News Intel GPUs - hot damn, a price cut!

[igor_kavinski]

Let's hope raytracing performance is better than AMD's.

 

[GodisanAtheist]

Intel Arc Update: Alchemist Laptops Q1, Desktops Q2; 4mil GPUs Total for 2022

www.anandtech.com

4 Million Units in '22 including Laptops/Workstations/AIBs is basically nothing (most of it will go to laptops from the sounds of it).

Smart business strategy (bootstrap GPUs by attaching them to Intel's locked in segments) but not great for DIY people.

Put all expectations of Intel bringing any relief to the GPU market at all this year on total pause. According to Ryan, NV/AMD shipped 40 million GPUs in 2021.

 

[igor_kavinski]

According to Ryan, NV/AMD shipped 40 million GPUs in 2021.

How much of that was unusable low end crap?

 

[GodisanAtheist]

How much of that was unusable low end crap?

- I'd wager not much given NV/AMD were pretty deep into their new line of cards for most of the year and didn't even really have low end options to manufacture and ship.

Even if hypothetically 50% of that was low end crap, that would make it 20 million AIB cards (not Laptop/Workstation class cards) shipped out vs 4 million Laptop/Workstation/AIB total.

The point being that Intel is going to be just as if not more scarce than NV and AMD have been over the last year.

 

[Grabo]

Intel Arc Update: Alchemist Laptops Q1, Desktops Q2; 4mil GPUs Total for 2022

www.anandtech.com

4 Million Units in '22 including Laptops/Workstations/AIBs is basically nothing (most of it will go to laptops from the sounds of it).

Smart business strategy (bootstrap GPUs by attaching them to Intel's locked in segments) but not great for DIY people.

Put all expectations of Intel bringing any relief to the GPU market at all this year on total pause. According to Ryan, NV/AMD shipped 40 million GPUs in 2021.

I'm still optimistic and believe what MLID said a couple of months or one ago, like I mentioned prior in this thread that maybe INTC will make a market impact in December.
Looks like kernel 5.18 will be receiving quite a lot of Alchemist attention ..still, it will likely have been available for half a year to a year by the time anyone can predictably get a hold of an Alchemist at a reasonable price.

 

[Ajay]

Intel Arc Update: Alchemist Laptops Q1, Desktops Q2; 4mil GPUs Total for 2022

www.anandtech.com

4 Million Units in '22 including Laptops/Workstations/AIBs is basically nothing (most of it will go to laptops from the sounds of it).

Smart business strategy (bootstrap GPUs by attaching them to Intel's locked in segments) but not great for DIY people.

Put all expectations of Intel bringing any relief to the GPU market at all this year on total pause. According to Ryan, NV/AMD shipped 40 million GPUs in 2021.

Wow, so only a 10% increase in total GPU supply. I seem to recall that AMD/NV will be manufacturing more GPUs in 2022 as well - so things are looking up a bit.
If I can find a reasonably priced card that can keep 95% of my frames above 60fps (1440p), I'll be happy.

 

[IntelUser2000]

Intel Arc Update: Alchemist Laptops Q1, Desktops Q2; 4mil GPUs Total for 2022

www.anandtech.com

4 Million Units in '22 including Laptops/Workstations/AIBs is basically nothing (most of it will go to laptops from the sounds of it).

4 million for a newcomer is very respectable. Even the iPhone with amazing growth only sold 1.5 million in the first year.

Also note there are *zero* ARC using devices that you can buy. Essentially you can assume the first quarter is not relevant to volume at all.

Also 4 million+ units for Intel represents all brand new products, while AMD/Nvidia has lots of previous generations and even archaic ones. If they can do 4.5 million this year, next year it might be 6 million just because it'll have full 12 months exposure, and maybe 8 million due to growth.

AMD in Q4 2021 had 17% GPU marketshare. At 50 million total units, 4 million units is already half of what AMD sells in total! This means with 8 million units Intel might equal the size of AMD dGPU share in a span of mere 18 months!

