Intel Confirms AIB Partners Are Building Custom Arc Alchemist Graphics Cards
In a wide ranging interview with a Japanese news outlet, Intel graphics boss Raja Koduri shared some insight into the development of Alchemist, the company's upcoming discrete GPU, and what it has planned. One of the more notable tidbits from the interview was in relation to Intel's add-in board (AIB) partners, and whether or not they will be releasing custom models.
Unless plans change between now and then, the answer is yes, Intel's hardware partners will offer up custom models rather than just sticking with Intel's reference design. This is the same approach AMD and NVIDIA take with their graphics cards—they each launch their own models (NVIDIA calls them Founders Edition), while AIB partners can decide whether to stick with reference designs, or offer up custom-cooled and overclocked variants. Or both.
"Partners and I think there will be a differentiation of ODM [original design manufacturer], and that will lead to the ultimate customer interest," Koduri replied when asked if Intel plans to release its own product model similar to NVIDIA's Founder Edition strategy.
Koduri also noted that Intel Arc as a whole is able to support professional graphics cards, similar to AMD's Radeon Pro and NVIDIA's Quadro product lines. However, he did not say whether Alchemist specifically will find its way to any professional SKUs.
Intel Turns To TSMC For Its Alchemist GPUs
Another interesting comment from the interview was in regards to Intel tapping TSMC to manufacture its Alchemist GPUs on its 6-nanometer node, rather than making the silicon in-house. What it boils down to are performance, cost, and capacity considerations, three factors that Koduri says are considered regardless of whether it decides to build chips in-house or externally.
"It is necessary to first determine the entire process that can be assumed at the start of design," Koduri said, adding that "other features such as how much operating frequency can be used are also important factors. Cost is also an issue. These three, that is, the cost, performance, and capacity are taken into consideration when deciding which process to use."
Intel could have decided to use its own '7' node (previously known as 10nm Enhanced SuperFin), as it is doing with Alder Lake, but ultimately decided that TSMC's 6nm node offered the best balance of the three factors Intel uses to determine the best manufacturing path.
Looking ahead, it sounds like Intel has not yet decided whether to task TSMC with making its future GPUs, such as Battlemage, or to produce them in-house. Both are options right now, to be decided at a later date, and could depend on how the Alchemist launch goes.
Some other notable mentions during the interview...
- XeSS is backwards compatible with current Xe GPUs
- Intel is not planning to support multi-GPU configurations on the consumer side
- New drivers will release regularly, including when new games come out
Check out the multi-page interview with Koduri at ASCII.jp for more of what he had to say.
[Update: A previous version of this article stated that Intel did not have the capacity to make Alchemist chips internally. An internal spokesperson reached out to say this was a misquote on ASCII.jp's part, and we have updated our story accordingly.]