News Intel GPUs - Falcon Shores cancelled

[Hans Gruber]

64 would not be needed as 32 Xe2 cores with 3GHz clock and game improvements due to architecture is enough to reach 4070S. 64 would mean 2x on top of that, which I don't believe they'll reach that on N5, especially with just 1.5x perf/watt improvement.

In fact, this would turn out to be an opposite of RDNA3 where the high level specs seem weak but they actually spent enough transistors to outperform it's high level specs.

XFX Radeon RX 7900 XTX Magnetic Air Review

The XFX Radeon RX 7900 XTX Magnetic Air comes with user-replaceable fans that pop out without any tools. It's a fantastic solution to clean your card and to simplify RMAs. The card achieved excellent performance results in our review, and the overclocking potential is highly impressive.

www.techpowerup.com

Lower resolutions/settings being on the level of other cards would indicate a ~10% improvement where the A770 would be on the 3060 Ti level. 25% improvement on top of that with clocks on Xe2 would be 4070S.

Assuming clocks are increased without exponential increases, 2x that would require going from 225W to something like 700W assuming GPU takes up about 160W, and way far past even 4090's power requirements. Yes it would also greatly outperform 4090 but it doesn't seem realistic.

Regarding Arrowlake: https://cdn.mos.cms.futurecdn.net/o4vQofD6ZXVJ43gfCqZXWU-1200-80.png.webp

On the older roadmaps the Alchemist+ parts were referred to as "G20". It would fit with the current leaks that it's Meteorlake's GPU with XMX and more caches added.

There is no point in making comparisons or assimilating what performance of Battlemage should be. Intel already admitted that ARC was a flawed/broken architecture from the design. They did a good job using software drivers to work around the broken aspects of the 1st generation Intel card. They said Battlemage would fix all the problems in the architecture that plagued ARC. Originally ARC was said to be on paper a bit better in theory to a 3070 and closer to a 3070ti in performance. Because of the hardware flaws that level of performance was never realized.

The real questions are what 5nm TSMC silicon will Intel use for Battlemage. Will it be N4 or N4P or N4X. Then compare the efficiency and power gains from N6 (Arc) to Battlemage. The other question is what performance to expect. I have heard anything from 4070 performance to 4070ti performance for the top Battlemage SKU. The other factor to consider is the GDDR6 memory and the speed of the memory Intel will use. The faster the better.

 

[DavidC1]

There is no point in making comparisons or assimilating what performance of Battlemage should be.

Sure we can. Because we have preliminary Lunarlake data from leakers and Intel themselves. The GPU moved from MTL's N5 to LNL's N3B. This is likely similar from N6 to an N4 variant. Before Lunarlake, it was nearly pure speculation.

Originally ARC was said to be on paper a bit better in theory to a 3070 and closer to a 3070ti in performance. Because of the hardware flaws that level of performance was never realized.

I'm not sure if 3070(nevermind a 3070 Ti) was even possible. AMD rarely catches up to Nvidia in efficiency in that regard either. And their mess ups are far less.

Some people/groups are simply better/smarter than others. The performance in 4K is what I see as max for A770, a 3060 Ti. It is difficult to accept a company wholly focused on an entirely different market segment with HD Audio mentality on top of that to be on par with a product from a company with peak in execution and record employee satisfaction and high retention rates among it's engineers.

Obviously "leakers" like MLID don't know anything in regards to what they are talking about. Sad reality we live in today, where the modern Press has almost zero knowledge about the news they are releasing.

 

[DavidC1]

Exist50 says that G31, which will be at 4070/70ti level of performance is mid-2025! The lower end G21 comes in 2024.

4070/ti performance is ok, but mid-2025 not really.

He also says that it's Celestial that'll make the leap to x80 level of performance and fix things such as compatibility issues much more. This means that there's still going to be a lot the driver needs to make up for. Looks like they might have missed their targets to reach "Enthusiast" with Battlemage.

The release is on a 2-year cycle ever since cost-cutting measures. So Celestial is a 2026-2027 part?