That's why giving up a product line in 3-4 years is asinine. It takes 3-4 years to take foothold assuming you execute.

 

[gdansk]

That's not bad at all. AMD is shipping less than 3 million GPU per quarter according to JPR. Given the low Q1 shipments and typical ramp Intel is likely to be at half of that by 22Q4.

 

[IntelUser2000]

That's not bad at all. AMD is shipping less than 3 million GPU per quarter according to JPR. Given the low Q1 shipments and typical ramp Intel is likely to be at half of that by 22Q4.

Yea, I don't know why people expect shipping millions of units in an established market is something that you can turn around on a dime.

By the way, Intel's 90%+ marketshare in servers took 10+ years to get there. We think that it took them like a year right? No, because 10 years ago the x86 server share was quite small.

This kind of growth is what most companies will kill for. Taking possibly 10% of the market is devastating to existing vendors. No wonder there are rumors of Nvidia wanting to go all-out in anti-competitive practices to lock out Intel. They worked for decades to get there and a newcomer(even if it's Intel) takes 10% of your pie in the first year with not a complete product line?

 

[Karnak]

4 million Units is just bad. Really bad. That's literally nothing.

Even if you're going by ~3 million/q for AMD that's about 12 million a year. And here we're talking about dGPUs only. No workstations, no mobile and no HPC/AI Instinct stuff.

Intels 4 million are for EVERYTHING. And I can guarantee you that they will prioritize mobile by a lot.

 

[IntelUser2000]

4 million Units is just bad. Really bad. That's literally nothing.

Sorry but this is a very misinformed line of thought. The enthusiast forum posters take mass production for granted. Note the word "Mass". It took Nvidia 30 years to get to the position they are in today. Shareholders will scream for 5% revenue drop.

And Intel might take 10% in the first year, a company with basically no credibility in the gaming market. If that kind of ramp continues year over year Nvidia should be scared.

 

[moinmoin]

I think whether 4 million is seen as little or much depends on whether one believes the current supply is enough to satisfy the current demand. My impression has been that supply had been nowhere close to satisfying the demand in the last couple years, so 4 million does seem little. A company of Intel's size should be able to push more if it really wanted, even for a "first product". But then there is also the question how much of the demand so far was crypto, and what happens when that ceases to exist.

 

[Karnak]

And Intel might take 10% in the first year,

Curious how that works given the fact we're talking about 4 million units for Intel's ENTIRE GPU business in 2022.

So like what? Maybe 2 million units for the entire desktop market in 2022? How do u want to reach 10% with that?

 

[IntelUser2000]

Curious how that works given the fact we're talking about 4 million units for Intel's ENTIRE GPU business in 2022.

So like what? Maybe 2 million units for the entire desktop market in 2022? How do u want to reach 10% with that?

The combined dGPU shipment for 2021 is about 50 million. And Q1 for Intel is a non-existent quarter. They also said 4 million+. That in 8-10 months is roughly 10%. How much of Nvidia's shipments are based on the archaic GT710s and the endlessly rebranded MX mobile GPUs?

That's not even my point anyways. It really doesn't matter whole lot for the first year whether it's 5 or 10%. That's a significant loss for existing players and a big gain for the newcomer.

And like it was pointed out numerous times before it's already half of AMD's dGPU size, in a worst case scenario since volume will continue to ramp quarter over quarter for a newcomer in the first year.

Again, the original iPhone shipped only 1.5 million units. You need years of consistent execution to take decent marketshare. There's no such thing as entering a new market and suddently taking 25% overnight.

A company of Intel's size should be able to push more if it really wanted, even for a "first product".

This is true in a sense but Intel has almost no credibility in the GPU market. So unless they are completely lying and are shipping 4 million but only quarter are being built, they still need to do immense amount of work to get the AIBs to use it and the partners to ship them in systems.

The AIBs and retailers aren't completely braindead either. Remember Lakefield? How many systems were featured with that CPU? 2? And 1 that actually sold? What about Kabylake-G? They sucked so no one used them Intel or not.

