When earlier this week it turned out that Intel sold as many discrete graphics processors as AMD in the fourth quarter of 2022, based on data from reputable Jon Peddie Research, we felt something was off. It turns out that JPR had inadvertently included 60,000 Ponte Vecchio data center GPUs in his calculations. Naturally, that threw the numbers off significantly.
Jon Peddie himself was kind enough to warn us that JPR's unit shipment estimates were based on Intel's recent financial report and ASPs and could be inaccurate. Indeed, they were, and Intel did not sell as many standalone GPUs as AMD last quarter.
6%, Not 9%
Intel included revenue it got from selling Ponte Vecchio compute GPUs into earnings of the division that sells consumer GPUs, which significantly increased its revenue.
Meanwhile, JPR assumed that the business unit only sold graphics processors from client PCs, so when it divided revenue by average selling price (ASP) it considered accurate for Intel Arc products, it got an incorrect number of client GPUs sold by Intel last quarter.
After correction, it turned out that out of 13 million standalone GPUs for desktops and notebooks sold last quarter, Nvidia shipped 85%, AMD supplied 9%, and Intel sold 6%.
In fact, controlling of 6% of the discrete GPU market is a big deal for a company that has been on this market for only two years. Keeping in mind that the bulk of Intel's GPU shipments are probably inexpensive notebook GPUs, its ASPs are behind those of AMD and Nvidia, but eventually the company will likely catch up on this front as well.
Here Is What Happened
While Intel is a relative newbie on the discrete GPU market, its product portfolio is now pretty broad. The company sells standalone Arc Alchemist GPUs for desktops and notebooks, it has Flex Series graphics cards tailored for remote rendering and media streaming (yet based on the same silicon as Arc products for client PCs), and it has Data Center GPU Max compute GPUs formerly codenamed Ponte Vecchio.
Intel typically included sales of its discrete graphics processors into revenue of its Accelerated Computing Systems and Graphics Group (AGX), but in Q4 2022 that division was split into two different branches with consumer GPUs moving to Client Compute Group (CCG) and server GPUs became part of the Data Center and AI (DCAI) division. Analysts from Jon Peddie Research asserted that sales of compute GPUs in Q4 were already folded into results on the DCAI, whereas sales of consumer GPUs in Q4 were included into results of AXG.
Meanwhile, AXG substantially increased its revenue in Q4, which JPR attributed to increased sales of its consumer GPUs. But in Q4 2022, Intel shipped its first large batch of Ponte Vecchio compute GPUs presumably for Argonne National Laboratory Aurora supercomputer. The machine requires 60,000 of Intel's Data Center GPU Max and a substantial part of them were shipped in Q4 2022, according to JPR's estimates. JPR now believes that sales of Ponte Vecchio were reported in AXG's results, so its unit shipments estimates were incorrect.
"We had been using an ASP for the dGPUs from AXG to arrive at unit shipments, assuming Ponte Vecchio was rolled up in the data center (DCG) group, renamed DCAI," the explanation by JPR reads. "The tricky part is the split of Xe to computer (DCAI) and AGX, and which bucket got credit for the Aurora shipments.
Since the split of the GPU group took place in Q4, we think the AXG got credit for all Xe shipments, which would include the 60,000-plus Ponte Vecchio dGPUs," JPR explained. "That spike in revenue threw off our ASP modeling and made it look like Intel had a big jump in shipments, and that dGPU shipments had gone up more than they had."
Jon Peddie Research has remodeled its Intel’s Q4 discrete GPU shipments by subtracting the 60,000+ expensive Ponte Vecchio compute GPUs and naturally got different results.
"We have never counted AMD or Intel GPU-compute GPUs in our quarterly reports and got caught by surprise by Intel," the company wrote. "We don’t think Intel intended to deliberately mislead the industry and simply isn’t used to dGPU consumer vs. data center GPU shipment differentiation — a dGPU is a dGPU (except they aren’t)."