Video Memory Discussion

[safar1980]

HBM Prices to Increase by 5-10% in 2025, Accounting for Over 30% of Total DRAM Value, Says TrendForce
Avril Wu, TrendForce Senior Research Vice President, reports that the HBM market is poised for robust growth, driven by significant pricing premiums and increased capacity needs for AI chips. HBM's unit sales price is several times higher than that of conventional DRAM and about five times that of DDR5. This pricing, combined with product iterations in AI chip technology that increase single-device HBM capacity, is expected to dramatically raise HBM's share in both the capacity and market value of the DRAM market from 2023 to 2025.

Specifically, HBM's share of total DRAM bit capacity is estimated to rise from 2% in 2023 to 5% in 2024 and surpass 10% by 2025. In terms of market value, HBM is projected to account for more than 20% of the total DRAM market value starting in 2024, potentially exceeding 30% by 2025.

[FP91G]

SK Hynix Reveals Plans For Cutting-Edge HBM4E Memory, Development Expected By 2026
The massive HBM industry growth has initiated the "fire" of innovations, with the Korean giant SK hynix revealing plans for HBM4E memory.

SK Hynix Looks To Establish New Benchmarks With HBM4E Through Integration of Semiconductors & Memory Into One Package
Right now, the AI industry sees HBM as a crucial component for progress in performance capabilities due to its immense importance in modern-day AI accelerators. Recently, we saw the widespread adoption of the HBM3E standard in newer AI GPUs such as the Blackwell B100 and the Instinct MI300X, which brought in a huge boost in performances; however, SK hynix has disclosed that this is just the start, as the Korean memory manufacturer plans on mass-adoption of the HBM4E somewhere within the next two years.
https://wccftech.com/sk-hynix-reveals-plans-for-cutting-edge-hbm4e-memory-development-expected-by-2026/

[safar1980]

SK Hynix Revealed Progress for HBM3e, Achieving Nearly 80% Yield
SK hynix has disclosed yield details regarding the company’s 5th generation High Bandwidth Memory (HBM), HBM3e, for the first time. According to a report from the Financial Times, citing Kwon Jae-soon, the head of yield at SK hynix, the memory giant has successfully reduced the time needed for mass production of HBM3e chips by 50%, while close to achieving the target yield of 80%.

This is better than the industry’s previous speculation, which estimated the yield of SK Hynix’s HBM3e to be between 60% and 70%, according to a report by Business Korea.

[FP91G]

SK Hynix Revealed Progress for HBM3e, Achieving Nearly 80% Yield
SK hynix has disclosed yield details regarding the company’s 5th generation High Bandwidth Memory (HBM), HBM3e, for the first time. According to a report from the Financial Times, citing Kwon Jae-soon, the head of yield at SK hynix, the memory giant has successfully reduced the time needed for mass production of HBM3e chips by 50%, while close to achieving the target yield of 80%.

This is better than the industry’s previous speculation, which estimated the yield of SK Hynix’s HBM3e to be between 60% and 70%, according to a report by Business Korea.

Good news for buyers of NVIDIA B100 or H200 in the future, but miners do not buy such video cards. The price is very high.
If AMD releases some modern analogue of VEGA on HBM3e, then it will be interesting to test.

[safar1980]

DDR6 And LPDDR6 Memory
Thanks to the increased bandwidth, the 10.667 Gbps per-pin data rate translates to about 28.5 GB/sec of memory bandwidth from a single LPDDR6 IC. If we talk about a typical desktop PC's "dual-channel" memory interface, we're looking at 228 GB/second of memory bandwidth. For comparison, a PC using DDR5-6400 memory only has 102.4 GB/second of memory bandwidth. In short, LPDDR6 looks like it will roughly double available system memory bandwidth once it hits.

But what about regular old desktop DDR6? The spec for it isn't final yet, but JEDEC's documents suggest that data rates for the socketable RAM could start as high as 8.8 Gbps and then scale all the way to 17.6 Gbps, or perhaps even as high as 21 Gbps. JEDEC's notes suggest that the draft spec for DDR6 should be released this year, and then the spec should be finalized in Q2 of 2025.

[Viktor_1509]

Thanks for sharing this insightful information about DDR6 and LPDDR6 memory advancements!

The leap in bandwidth for LPDDR6 is indeed impressive, with 10.667 Gbps per-pin translating to about 28.5 GB/sec from a single LPDDR6 IC. When you consider a typical desktop PC's dual-channel memory interface, the potential 228 GB/second of memory bandwidth is a significant boost compared to the 102.4 GB/second with DDR5-6400 memory. This essentially doubles the available system memory bandwidth, which is quite remarkable.

