All they have to do is use their latest mature node (20A?) to produce and release in volume an Alchemist Arc A790+ with 64GB GDDR6X VRAM for just 400-500$ (with taxes) in the next 3 months, and it'll be flying off the shelves in no time! Is it so difficult?

If they were to sell just 10 million Arc 790+ 64GB cards at a profit of 200 dollars each, they would have made a total profit of 2 billion. They already have the factories, and they continue to pay salaries just to keep them running on idle.

Imagine if AMD brought out a bid too, it'd be great for memes.

unoficialoficial said:

All they have to do is use their latest mature node (20A?) to produce and release in volume an Alchemist Arc A790+ with 64GB GDDR6X VRAM for just 400-500$ (with taxes) in the next 3 months, and it'll be flying off the shelves in no time! Is it so difficult?

If they were to sell just 10 million Arc 790+ 64GB cards at a profit of 200 dollars each, they would have made a total profit of 2 billion. They already have the factories, and they continue to pay salaries just to keep them running on idle.

Click to expand...

That assumes a lot of things, like that the yields for such a large chip would be anywhere decent, or that you could make a 64 GB card for $4-500 with a $200 margin.

Looking at how much 48GB cards cost.....yeah nah mate.

Theinsanegamer said:

That assumes a lot of things, like that the yields for such a large chip would be anywhere decent, or that you could make a 64 GB card for $4-500 with a $200 margin.

Looking at how much 48GB cards cost.....yeah nah mate.

Click to expand...

All these are rough estimation, but the average bulk price for 1 GB of GDDR6X is around $2.30 (source: dramexchange.com, the site list prices for 8 Gb modules in the GDDR spot price section, which is 1 GB).

So, 64 GB costs around $150 in bulk. That means $150 for the rest (mainly the GPU chip, the capacitors are very cheap, a 300 mm wafer have cost ~10K and gives ~100 large chips with 70% yield which is ~100$ per chip) and $200 profit for a final price of 500$. They will print the chip in-house in their own fabs, which can reduce the costs even farther because they already pay for salaries even if they don’t produce anything and the RnD cost have been amortised. So around $100 per chip and $50 for the cooler, packaging and transport.

I think it'll sell well (over 10 million cards in the first 2 years) even at 800$! I don't know exactly what yields they have on this node (20A), but I suppose it's over 70% or else the node isn't approved, 70% is the minimum. They've been doing this for over 4 decades, and they recently bought the most modern extreme ultraviolet machines. If after over 40 years they don't know how to design and print chips efficiently with good yields, who knows?

OMG, this is the end and everyone is smelling the blood....

Axeia said:

Imagine if AMD brought out a bid too, it'd be great for memes.

Click to expand...

It would never past the approval process.

Not to mention, Intel is a sinking ship.

Intel is daft to not consider this. Arm is a much more efficient Instruction set Vs X86 and even if this doesn't happen now in the future intel will have no choice but to make an Arm chip, especially if Arm takes off in windows devices.

Lounds said:

Intel is daft to not consider this. Arm is a much more efficient Instruction set Vs X86 and even if this doesn't happen now in the future intel will have no choice but to make an Arm chip, especially if Arm takes off in windows devices.

Click to expand...

It might but as far I understand imagine repack a game ,watch YouTube in 4k and just browse internet....I don't think arm would perform well in multitasking like these anytime soon.

This is how intel will steer the ship and make it out of the storm. Tales like these make the news. It's against the odds.

But is intel really against the odds? The only major obstacle is AMD. That's it.
TSMC is also a major player but tsmc doesn't have the capacity intel would for its own needs.

The market refreshes itself probably. Intel still has huge revenue.

https://futurumgroup.com/insights/i...e-navigating-challenges-and-strategic-shifts/

Lounds said:

Intel is daft to not consider this. Arm is a much more efficient Instruction set Vs X86 and even if this doesn't happen now in the future intel will have no choice but to make an Arm chip, especially if Arm takes off in windows devices.

Click to expand...

There is no major difference on efficiency ARM vs x86. Just because there are some ARM-architectures that are made only efficiency in mind, does not mean there is no way to make x86 chip with equal efficiency.

