARC is a flop for some reason, their bread and butter manufacturing isn't affected.

 

[evdjj3j]

I can't believe Intel is still trying to run this con.

 

[evdjj3j]

I like how the author uses AMD as an example since Intel CPUs all have GPUs the size difference between an Intel CPU and a GPU is not nearly as dramatic.

 

[peachpuff]

Reason intel won't make gpus in their fabs "our xeon chips are more profitable".

 

[Eximo]

Big GPUs, and we know Intel 10nm yields took a long time to dial in. TSMCs more mature node makes more sense from that perspective alone.

And I agree, the potential for a complete market failure is not unknown to Intel. Having this built on someone else's fab means they lose a lot less if it completely fails to gain any market traction. Which this first round may have a lot of trouble with.

Unless they get the drivers right this could all be a waste of time.

Though like what AMD seems to be planning, slapping one of those dies into a true Intel APU would not be beyond them (heck they might have that in mind, Intel has used Vega once and it was pretty decent) Intel 8+4 + 128 EU, that would be something else.

 

[ezst036]

moving GPU production to TSMC means Intel will be able to produce more devices in a silicon-hungry market while simultaneously assuring its competition can't make as many chips as they could sell. That's one way to beat the competition.

This.

Intel has huge amounts of clout which makes it very difficult for TSMC or anybody else to simply tell them no. In the short term, sending Arc over to TSMC is like an 8x win-win-win for Intel. Even if Arc is an under-performer in the short term this will end up constraining GPU availability even further ensuring that every single Arc sells out. The one that's available is the one you'll buy - somebody will.

In the long run once Intel gets their fabs up to date and good enough to build GPUs this will be a good thing and free up massive amounts of GPU manufacturing capacity.

In the mean time, its a good thing that Nvidia (and seemingly AMD is following now) is using failed GPUs re-spun as miner cards instead of simply throwing them in the trash. Hopefully Intel eventually does the same and we get three brands of miner cards out of it all instead of just CMP HX.

 

[zodiacfml]

more likely supply chain efficiency. TSMC already does this GPU fab work for others that it is trivial. Doing it the other way, Intel making GPUs while outsourcing some CPUs to TSMC, is simply more effort.

 

[JayNor]

Intel is building Xe-HP and Xe-LP in its own fabs, so it doesn't seem to be solely a GPU/non-GPU thing.

The decision to build Xe-HPG at TSM was probably made three years ago, before Intel had their 10nm yield issues solved.

I say "solved" because Intel stated recently that they are now building more 10nm chips than 14nm. Also, the 10nm chips they are building include server chips with relatively high core counts.

Intel builds other chips at TSM. They have made statements about 20% of their volume previously. We know about the eyeq5, habana gaudi and perhaps the easic chips. Now that they are more open about the GPUs being built there, maybe there will be more visibility about the other products.

 

[InvalidError]

Plus if ARC is a flop for some reason, their bread and butter manufacturing isn't affected.

My prediction is that Alchemist will be the Broadwell of GPUs: it exists on paper but few beyond select OEMs will actually have any for sale. Battlemage might be Intel's Skylake GPU that is actually worth having and can be obtained without jumping through hoops.

 

[Co BIY]

Basically a rational allotment of the total marketplace production capacity to the most profitable products.

No fanboys at TSMC or Intel - Hardnosed Calculators.

 

[Chung Leong]

The move should help Intel silence calls by activist investors to split the company. Profit margins on the outsourced Xe-HPG will look abysmal compared to the profit margins on the in-house Xe-HPC and Xe-HP.

 

[InvalidError]

The move should help Intel silence calls by activist investors to split the company. Profit margins on the outsourced Xe-HPG will look abysmal compared to the profit margins on the in-house Xe-HPC and Xe-HP.

That doesn't really make any sense: HPG is a byproduct of Intel's HPC ambitions, a way to seed the platform to developers out there and possibly turn a profit along the way. Without it, Intel would need to waste about just as much time and resources making a low-end dedicated HPC SKUs for training and software development purposes, increasing the barriers to entry for HPC that much further. Kind of the same way that Intel's consumer x86 CPUs are a software development vehicle to upsell Xeons when you need to scale beyond consumer stuff.

HPG as a stand-alone company that has to duplicate all of the HPC R&D along with everything needed to support a company without datacenter and HPC sales prospects, CPU sales, existing customer base, etc. would have a very hard time becoming viable on its own. With all rumors pointing toward next-gen's massive performance gains coming with similarly substantial MSRP increases, GPUs are about to become really profitable if you can make anything decent by contemporary standards. As long as Intel manages to get HPG up to speed, the margins should be similar to its mid-range (i3-i7) consumer CPUs.

Worst case, Intel is no stranger to spending billions trying to break into new markets and then having to eat the exit loss because it isn't happening.