 

[Leeea]

I am just happy Intel is still in it. That they did not give up and walk away.

 

[poke01]

I am just happy Intel is still in it. That they did not give up and walk away.

They can't leave mid way, what else would Intel use for their iGPU IP in their laptop SoC's?

SoC's are the future and the GPU is now more important than ever. As long as the ARC lives on, even in laptop form its good. They can always fund more dGPUs every year when the company is in a better situation.

Edit: fixed grammer

 

[DAPUNISHER]

Intel already has branding on their side, as it is a household name. Gandalf/Battlemage video card day one reviews will be all about the pricing, I predict. I expect drivers to be solid with ARC's failings fixed. The ray tracing and hardware upscaling will extend its lead over AMD. They only need offer the same or more ram as the other 2, with more fps per dollar, and the tech press will enthusiastically recommend Intel.

The media capabilities are already excellent. ISV support will keep getting better. The high volume tiers are ripe for the picking, as I think most gamers are tired of only having two choices. Raster and lower pricing, or raytracing and upscaling superiority for a bit of a premium. Enter player 3 with products that blend all of it at the right price, with strong endorsements from reviews, and good things should happen.

 

[DavidC1]

The cancellation of G10 and the late release suggests to me that they had problems getting Battlemage out and they went through at least a significant silicon revision, if not a design change.

G10 was originally projected to be faster with 56/64 Xe cores and higher clocks, not G31's 32 Xe combined with better perf/shader.

Alternately maybe they went with 32 Xe cores to reduce silicon costs. My hopes are then it's a 300mm2 die so they can justify selling at A770 like prices for longer. If it's going against Geforce 5xxx GPUs, then it'll be x60 Ti level again.

 

[cebri1]

Best hope for Intel right now is that according to the rumors, RX8000 series is a mere refresh with better power consumption and that the 5xxx series mid range is probably 12 months away. They will have a good window of opportunity of at least 7-10 months to be a viable option in the mid upper range (4070S-80 series).

XeSS kicking FSR ass in terms of image quality is another thing they could have going on their side.

Edit: 4nm https://www.techpowerup.com/324197/intel-arc-xe2-battlemage-discrete-gpus-made-on-tsmc-4-nm-process

 

[Mopetar]

Really they just need to focus on having a good launch with drivers in good shape. Expecting them to take the industry by storm is going a bit to far, but leaving everyone with the impression of capable execution means that even if someone doesn't consider you this time, they'll still have reason to in the future.

It's a little bit odd because if I told myself ten years ago that a decade from now I might have a computer with an AMD CPU and an Intel GPU I'm not quite sure how I'd process that information.

 

[igor_kavinski]

It's a little bit odd because if I told myself ten years ago that a decade from now I might have a computer with an AMD CPU and an Intel GPU I'm not quite sure how I'd process that information.

Technically, Intel's design has its roots in AMD's GPU design philosophy: great specs on paper, need to wait for the drivers to mature to get the most out of the design. Nvidia is boring. There is almost no drama regarding the synergy between hardware and software. Everything works, as long as you don't mind paying through the nose, dealing with their crappy Geforce experience software and knowing that they will stop caring about your model the moment they launch the nextgen.

Personally, I would get a mid-range Intel or AMD to enjoy current gen graphics features and keep a previous gen used Geforce as backup.

 

[poke01]

Geforce experience

the new Nvidia App is much better, the old geforce app needs to go away.

Nvidia is boring. There is almost no drama regarding the synergy between hardware and software

True, but boring means stable and that is very important to gamers and pros, this goes a long way in creating mindshare.

Expecting them to take the industry by storm is going a bit to far

Intel just needs 4070/S perf at $350 to $400 with 16GB vram and they will break AMD.

 

[igor_kavinski]

the new Nvidia App is much better, the old geforce app needs to go away.

Thanks, didn't know that existed but still in beta.

