Intel Arc "Battlemage" GPUs Confirmed for 2024 Release

[btarunr]

Intel, in a company presentation made to its channel partners, confirmed that it is looking to release its next generation Arc Xe² discrete GPU lineup, codenamed "Battlemage." This would be Intel's second rodeo with high performance gaming graphics since its 2022 return to the segment with the Arc "Alchemist" series. The One Intel presentation slide talks about what to look forward to from the company in the client segment, in the coming year. The slide states that PC processor, workstation processor, and discrete GPU segments will each see upcoming products, which can be seen as a confirmation for a 2024 launch of "Battlemage." Older company slides had illustrated that the launch of "Battlemage" would be timed around that of the company's "Meteor Lake" and "Arrow Lake" client processors. The company is expected to launch "Arrow Lake" sometime in 2024. With "Battlemage," Intel is looking to offer a linear increase in performance, along with new hardware capabilities. The discrete GPUs from this family are expected to be built on a 4 nm-class foundry node by TSMC.



View at TechPowerUp Main Site | Source

 

[KrazyT]

Wait & See, but very curious about this one !

 

[bug]

If this is just a tiled Alchemist, it may help reduce the MSRP, but it probably won't help with TDP, which is on the high side.
Wait and see ✓
Fingers crossed ✓
Holding breath ✗

 

[phanbuey]

Well at least they still make top-notch marketing slides.

 

[bug]

Well at least they still make top-notch marketing slides.

They do, but these aren't those. These are extremely generic.

 

[80-watt Hamster]

They do, but these aren't those. These are extremely generic.

They're also not marketing slides:

... in a company presentation made to its channel partners ...

 

[R0H1T]

Considering the last set of Arc GPU's had so little availability & the rumors of Intel abandoning the (consumer?)space post battlemage would "channel partners" really want to gamble billions on this roadmap?

 

[theouto]

Well at least they still make top-notch marketing slides.

After core truths, we were all worrying for intel, fearing that perhaps they'd lost their mojo

 

[bug]

Considering the last set of Arc GPU's had so little availability & the rumors of Intel abandoning the (consumer?)space post battlemage would "channel partners" really want to gamble billions on this roadmap?

Gamble? Do you honestly think partner have nothing more to base their plans on, other than internet rumors?
And even if Intel was exiting the business, what would those billions be poured into? It's not like partners design fully working cards, sign supply deals and update their retail channels, all before seeing a working GPU.

 

[R0H1T]

With the MTL performance numbers being bad, at least initially, this would pile on top of that. And no you definitely do need some amount of assurance that they're not pouring money in a blackhole, we've seen what Intel's done with mobile/tablet space a decade back & they poured probably 10x the amount(subsidies) through contra revenue & even then it just went poof almost overnight!

 

[bug]

With the MTL performance numbers being bad, at least initially, this would pile on top of that. And no you definitely do need some amount of assurance that they're not pouring money in a blackhole, we've seen what Intel's done with mobile/tablet space a decade back & they poured probably 10x the amount(subsidies) through contra revenue & even then it just went poof almost overnight!

What is this black hole you talk about? Partners are currently planning for Battlemage which apparently will happen. They're not planning anything else, unless Intel gave them some solid reason to do so.

And about MTL number being bad, have you seen this? https://www.phoronix.com/review/meteor-lake-arc-graphics

 

[Vayra86]

I guess Intel needs to gather its party before venturing forth.

Q4 2024? We've already known about something something Battlemage in H2 2024 for nearly a year...

If this is just a tiled Alchemist, it may help reduce the MSRP, but it probably won't help with TDP, which is on the high side.
Wait and see ✓
Fingers crossed ✓
Holding breath ✗

Alchemist isn't built on 4nm though, so they'll get that sweet shrink juice

Well at least they still make top-notch marketing slides.

Here's a wild thought. The chips on this slide align with the input quality of the render from gen to gen, the rest is all gonna be XeSS

 

[pressing on]

With the MTL performance numbers being bad, at least initially, this would pile on top of that.

MTL performance numbers so far show that MTL single-thread performance is only the same or slightly less than the previous gen. Multi-thread performance is a different issue. For example this graph shows the relative performance of the MTL 185H versus the 13800H in CPU-Z bench. Both are 6P + 8E parts, top speed of the 185H is 5.1 GHz against 5.2 GHz for the 13800H.

 

[SSGBryan]

Considering the last set of Arc GPU's had so little availability & the rumors of Intel abandoning the (consumer?)space post battlemage would "channel partners" really want to gamble billions on this roadmap?

The rumors all came from 1 place, Moore's Law is Dead.

In fairness, what he reported was that Intel was sitting on the fence about whether to go forward or not. The internet, being the internet, extrapolated that to "It's dead Jim."

I have been quite pleased with my a770; It is a solid 1440p gaming card, and is better for productivity than the RTX 3060 that was replaced.


The performance goals of the top battlemage gpu is 4070ti in gaming and 4080 in compute. Looking forward to getting one this summer.

 

[Vayra86]

The performance goals of the top battlemage gpu is 4070ti in gaming and 4080 in compute. Looking forward to getting one this summer.