Perhaps it won't pan out the way they say it will but taking the numbers by face value I'd say 4 million for the 1st year is good to darn impressive.

In reality they only have slightly more power to influence the market.

 

[gdansk]

I suppose it's a matter of expectations. Ramping up to 40-50% of one of the entrenched players in about 9 months seems pretty good to me.

But if you were expecting big market pricing impacts it will disappoint. People just want to buy Nvidia in the end and Intel can't change that anytime soon.

 

[eek2121]

Intel Arc Update: Alchemist Laptops Q1, Desktops Q2; 4mil GPUs Total for 2022

www.anandtech.com

4 Million Units in '22 including Laptops/Workstations/AIBs is basically nothing (most of it will go to laptops from the sounds of it).

Smart business strategy (bootstrap GPUs by attaching them to Intel's locked in segments) but not great for DIY people.

Put all expectations of Intel bringing any relief to the GPU market at all this year on total pause. According to Ryan, NV/AMD shipped 40 million GPUs in 2021.

Wow, so only a 10% increase in total GPU supply. I seem to recall that AMD/NV will be manufacturing more GPUs in 2022 as well - so things are looking up a bit.
If I can find a reasonably priced card that can keep 95% of my frames above 60fps (1440p), I'll be happy.

GPU prices are heading down as it is. Add 10% more GPUs to the mix and prices will come down faster.

 

[DrMrLordX]

Intel probably isn't getting much N6. They're having to spread it out between Arc and Ponte Vecchio as well.

 

[Frenetic Pony]

That's not bad at all. AMD is shipping less than 3 million GPU per quarter according to JPR. Given the low Q1 shipments and typical ramp Intel is likely to be at half of that by 22Q4.

I expect that's mostly due to extreme demand for consoles. They probably have some sort of contractual priority for building those which have shipped in the tens of millions, and combined with shortages, well you get relatively few GPUs despite demand there as well.

As a complete aside, looks like Battlemage is indeed at least partially on Intel 4. Didn't see anything about desktop being there as well as laptop, but it wouldn't be the most surprising.

 

[IntelUser2000]

As a complete aside, looks like Battlemage is indeed at least partially on Intel 4.

Uh, for Meteorlake it's the compute tile that's on Intel 4. The GPU is using TSMC's process.

 

[psolord]

4 million gpus in 2022, are 4 million gpus MORE, than what would have existed otherwise. So I'LL TAKE IT!

 

[NTMBK]

Intel Goes Full XPU: Falcon Shores to Combine x86 and Xe For Supercomputers

www.anandtech.com

Falcon Shore sounds like a HPC APU, built with tiles. What are expecting, HBM3 local memory backed by DDR5, similar to the HBM2e version of Sapphire Rapids? Could be interesting tech for a high performance gaming laptop- take something like one CPU and one GPU tile, and put a cutdown version into a ~100W total TDP.

 

[Frenetic Pony]

Uh, for Meteorlake it's the compute tile that's on Intel 4. The GPU is using TSMC's process.

Ok so they only specify CPU tile being taped out on Intel 4, missed that. But can't find any specifics on the GPU tile. The slide I posted certainly seems to suggest integrated graphics is on Intel 4, but another suggests some part of Meteor Lake or Lunar Lake is also going to be on TSMC 3nm. I don't see any specifics though?

 

[NTMBK]

Ok so they only specify CPU tile being taped out on Intel 4, missed that. But can't find any specifics on the GPU tile. The slide I posted certainly seems to suggest integrated graphics is on Intel 4, but another suggests some part of Meteor Lake or Lunar Lake is also going to be on TSMC 3nm. I don't see any specifics though?

If they aren't shouting about the Intel process they're using, you can bet it's not on Intel.

 

[Leeea]

Let's hope raytracing performance is better than AMD's.

That is very unlikely.

 

[ultimatebob]

4 million gpus in 2022, are 4 million gpus MORE, than what would have existed otherwise. So I'LL TAKE IT!

Doesn't Intel have their own fabs, though? They should be able to pump out much more product than this if they really wanted to.