There are only two companies that legitimately could and have the motivation to buy Intel.

First is Nvidia, Nvidia has soooo much money now and, if so desired, could pretty much buy all or any part of Intel the wished. This would give them overwhelming dominance in the compute space, especially the server and data centers. My only concern would it pass the monopoly test?

The second company would be Broadcom. They could buy a large portion of Intel, specifically Intel's chip division. This is literally how Broadcom has grown so large over the last decade. It's core model has been aggressive acquisitions of tech companies. However, Broadcom is headquartered in the U.S. but has roots in Singapore, which may be a non-starter for the US government.

unoficialoficial said:

All these are rough estimation, but the average bulk price for 1 GB of GDDR6X is around $2.30 (source: dramexchange.com, the site list prices for 8 Gb modules in the GDDR spot price section, which is 1 GB).

So, 64 GB costs around $150 in bulk. That means $150 for the rest (mainly the GPU chip, the capacitors are very cheap, a 300 mm wafer have cost ~10K and gives ~100 large chips with 70% yield which is ~100$ per chip) and $200 profit for a final price of 500$. They will print the chip in-house in their own fabs, which can reduce the costs even farther because they already pay for salaries even if they don’t produce anything and the RnD cost have been amortised. So around $100 per chip and $50 for the cooler, packaging and transport.

I think it'll sell well (over 10 million cards in the first 2 years) even at 800$! I don't know exactly what yields they have on this node (20A), but I suppose it's over 70% or else the node isn't approved, 70% is the minimum. They've been doing this for over 4 decades, and they recently bought the most modern extreme ultraviolet machines. If after over 40 years they don't know how to design and print chips efficiently with good yields, who knows?

Click to expand...

There's just a lot wrong with this. You havent calculated in the commission from the retailers, or taxes, and you certainly underestimate costs for things like shipping. The site you linked doesnt list GDDR6X memory anywhere and any reference to its price I can find shows it closer to $3 per GB.You also dont include things like R+D costs for designing a new card, the fact that intel doesnt make its own cards (even the intel branded cards were done with AIBs, not intel internally) or the cost of continued support.

Above all else, you are running on this false premise that all intel has to do is sell this mystical A790+ to tens of millions at $500. In case you didnt know, the A770 sales are abysmal, with intel acquiring a 0% market share last quarter for sales. That's a $300 card. Who would pay $500 for a slightly larger intel card with useless amounts of VRAM on it? AI? the competition will destroy said card. That $16GB 770 is way cheaper by all metrics and yet it's not selling at all, what makes you think an A790 would be selling RTX 4070 levels?

All this is also based on the 20A node, a node INTEL THEMSELVES IS NOT USING FOR THEIR OWN CHIPS (did you miss this news?), being able to make such a GPU. Absolutely delusional.

It IS, in fact, that hard to make profits. Which is why we've heard multiple stories of AIBs struggling to turn a profit. I'm sure your degree in marketing with a minor in advanced manufacturing (that you have, right?) lends you to think this is possible, but lets face it, alchemist was a DOA design. Battlemage needs to hit it out of the park or drop out altogether.

Theinsanegamer said:

There's just a lot wrong with this. You havent calculated in the commission from the retailers, or taxes, and you certainly underestimate costs for things like shipping. The site you linked doesnt list GDDR6X memory anywhere and any reference to its price I can find shows it closer to $3 per GB.You also dont include things like R+D costs for designing a new card, the fact that intel doesnt make its own cards (even the intel branded cards were done with AIBs, not intel internally) or the cost of continued support.

Above all else, you are running on this false premise that all intel has to do is sell this mystical A790+ to tens of millions at $500. In case you didnt know, the A770 sales are abysmal, with intel acquiring a 0% market share last quarter for sales. That's a $300 card. Who would pay $500 for a slightly larger intel card with useless amounts of VRAM on it? AI? the competition will destroy said card. That $16GB 770 is way cheaper by all metrics and yet it's not selling at all, what makes you think an A790 would be selling RTX 4070 levels?

