News Intel GPUs - Battlemage rumoured cancelled (again)

[adroc_thurston]

How certain are you about it?

yea

Gone for good? Or will it rear its lovely head in 2025?

well it exists, not that BMG is a viable product.

Why are you calling it violent?

See, PPA is a nice triangle.

so Intel has a short window of opportunity to flood the market with cheap Battlemage cards.

Patty G. is really-really not keen on running FCF negative ops so the flood™ is impossible.
It'll be priced to market prices.

 

[igor_kavinski]

well it exists, not that BMG is a viable product.

That's what everyone here wants.

It'll be priced to market prices.

That's fine. Just like AMD prices them and then has to discount them later. Fine by us.

 

[Aapje]

5060 doesn't seem to be on the radar in 2024 so Intel has a short window of opportunity to flood the market with cheap Battlemage cards.

A flood would be dumb unless we somehow get a new mining boom or such. There are not enough buyers unless Battlemage will be some amazing leap ahead, but that's very unlikely.

 

[igor_kavinski]

A flood would be dumb

If multiple models from multiple AIBs are not staring consumers right in their faces, Intel won't get the visibility they need when people are shopping for graphics cards.

 

[Heartbreaker]

The IGPs are perfectly fine for everything but gaming though.

That isn't really a sustainable position if your competition is good for everything including gaming.

They are not perfectly fine to compete with larger future iGPU/APU parts, from AMD and potentially more on the Windows-ARM side.

APU/iGPU competition will scale up in the future (soon rumored AMD Strix Halo), and for Intel to compete in that space they need to have a competitive bigger than current iGPUs.

Intels problem is not that it's some kind of major challenge putting a GPU on a separate board, it's that they don't scale well, and no one ever expected much from their iGPUs.

The Windows PC space is likely going to shift more and more to ARM, especially in laptops, and that bring additional competitors, like Qualcomm, and possible NVidia will re-enter the space.

GPU could become a major point of differentiation for Win-ARM SoCs, and Intel iGPU sucking will no longer be an option if they want to compete.

dGPU is the crucible for Intel GPU tech, that forces them to improve and scale their technology.

 

[NTMBK]

I do wonder how much future the Intel dGPU has, now that Intel are (apparently) committed to spinning off their fabs.



GPUs made a ton of sense as a fab filler - high volume, big dies, need cutting edge process node tech to be competitive. (Talking long term, assuming Intel can get competitive with TSMC again, which is a big assumption.) But an independent fab can just win contracts with AMD and Nvidia, if their process is good enough- why keep propping up uncompetitive designs from Intel in that case? It makes a lot less strategic sense now.

 

[Aapje]

@NTMBK

Of course the idea is that Intel's GPU designs will be competitive, hopefully before the AI boom dies. It would be pretty sad if Intel misses both the mining boom and the AI boom, and gets competitive just as the market craters.

With the company split in two halves, neither half can blame the other anymore. Intel Foundry can claim that their side is great and that Intel Design is the problem, but if Nvidia and other customers also reject Intel Foundry, then its clear that they are failing and refuse to fix their own issues that make others reject them.

And if Intel Foundry makes great products for other customers, but Intel Design products perform poorly, then the design guys need to fix their problems and can't blame the foundry.

 

[Mopetar]

I don't think splitting the company in half fixes any of the fundamental problems that led the company to their present state. If the same idiots are in charge that let both fab and design stagnate, then you have two underperforming halves that realize how much of the problem they were all along.

Has no one considered that they were both holding each other back?

 

[CouncilorIrissa]

See, PPA is a nice triangle.

Still the same level of deficit that ACM had compared to N23?

 

[Aapje]

I don't think splitting the company in half fixes any of the fundamental problems that led the company to their present state.

No, but it makes it so that they can't blame the other side for their own failures.

Has no one considered that they were both holding each other back?

But that is exactly why they are working towards a split, so both of them can and have to show their own qualities, by working with other design teams/foundries.

 

[KompuKare]

Very bad news if true...
...Since I was waiting for Intel to drive some competition so that...
...I can buy any future AMD GPU cheaper of course!

 

[DAPUNISHER]

I am with igor on this one; in denial. I expect it to show up by the holidays. TAP said they were working hard on it. To reuse the joke someone else here made - A wizard is never late, nor is he early. He arrives precisely when he means to. The boys were getting schlumped in goblin town and Gandalf appeared from nowhere and saved the day. Maybe they know who this leaker is and they are intentionally hiding it from them. I'll be back after another glass of copium laced blueberry wine.

 

[reggie_fils_aime]

I also don't quite believe the Battlemage Is Dead rumor. The silicon is clearly already done, it's showing up in leaks here and there, and there's clearly huge issues inside of Intel in terms of communication. You could probably call any one employee there and they'll give you three different answers - AXG Is dead, Battlemage is going to compete with the 4070 Ti, Intel has converted to run on biodiesel, etc.

 

[DavidC1]

I also don't want to believe it but they still haven't got the company out of the slowly-swirling-down-the-drain phase.

Maybe these are just false rumors but if they fail on the Battlemage generation, the future of dGPUs are going to be very murky. The company does not have a history of sticking with something, which might be a big reason why they can't expand beyond their bread and butter. They need to stick through it even if it's very difficult.

