We know Intel has assembled a team to build its first modern discrete GPU, which if all goes to plan, will arrive next year. What we do not know, however, is what market segment it will target out of the gate. Raja Koduri, chief architect and senior vice president of Architecture, Software, and Graphics at Intel, may have shed some light on that in a recent interview.
The interview spans just over 32 minutes, and part of the discussion apparently revolves around GPUs and Intel's desire to come out swinging at around the $200 price point. I say "apparently" because unfortunately, the entire video is dubbed over in Russian. [EDIT: And apparently that was not actually the case--see our update at the end of this article]. Here's the video...
While I do not speak a lick of Russian, a user on Reddit (taryakun) posted a supposed translation of the pertinent part of the interview.
"Our strategy revolves around price, not performance. First are GPUs for everyone at $200 price, then the same architecture but with the higher amount of HBM memory for data centers... Our strategy in 2-3 years is to release whole family of GPUs, from integrated graphics and popular discrete graphics, to data centers GPUs," Koduri said, according to the translation.
There are multiple reports on the interview, and as far as I can tell, they are all rooted in taryakun's translation from Russian to English.
Let's assume it is accurate, for the sake of discussion. If so, the desired price point is not surprising. It was always a long shot that Intel could shake things up in the high end of the market right out of the gate. Designing and building a GPU architecture takes time, after all, and aiming for 2020 is an aggressive release schedule.
That's not to say that Intel will not have a high end competitor sometime down the line. It very well could, but initially, a $200 GPU for the mainstream market makes sense. AMD has sort of taken the same path, even if not entirely by design. For the past several years, Polaris has mostly competed in the mainstream market, leaving NVIDIA all by itself at the top end. Then Vega came along, which upped the ante to an extent, and now Navi has pushed AMD into a higher performance tier.
Interestingly, according to a translated ComputerBase article, Koduri also mentioned using a "larger amount of HBM memory" on an enterprise graphics card, sometime after Intel hits the mainstream consumer market. Margins are typically higher in the enterprise, which might be why Intel figures using HBM memory makes sense.
On the consumer side, Intel has been campaigning its efforts, and its "Odyssey" into graphics. The takeaway is that Intel is serious about delivering discrete GPUs to gamers. I'm not holding my breath for ray tracing support at the outset, but given Intel's hype on the subject (of graphics), such a GPU will probably manifest at some point.
One other thing worth noting—Intel this week finally announced its first volume roll out of 10nm processors, and they bring with them impressive integrated graphics for laptops and other mobile platforms. We posted a handful of Ice Lake benchmarks that is worth checking out when you have a free moment.
A spokesperson for Intel reached out to us and let us know that some things had been misconstrued and/or lost in translation. Rather than targeting the $200 price point initially, Koduri was making the point that not all users will buy a $500-$600 graphics card, and that Intel's strategy revolves around going for the full stack that ranges from client to the data center. The reference to $200 was an example of entry level pricing for client GPUs, and not confirmation of Intel's discrete GPU strategy.
Intel also provided us with a correct translation what exactly Koduri said...
"Not everybody will buy a $500-$600 card, but there are enough people buying those too—so that’s a great market. So the strategy we’re taking is we’re not really worried about the performance range, the cost range and all because eventually our architecture as I’ve publically said, has to hit from mainstream [which starts around $200] all the way to Data Center class graphics with HBM memories which will be expensive. We have to hit everything, it’s just a matter of where do you start? The first one? The second one? The Third one? And the strategy that we have within a period of roughly—let’s call it 2-3 years—to have the full stack," Koduri said.
Image Source: PRO Hi-Tech (YouTube)