 

[coercitiv]

It's a little bit odd because if I told myself ten years ago that a decade from now I might have a computer with an AMD CPU and an Intel GPU I'm not quite sure how I'd process that information.

I'm having issues processing it today, let alone a decade ago.

 

[igor_kavinski]

The awesome thing is that my local Amazon shows the A770 as the cheapest 16GB card, almost a $100 cheaper than 7600 XT. And when they release their AI playground software, it could become a nice entry level AI/gaming gift to friends and family. That Intel would enable such a reality, is pretty unreal.

 

[beginner99]

Intel just needs 4070/S perf at $350 to $400 with 16GB vram and they will break AMD.

I mean AMD isn't asleep and not too far away from these specs. But yeah if that happens and the intel card is actually "good" as in can do basic AI tasks like stable diffusion then AMD might be in trouble.

I myself while AI would be cool to play around, as i play older games I would likely not risk it these older games not working properly or at all.

 

[Hans Gruber]

It would be nice to know which 4nm process Intel will be using. Will it be N4 or N4P or even N4X. I know TSMC has been shifting their customers to N4P while waiting for 3nm silicon to become available. Intel should have big efficiency gains over N6 regardless of the process they are using for Battlemage.

 

[igor_kavinski]

I myself while AI would be cool to play around, as i play older games I would likely not risk it these older games not working properly or at all.

Some games could still work if Intel gives an option in their control panel to have the card pretend to be AMD or Nvidia

I think the issue is that a lot of older games have two separate codepaths for AMD/NVIDIA and when they encounter unknown card, they crap out coz the developer didn't have the foresight to implement a generic path.

 

[NTMBK]

Intel already has branding on their side, as it is a household name.

Yeah, about that... https://www.tomshardware.com/tech-industry/amd-now-has-better-brand-recognition-than-intel

 

[DAPUNISHER]

Yeah, about that... https://www.tomshardware.com/tech-industry/amd-now-has-better-brand-recognition-than-intel

"“isolates the value generated by the strength of the brand alone in the minds of consumers.”"

Damn, there's another wager I would have lost.

 

[Ghostsonplanets]

https://twitter.com/x/status/1810553552311611569

It would be nice to know which 4nm process Intel will be using. Will it be N4 or N4P or even N4X. I know TSMC has been shifting their customers to N4P while waiting for 3nm silicon to become available. Intel should have big efficiency gains over N6 regardless of the process they are using for Battlemage.

N4P is the rumor

 

[KompuKare]

Would be a fun comparison if all three next gen are on the same node.
Certainly makes any PPA and costs calculations a lot easier

 

[Kepler_L2]

https://twitter.com/x/status/1810553552311611569

N4P is the rumor

So they still have terrible PPA. Sad stuff.

 

[gaav87]

https://twitter.com/x/status/1810553552311611569

N4P is the rumor

How can it be 400mm2 for 32xe2 cores. What about the shiping manifests 56/64xe2 versions that would put them above reticle limit.
8 cores of Xe2_lpg die area is ~40.7mm2 Media is ~13.5mm2 display ~7mm2 Assuming 1 display engine and 1 media engine 20Xe2 core should be: 122,25mm2 on N3E ? and around 153,625mm2 on N4 [N5->N3E is 1.3x chip density N5-N4 is 1.06x]
Im missing L2$ controls, 2 gddr phys or 1 gddr phys (192/256bit) and l2$
Still 270mm2 for 20xe2 cores and 400mm2 for 32xe2 seems ridiculous too me.

After some calculations i would put this at
32Xe2: 205-275mm2
64Xe2: 410-550mm2

If 32xe2 is 400mm2 and 4070non-super competitor bmg is DOA... 4070ti is 294.5mm2

 

[Ghostsonplanets]

So they still have terrible PPA. Sad stuff.

For comparison AD107 is 24SMs/3072 Shaders on a ~160mm² die on 4N

20 Xe² is 160 EUs/2560 Shaders at 270mm² on N4P

Terrible PPA if so, yes.

 

[poke01]

The 4070S is 294mm2. So yeah bad.