All this is also based on the 20A node, a node INTEL THEMSELVES IS NOT USING FOR THEIR OWN CHIPS (did you miss this news?), being able to make such a GPU. Absolutely delusional.

It IS, in fact, that hard to make profits. Which is why we've heard multiple stories of AIBs struggling to turn a profit. I'm sure your degree in marketing with a minor in advanced manufacturing (that you have, right?) lends you to think this is possible, but lets face it, alchemist was a DOA design. Battlemage needs to hit it out of the park or drop out altogether.

Click to expand...

Most people don't think of Intel when it comes to GPUs. Without some kind of spectacular competitive advantage, they will never hear about Intel in GPUs.

Porkous said:

This is how intel will steer the ship and make it out of the storm. Tales like these make the news. It's against the odds.

But is intel really against the odds? The only major obstacle is AMD. That's it.
TSMC is also a major player but tsmc doesn't have the capacity intel would for its own needs.

The market refreshes itself probably. Intel still has huge revenue.

https://futurumgroup.com/insights/i...e-navigating-challenges-and-strategic-shifts/

Click to expand...

No, Intel is facing AMD for CPUs, TSMC for manufacturing and Nvidia for GPUs. And they are losing on all fields.

Intel has no chance in hell. Going against TSMC was their death warrant.

unoficialoficial said:

Most people don't think of Intel when it comes to GPUs. Without some kind of spectacular competitive advantage, they will never hear about Intel in GPUs.

Click to expand...

Then they need extra aggressive marketing, which adds to the total production cost, making the fairy tale A790 look more like a unicorn.

redgarl said:

It would never past the approval process.

Not to mention, Intel is a sinking ship.

Click to expand...

I just want to see it for the fun of it. I don't think its a realistic option although tbf I bet if AMD made a bid on the fabs it would be fine (moving that from one to the other x86 player does no harm).
But yes the design part of Intel would have to go to literally any company that is not AMD (NVIDIA would be problematic as well - every other company is probably fair game).

Intel might be cooking something up and making a comeback though, at least one big enough for their value to rise higher then it currently is. I'd say they're doing the right thing by not selling (yet).

UdyrL said:

Then they need extra aggressive marketing, which adds to the total production cost, making the fairy tale A790 look more like a unicorn.

Click to expand...

You don't understand me. Imagine a company finds a new metal process and builds a car out of titanium alloy, which is lighter than aluminium and steel and stronger than both and has better corrosion resistance etc. Imagine if that company had found a way to give the car's processing unit the ability to have an autopilot at level 5 and running and an llm with 1 trillion parameters. Imagine if that company had found a way to split water into H and O2 with minimal energy, and the car ran on hydrogen and was filled with free water that was split into H and O on the fly inside the car. Imagine this company selling this car at half the price of ordinary cars.

After these achievements, this company only needs to run only one advertisement to show these "spectacular competitive advantages" and the social media will catch fire about this car. People will stay overnight to buy the car. The concept of the car will become synonymous with the company's name. They will call their friends and tell them they have bought a new car, and the answer will be ah from that company.

If a new company comes along and says, you know, ah, I've made a car like Ford, a bit weaker, but also cheaper, ok, it overheats sometimes, but it's cheaper as we said, oh, and it burns some oil, but it's chepar, don't forget, well, that's not convincing. They will buy Ford because it has a reputation and they know it.

HardReset said:

There is no major difference on efficiency ARM vs x86. Just because there are some ARM-architectures that are made only efficiency in mind, does not mean there is no way to make x86 chip with equal efficiency.

Click to expand...

x86 carries the burden of supporting legacy PC applications, so that the SW that rruns 20 years ago still works today. If Intel or AMD were to decide that they will build a brand new x86 tomorrow that only caters to power efficiency without the need for backward compatibility, then I think the efficiency discussion will be very different

dCasualGamer said:

x86 carries the burden of supporting legacy PC applications, so that the SW that rruns 20 years ago still works today. If Intel or AMD were to decide that they will build a brand new x86 tomorrow that only caters to power efficiency without the need for backward compatibility, then I think the efficiency discussion will be very different

Click to expand...

