Owing to BOM cost limitations, DDR5 integration for both Intel and AMD PC platforms will be delayed to 2022

The majority of PCs are based on the Intel platform, which accounts for more than 70% of total PC shipment. As PC consumers are extremely sensitive to device prices, and DDR5 memory will command a premium over DDR4 during the initial stages of release, Intel will not rush to offer DDR5 support on its platforms. Although the company’s roadmap had previously suggested that it will offer DDR5 support first with Tiger Lake-H (highest-end configuration only), planned for release in 2021, Intel ultimately decided to push back official DDR5 support potentially to the start of 2022 at the earliest, with its Alder Lake platform instead.

On the other hand, the second-largest PC platform supplier, AMD, whose CPUs are found in more than 20% of total PC shipment, will similarly offer DDR5 memory support with its 5nm CPUs in 2022 at the earliest. Both Intel and AMD are unable to offer DDR5 support until 2022 due to the BOM costs involved, meaning DDR5 memory will remain in the product development and verification phases in 2020 and 2021.

Intel and AMD may likely mass produce server platforms supporting DDR5 server DRAM in 2022

As with the PC market, Intel is the dominant platform supplier in the server market, accounting for more than 90% of total server shipment. Since this market is less sensitive to manufacturing costs compared to the PC market, Intel will be faster to introduce DDR5-compatible platforms in the server market than in the PC market. The company is expected to first bring DDR5-compatible products to the server market with its Eagle Stream platform, expected to enter small-batch production in 2H21. TrendForce expects mass production of Intel platforms supporting DDR5 server DRAM to gradually ramp up in 2022 and replace existing platforms supporting DDR4 DRAM.

Trailing behind Intel is AMD, with an approximately 10% market share in terms of total server shipment. AMD’s upcoming Milan platform will continue to support DDR4 only. AMD will unlikely support DDR5 until it releases the Genoa server platform, although this is yet to be confirmed. The Genoa platform will enter its testing phase in 2021, with mass production expected to kick off in 2022. Should AMD officially offer DDR5 support with Genoa, that means AMD’s server platforms will not transition into DDR5 support on a large scale until 2023 at the earliest.

Owing to high-end smartphone chips from Qualcomm and MediaTek, LPDDR5 is expected to overtake DDR5 in penetration rate

As the dominant mainstream suppliers of smartphone APs, Qualcomm and MediaTek collectively occupy close to 70% of all smartphone shipments this year. Of the two companies, Qualcomm is relatively more committed to LPDDR5 adoption, since it has already incorporated the next-generation memory into its solutions this year. Qualcomm released its flagship, LPDDR5-compatible chip Snapdragon 865 in early 2020, while the upcoming 870 series of flagship chips will also support LPDDR5 memory.

On the other hand, MediaTek’s roadmap of LPDDR5 adoption lags behind Qualcomm, since its high-end 5G chips, including reference Dimensity SoCs, support only LPDDR4 memory at the moment. However, MediaTek is projected to release at least two (as yet unnamed) chips that will be the first in its product portfolio to support LPDDR5; these chips are likely to be available in the market in 1H21.

The price gap between LPDDR5 and LPDDR4X has been narrowed to under 10%, owing to the strategic marketing efforts of DRAM suppliers. This narrowing will contribute to the popularity of the next-generation memory. Looking ahead to 2021, as Qualcomm’s 870 series of flagship chips and MediaTek’s next-generation chips will support LPDDR5, along with the fact that LPDDR5 is both faster and more power-efficient than its predecessor, LPDDR5 is expected to reach a penetration rate of 18.5%. It should be pointed out that the future development of LPDDR5 will depend on its price difference from LPDDR4X. Once this difference becomes negligible, the penetration rate of LPDDR5, compared to that of DDR5, will be more rapid and noticeable.

GDDR6 has become the mainstream graphics DRAM solution and standard spec on new graphics cards from Nvidia and AMD

Nvidia, the dominant supplier in the GPU industry, accounts for about 75% of all graphics card shipment, and the company’s Turing-based GTX and RTX graphics cards, released in 2018, are equipped with GDDR6 memory in the mid-range and high-end product lines, while all cards in the upgraded GTX and RTX series, dubbed the “Super” series, were equipped with GDDR6.

Likewise, GDDR6 is standard spec across all graphics cards with Ampere GPUs, released in 3Q20. The second-largest GPU supplier, AMD, occupies about 25% of graphics card shipment. AMD’s Navi GPUs, released in 2019, are all equipped with GDDR6 memory. GDDR6 will continue to remain standard spec for AMD’s next-generation Big Navi GPU, announced in 3Q20. Big Navi will enjoy increased GDDR6 content per box (that is, DRAM capacity), thereby quickly raising the penetration rate of GDDR6 memory.

Graphics cards from Nvidia and AMD account for nearly 70% of all graphics DRAM bit consumption. Therefore, given the two companies’ active support for GDDR6, the price differences between GDDR6 and GDDR5 are continuously decreasing (at the moment no more than 10%, for the most part; certain suppliers have even quoted GDDR6 at the same price as GDDR5).

Aside from graphics cards, game consoles account for about 30% of graphics DRAM bit consumption. Microsoft Xbox Series X and Sony PS5, both set to be released in 4Q20, will be equipped with 16GB of GDDR6. Not only is this double the amount of GDDR5 memory equipped on current-gen consoles, but it also far exceeds the 6 to 8GB equipped on current mainstream graphics cards.

TrendForce indicates that the dominant platform suppliers of the above product categories have a considerable influence over the penetration and integration of next-generation DRAMs. DRAM manufacturers are generally capable of producing DRAM chips or modules within one year from JEDEC’s release of exact specifications for the next-generation DRAM products. However, without support from the platform end, the penetration rate growth of next-generation DRAMs will be constrained even if DRAM manufacturers are ready to release their own next-generation DRAM product mixes.