If they get there in Q2 2024 they're really playing along with the big boys, and that's not bad considering this is their second gen. Nvidia might have to start actually doing something, I don't think their RT (or DLSS) button is going to keep working for them when consoles and two of three gpu players are not all in on RT, but do have a strong upscale solution.

Against just AMD they can still say they're right and the other camp is wrong. But when Intel also just sticks to common sense and doesn't push RT heavily (which I think they won't, because they simply can't)... things change. Now suddenly Nvidia is the odd one out, between everything else moving in gaming they have a handful of 'awesome RT' pushes.

Also take note of the fact the next gen console is going to be AMD again, so this situation won't get better for Nv either.

 

[Squared]

I think Intel has been much smarter about marketing Battlemage than they were Alchemist and Meteor Lake and have been with Arrow Lake and Lunar Lake. All of these Intel keeps talking up and leading us to believe they're coming soon. For Battlemage I think Intel would do well to just shut up until its release is actually close. Tell the people who actually need to know (like the channel partners in this case) and don't tell the consumers until you have a working product to show off and a real launch date.

 

[80-watt Hamster]

The performance goals of the top battlemage gpu is 4070ti in gaming and 4080 in compute. Looking forward to getting one this summer.

Does that mean double precision in hardware? That was my biggest disappointment in Alchemist: no F@H.

 

[AnotherReader]

The rumors all came from 1 place, Moore's Law is Dead.

In fairness, what he reported was that Intel was sitting on the fence about whether to go forward or not. The internet, being the internet, extrapolated that to "It's dead Jim."

I have been quite pleased with my a770; It is a solid 1440p gaming card, and is better for productivity than the RTX 3060 that was replaced.


The performance goals of the top battlemage gpu is 4070ti in gaming and 4080 in compute. Looking forward to getting one this summer.

I would temper my expectations if I were you. The 4070 Ti is much faster than the A770. A node shrink is unlikely to make up for that gap; the 4070 is a more realistic target and there would be no shame in matching it. At 1440p, the 4070 is nearly 60% faster than the A770.

 

[ZoneDymo]

I would temper my expectations if I were you. The 4070 Ti is much faster than the A770. A node shrink is unlikely to make up for that gap; the 4070 is a more realistic target and there would be no shame in matching it. At 1440p, the 4070 is nearly 60% faster than the A770.

View attachment 326226

hell I would be happy if it would finally match the RTX3070 as the specs of the A770 should allow.....

but far more important is that damn price, Intel needs to be far lower to be able to catch a lot of customers and build a reputation, I would not mind an A770 16gb but that price is just way too high for waht it is.

 

[jallenlabs]

Love my ARC card. After a bit of tinkering, its great for video editing. I dont game on it, but its been stable and fast for me.

 

[SSGBryan]

I would temper my expectations if I were you. The 4070 Ti is much faster than the A770. A node shrink is unlikely to make up for that gap; the 4070 is a more realistic target and there would be no shame in matching it. At 1440p, the 4070 is nearly 60% faster than the A770.

View attachment 326226

Have you looked at the specs for the top battlemage die?

A770 - 32 Xe cores
B770 - 64 Xe cores

A770 - 4096 FP32 cores
B770 - 8192 FP32 cores

A770 - 2.1Ghz clock
B770 - 3.0Ghz clock

A770 - 225 watt power draw
B770 - <225 watt power draw.

 

[AnotherReader]

Have you looked at the specs for the top battlemage die?

A770 - 32 Xe cores
B770 - 64 Xe cores

A770 - 4096 FP32 cores
B770 - 8192 FP32 cores

A770 - 2.1Ghz clock
B770 - 3.0Ghz clock

A770 - 225 watt power draw
B770 - <225 watt power draw.

Intel is moving from TSMC's N6 to N5 so all of those specs seem rather unlikely. Even Nvidia didn't see a jump like that despite moving from a process that was inferior to TSMC's N7.

 

[Squared]

Xe 1 in Meteor Lake is built with TSMC N5 (and the memory controller is on N6) and according to Phoronix it's very competitive in performance and power efficiency with RDNA3 in Phoenix, which was built with TSMC N4. So Battlemage has a fair shot at being competitive if it arrives inside 2024 and is built with at least TSMC N4.

 

[Minus Infinity]

If we are to believe all the leaks coming out of Intel, top tier Battlemage which was once targetting 4080 in raster is apparently going to be closer to 4070 in raster. It may be faster than 4080 in compute. Battlemage doesn't appear to be able to hit its original goals and the fact the top tier is only 225W gives a clue. I doubt it is super efficient. Still even 4070 level raster with good RTung at say $449 or less would be welcome addition. But then again RDNA 4 N48 is said to be 7900 levels of raster and Nvidia will have Blackwell for Battelmage to contend with. Also IIRC Arrow Lake iGPU is not Battlemage but Alchemist+

 

[MrMilli]

Intel is moving from TSMC's N6 to N5 so all of those specs seem rather unlikely. Even Nvidia didn't see a jump like that despite moving from a process that was inferior to TSMC's N7.

I think Intel will easily double its current performance. N5 alone will be enough to double the Xe cores to 64.
Hopefully Battlemage won't be just a simple die shrink.
If they fix one of these issues then they will be competitive:
- horrible cache/memory latency/bandwidth
- core dependency on high parallelization