 

[igor_kavinski]

The company does not have a history of sticking with something, which might be a big reason why they can't expand beyond their bread and butter.

X86-64 won't last forever. They need to take the necessary steps to prevent their own demise.

 

[Thunder 57]

X86-64 won't last forever. They need to take the necessary steps to prevent their own demise.

x86-128!

 

[igor_kavinski]

x86-128!

They didn't have the courage to even do 64-bits. AMD forced their hand.

 

[Thunder 57]

They didn't have the courage to even do 64-bits. AMD forced their hand.

Ahem, is IA-64 a joke to you? Because it sure is to the rest of us!

 

[igor_kavinski]

Ahem, is IA-64 a joke to you?

No feelings evoked by it. If it didn't happen in client computing, IT DIDN'T HAPPEN!

At least that's how I think

 

[Mahboi]

How certain are you about it? Gone for good? Or will it rear its lovely head in 2025?

Lovely head?
ALM was meant to battle RDNA 2. Came out 6 months before RDNA 3. Sort of.
BMG was meant to battle RDNA 3, if it came out today, it would be less than 6 months before RDNA 4.

It's dead Jim. It's very, very dead.

Why are you calling it violent? I hate it when I have to try to decipher your riddles.

He means that if you pinned BMG vs RDNA 4 or Blackwell, it would be like a toddler meeting an MMA fighter in the ring.

What exactly happened? RT couldn't be the last straw that broke Intel's GPU camel's back.

RT fine, Raster ded. Drivers went from sunk in the ocean to unable to get out of the swamp. Better is still bad.

 

[KompuKare]

X86-64 won't last forever. They need to take the necessary steps to prevent their own demise.

How about creating a "GPU" with the x86 everywhere philosophy!

Because the 8086 was such a great chip, and getting lucky with IBM PC gig had no bearing whatsoever!

Intel must have been reading "computer history according to Intel" when they started on Larrabee, or were just mad.

 

[Hitman928]

How about creating a "GPU" with the x86 everywhere philosophy!

Because the 8086 was such a great chip, and getting lucky with IBM PC gig had no bearing whatsoever!

Intel must have been reading "computer history according to Intel" when they started on Larrabee, or were just mad.

Don't jinx us, Intel's current CEO was the one spearheading the Larrabee project when he was there the first time as a VP and still believes that it would not only have been successful, but would have allowed Intel to be a leader in AI if Intel hadn't pushed him out and killed off the project.

Intel CEO Pat Gelsinger candidly reveals where Intel dropped the ball in recent years

Mistakes and what-ifs.

www.pcgamer.com

 

[igor_kavinski]

Lovely head?

I would buy the A770 today but I want V2.0 of Intel dGPU.

He means that if you pinned BMG vs RDNA 4 or Blackwell, it would be like a toddler meeting an MMA fighter in the ring.

MMA fighter wouldn't lay a finger on the toddler. Analogy failed. Try again

RT fine, Raster ded. Drivers went from sunk in the ocean to unable to get out of the swamp. Better is still bad.

Referring to any particular instance? They are still releasing regular drivers with fixes.

 

[GodisanAtheist]

Don't jinx us, Intel's current CEO was the one spearheading the Larrabee project when he was there the first time as a VP and still believes that it would not only have been successful, but would have allowed Intel to be a leader in AI if Intel hadn't pushed him out and killed off the project.

Intel CEO Pat Gelsinger candidly reveals where Intel dropped the ball in recent years

Mistakes and what-ifs.

www.pcgamer.com

-Yikes!

I ask again how does someone manage to swing a CEO gig?

Cause I can kill a company twice as fast for half as much as most of these guys seem to...

 

[DavidC1]

Lovely head?
ALM was meant to battle RDNA 2. Came out 6 months before RDNA 3. Sort of.
BMG was meant to battle RDNA 3, if it came out today, it would be less than 6 months before RDNA 4.

It's ACM, not ALM.

2x performance and perf/watt along with many minor fixes existing in Alchemist will make it a far better product. Chips and Cheese article convinces me that lot of the seeming driver related problems will either be significantly mitigated or make it easier to fix by drivers with the new hardware. Previous Intel GPU problems thought to be driver related turned out to be hardware after all.

-Imbalanced architecture
-Idle power management issue
-ReBar performance and compatibility hit

Just think of ReBar. They are aiming Alchemist at "cheap systems", yet is viable in only relatively newer ones. I could stick my RX 480 in my Core 2 system no problem. How silly is that? The A380 is a mere $170 CDN, yet I need to go buy a motherboard, CPU, and new RAM to make it work properly.

Idle power isn't that big of an issue, but certainly there are imbalances in the architecture, hence horrible performance in some games, and even things like why it doesn't perform well in lower resolutions.

Even in die and power use it'll be far more competitive.

Alchemist:
-400mm2 N6 die performing like Samsung 8N 270mm2
-No comparison for power use. Turing generation equivalent at best.

To

Battlemage -
-400mm N4 die performing like 300mm2 N4 from Nvidia.
-Huge power efficiency gains. This is the generation where Ashraf Eassa said were aiming for "leadership perf/watt". With the delays it may not be #1, but may be in the ballpark.