Not very different because many legacy instructions are put on non-sensitive areas on CPU (=are much slower than more widely used instructions) or are even executed on microcode (run even slower). Efficiency loss is very small, most disadvantage comes from need to waste die space. Also depending on how fast that legacy stuff must be, it may cause small performance loss too.

But for efficiency, difference is so small it doesn't make any major difference. Difference comes from fact that most x86 CPUs are designed for high clock speeds and most ARMs are designed for low clock speeds.

unoficialoficial said:

You don't understand me. Imagine a company finds a new metal process and builds a car out of titanium alloy, which is lighter than aluminium and steel and stronger than both and has better corrosion resistance etc. Imagine if that company had found a way to give the car's processing unit the ability to have an autopilot at level 5 and running and an llm with 1 trillion parameters. Imagine if that company had found a way to split water into H and O2 with minimal energy, and the car ran on hydrogen and was filled with free water that was split into H and O on the fly inside the car. Imagine this company selling this car at half the price of ordinary cars.

After these achievements, this company only needs to run only one advertisement to show these "spectacular competitive advantages" and the social media will catch fire about this car. People will stay overnight to buy the car. The concept of the car will become synonymous with the company's name. They will call their friends and tell them they have bought a new car, and the answer will be ah from that company.

If a new company comes along and says, you know, ah, I've made a car like Ford, a bit weaker, but also cheaper, ok, it overheats sometimes, but it's cheaper as we said, oh, and it burns some oil, but it's chepar, don't forget, well, that's not convincing. They will buy Ford because it has a reputation and they know it.

Click to expand...

An A790 is not any of those incredible findings you mentioned. Intel would not be discovering the "golden egg" of the GPUs doing what you propose. Nvidia could easily outdo it in a blink, yet they're not doing it.

UdyrL said:

An A790 is not any of those incredible findings you mentioned. Intel would not be discovering the "golden egg" of the GPUs doing what you propose. Nvidia could easily outdo it in a blink, yet they're not doing it.

Click to expand...

It's not that, but it's a good solid first step towards that. As for the competition's reaction, If Nvidia initially sells the 3090 with 32GB GDDR7 for say $3000 (in reality as a fabless company they can't sell lower because of scalpers even if they wanted to) and Intel comes out with a 128GB GDDR6X model for $1000 that does x3 better in AI inference than the x3 more expensive 3090 and Nvidia suddenly comes out with a 3090 with 128GB model for 2000$ as a reaction, then it will be a loss of face and a loss of reputation for Nvidia that will last as a memory for years. Because people will think that they could have gotten more value for a better price and they didn't just do it to steal their money (and personal money is not like company money which is tax deductible, you fight with your wife to take a few thousand dollars from your kids). So they will never buy another Nvidia product for at least the next 5 years. It's about the same thing that happened in the past with Intel, who kept offering the same product with minor improvements until AMD and TMSC(with some secrete sauce to high yields, maybe the use some kind of plasma (argon oxygen?)to purify the environment) came along to compete. Enthusiasts left Intel and OEMs stayed because Intel could guarantee the volumes they wanted because they had the factories. But after a while, demand fell and the fabs sat idle, burning money. And Intel went into a cycle of negative feedback, and that's why today they buy chips through TMSC, burning money, while their own fabs sit idle, burning money.

redgarl said:

No, Intel is facing AMD for CPUs, TSMC for manufacturing and Nvidia for GPUs. And they are losing on all fields.

Intel has no chance in hell. Going against TSMC was their death warrant.

Click to expand...

Intel is not really loosing in all fields. They dominate laptop and oem market and this market is much bigger than desktop market.

They sold more in datacenter market than AMD last quarter as well. This is not a secret you know. Have a look at their financials

The only market they struggle in is GPU which was expected when you enter a brand new market and start from scratch. They are not giving up and Battlemode and Xe2 looks good.

You root for Intel to die and let TSMC get monopoly!?

However, Intel won't die. Things are looking good, with Lunar and Arrow lake. Next gen datacenter CPUs on own 18A node and next gen GPUs coming in just a few months.

Worst times for Intel are over. AMD has been in a place like this many times before and came back. AMD has been close to bankrupcy at least 5 times in the last 30 years.