In fact, Intel hasn’t caught up with AMD when it comes to graphics card market share – an error has crept into the calculations

It was recently announced that Intel has allegedly caught up with AMD in terms of market share for discrete graphics cards. This data was provided by authoritative analysis agency Jon Peddie Research (JPR), so there was no doubt about its reliability. However, the data turned out to be still wrong: analysts included data center accelerators in the statistics, which skewed the data.

Recall that earlier this week, JPR released data on discrete graphics card sales for the fourth quarter of 2022, which indicated that both Intel and AMD had the same market share — 9% each. However, these conclusions are based on flawed data that included 60,000 Intel Ponte Vecchio compute accelerators in addition to consumer desktop and mobile Intel Arc graphics cards. This significantly skewed the statistics.

That’s because Intel included the revenue from the sale of the Ponte Vecchio accelerators in the earnings of the division that sells consumer graphics cards, which significantly boosted the earnings of this division in the quarterly report. Meanwhile, JPR speculated that the division only sold consumer graphics cards. As a result, analysts got the wrong number of GPUs Intel sold last quarter when dividing revenue by the average selling price of consumer Intel Arc cards.

After fixing the bug, it turned out that of the 13 million discrete desktop and laptop GPUs sold last quarter, 85% were NVIDIA, 9% AMD, and 6% Intel. In fact, controlling such a share of the discrete GPU market is a major achievement for a company that’s only been in the market for two years. Given that the bulk of Intel’s graphics shipments are likely to be low-end laptop GPUs, the mid-range price lags behind AMD and NVIDIA, but over time Intel is likely to catch up on that front as well.

Although Intel is a relative newcomer to the discrete GPU market, its product portfolio is already quite diverse. The company sells Arc Alchemist graphics cards for desktops and laptops, it has Flex graphics cards on the same GPUs but designed for data centers for remote rendering and media streaming, and Intel has Data Center GPU Max compute accelerators, formerly known as Ponte Vecchio. Note that all GPUs mentioned, both consumer and server, are based on the Intel X-Architecture.^e.

Intel used to include sales of its discrete GPUs in Accelerated Computing Systems and Graphics Group (AGX) sales, but in Q4 2022 this division was split into two branches: consumer GPUs were moved to Client Compute Group (CCG) and Server GPUs became part of the data center and AI (DCAI). Analysts initially thought that fourth-quarter 2022 Ponte Vecchio accelerator sales were already included in DCAI’s numbers, while consumer GPU sales were included in AXG data.

AXG’s revenue rose sharply in the fourth quarter, which JPR attributed to rising sales of consumer GPUs. But also in the last quarter of last year, Intel delivered the first large batch of Ponte Vecchio computer accelerators, presumably for the Aurora supercomputer at Argonne National Laboratory. This machine will require more than 60,000 Intel Data Center GPU Max accelerators, and JPR estimates a significant portion of these shipped in Q4 2022. JPR now believes that Ponte Vecchio sales were reflected in AXG results, causing its original estimates of Intel graphics card shipments to be wrong.

“We used average retail price for AXG discrete graphics cards to calculate unit shipments, assuming Ponte Vecchio is included in DCAI, JPR explained. — The tricky part is splitting X^e about the DCAI and AGX groups, and which group received revenue from supplies for Aurora“.

“Since the GPU group split occurred in the fourth quarter, we believe AXG is showing revenue from all shipments GPU based on Xe architecture, including 60,000 Ponte Vecchio accelerators, — explained in JPR. — This profit increase disrupted our average selling price modeling and gave the impression that Intel had a large increase in shipments and that dGPU shipments were increasing more than they actually were.“.

“We never considered AMD or Intel accelerators in our reports and were surprised by Intel, wrote JPR. – We don’t think Intel intentionally misled the industry and we’re just not used to distinguishing consumer and data center shipments of discrete GPUs – a discrete GPU is a discrete GPU (except when it’s not).“.