TSMC has reportedly won some very large production orders from Intel. The orders are not only for the next-generation Battlemage graphics processing units, but also for the follow-up Celestial architecture GPUs. Taiwan’s Commercial Times cites industry sources for its insider info, which also includes some tantalizing news nuggets regarding processes, timings and volume.
Whether Intel’s first-generation consumer GPUs have been a success is open to some interpretation. The Intel Arc Alchemist graphics cards (using TSMC 6nm silicon) were terribly late, but are now widely available and appear to be heavily discounted to shift inventory. We have reviews of Intel's Arc A770 16GB, Arc A750, and Arc A380, and thanks to the competitive pricing, some of these even rank among the best graphics cards. Intel offers a well-known brand and a third GPU architecture choice for enthusiasts who don’t blindly follow brands. Most positively, the driver team has been beating expectations with the rate and quality of updates over recent months.
Taiwanese insiders say that Intel is continuing its next-generation GPU R&D programs as previously planned. You may be eager to hear when these generations of GPUs might become available. According to the source, “Intel will launch the second-generation Battlemage graphics chip with the Xe2 architecture in the second half of 2024, and the third-generation Celestial graphics chip with the Xe3 architecture in the second half of 2026.” It is also claimed that Battlemage GPUs will be fabricated on TSMC 4nm, and Celestial on 3nm processes at TSMC.
There have been some Intel naysayers and Intel doomsters — and the recent departure of Raja Koduri doesn’t look very positive. Nevertheless, the Commercial Times insiders claim Intel has clear sight of strong demand for graphics. From the consumer side, gaming / eSports continues to be strong, and in business / enterprise there is still a growing need for accelerators for AI and other intensive computing tasks.
The last official information we got regarding Battlemage was earlier this year, via a tech press interview with Intel Fellow Tom Peterson. He made it clear that Intel had learned from several mistakes with Alchemist, and where that generation had four targeted microarchitectures, the second-gen will focus on two: Xe2-LPG and Xe2-HPG. It is hoped this focus will prevent delays and streamline driver development, as well as being more economical.
Intel has quite a lot of time to fill between now and the rumored H2 2024 launch of Battlemage. With new Intel hardware seemingly a long way off, new AMD and Nvidia products are sure to emerge in the meantime. So Intel’s driver team will have to keep working hard pushing updates and tuning performance to get the most performance possible from existing silicon. Hopefully that means when Battlemage does arrive, drivers will be the last thing potential buyers have to worry about.
So we do now give kudos to a multi billion dollar company.. that released a highly dysfunctional and broken product and is now fixing their mutiple own fk ups? Something that should have beeen actually a given ???
Come on! I know what you mean, but the way you say it sounds just deliberately wrong.
I highly disagree with your statement on this one, but I can at least agree to the rest of what you said.
We do it all the time but your not wrong...
The bar wasn't just lower, it was buried under six feet of fud. Its amazing Intel managed to dig the danged thing up and get their card functional at all. I honestly expected Intel to already have killed the project at this point. Much to my delight they haven't as we are in desperate need of more competition in the GPU space. The question is will Intel get competitive or fold first?
The crypto-craze days are gone. AMD and Nvidia have working GPUs gathering dust on store shelves and warehouses right now with sales at 20+ years lows. Having products on the market isn't enough anymore, they also need to be priced low enough for normal gamers to actually want to buy them. Intel also bears the additional burden of new-kid-on-the-block status and that will take 2-3 technically and marketably successful generations to achieve. If it tries to sell almost exactly the same performance at almost exactly the same price as AMD or Nvidia, it will fail.
I believed adding Intel's manufacturing weight to the GPU market supply would be very beneficial. It also made more sense for continuity with it's iGPUs.
Furthermore if their previous leaked "target of <225 watts" holds true, and assuming an enormous jump in efficiency I think they're going to land somewhere around 3080/4070Ti levels of performance, and competing with a later generation at the same time.
Right now, the A750 barely competes with AMD and Nvidia GPUs that have half as much raw compute performance with a few games faring drastically worse. Regardless of how much Intel's drivers may have improved since launch, they still have a very long way to go.