As for regular desktop DDR6, it's exciting to see the projected data rates ranging from 8.8 Gbps up to potentially 21 Gbps. With JEDEC's draft spec expected this year and finalization anticipated by Q2 of 2025, we're on the cusp of a new era in memory technology that could substantially enhance overall system performance.

Looking forward to seeing how these advancements will impact both consumer and enterprise computing in the near future!
__________________________
https://tradernew.pro/

[FP91G]

Thanks for sharing this insightful information about DDR6 and LPDDR6 memory advancements!

The leap in bandwidth for LPDDR6 is indeed impressive, with 10.667 Gbps per-pin translating to about 28.5 GB/sec from a single LPDDR6 IC. When you consider a typical desktop PC's dual-channel memory interface, the potential 228 GB/second of memory bandwidth is a significant boost compared to the 102.4 GB/second with DDR5-6400 memory. This essentially doubles the available system memory bandwidth, which is quite remarkable.

As for regular desktop DDR6, it's exciting to see the projected data rates ranging from 8.8 Gbps up to potentially 21 Gbps. With JEDEC's draft spec expected this year and finalization anticipated by Q2 of 2025, we're on the cusp of a new era in memory technology that could substantially enhance overall system performance.

Looking forward to seeing how these advancements will impact both consumer and enterprise computing in the near future!

I have always said that the main clients of video card stores are miners. If there is nothing to mine with a good profit and the payback is approaching 2 years, then prices in stores are much lower than recommended. Most gechmers are played on old video cards, which is confirmed by the report from Steam.

[FP91G]

RAM chip manufacturer Micron has recently been making some interesting claims about its next generation of ultra-fast memory for graphics cards, GDDR7. Compared to what's currently being used (GDDR6X and GDDR6), Micron says its forthcoming tech is "expected to achieve greater than 30% improvement in frames per second for ray tracing and rasterization workloads."

That's a remarkable performance boost, no matter how you look at it, and it's normally the preserve of the significant architectural changes a new GPU design brings in. However, while it's certainly true that the data transfer rate and bandwidth of GDDR7 will be at least 30% greater than the fastest GDDR6/6X on offer right now, it's a different story altogether when it comes to actual games and applications.
https://www.pcgamer.com/hardware/graphics-cards/micron-expects-gddr7-will-improve-ray-tracing-and-rasterization-performance-by-more-than-30-compared-to-previous-gen-vram/

[safar1980]

GPUs poised for major performance enhancements under preliminary HBM4 specs
HBM4 is designed to further enhance data processing rates, offering higher bandwidth and increased capacity per die and/or stack compared to its predecessor, HBM3. It also aims to maintain lower power consumption, which is crucial for large-scale computing operations.

Technical advancements include a doubled channel count per stack compared to HBM3, a larger physical footprint, compatibility with HBM3 through a single controller, and support for 24 Gb and 32 Gb layers. There is also an initial agreement on speed bins up to 6.4 Gbps, with discussions about higher frequencies. Missing from the specs is the integration of HBM4 memory directly on processors, which Tom's Hardware says is perhaps the most intriguing part about the new type of memory.

HBM4 is particularly important for generative artificial intelligence, high-performance computing, high-end graphics cards, and servers. In particular, AI applications will benefit from the data processing and memory capabilities the standard will offer, allowing AI applications to handle larger datasets and perform complex calculations more quickly.

[FP91G]

Micron GDDR7 Memory Official: 32 Gbps Speeds, Over 1.5 TB/s Bandwidth & 30% Improvement In Game Performance Across All Resolutions

Leveraging 1β (1-beta) DRAM technology and innovative architecture, Micron GDDR7 delivers 32 Gb/s high-performance memory in a power-optimized design. With over 1.5 TB/s of system bandwidth, which is up to 60% higher bandwidth than GDDR6, and four independent channels to optimize workloads, GDDR7 memory enables faster response times, smoother gameplay, and reduced processing times.
https://wccftech.com/micron-gddr7-memory-32-gbps-speeds-over-1-5-tbps-bandwidth-30-percent-faster-game-performance-all-resolutions/

[FP91G]

The DDR4 Price Spike: What’s Behind the Surge
https://info.fusionww.com/blog/the-ddr4-price-spike-whats-behind-the-surge
In late May 2025, the memory market saw a sharp and sudden movement: DDR4 spot prices surged significantly. An 8Gb DDR4 that cost $3.20 is now hovering near $8. For 16gb chips, the leap was just as dramatic, from $5.50 to ~$16. 

But this isn’t a short-term spike. It is the result of a deeper transformation in how memory supply chains are structured and managed, and businesses need to take notice.

At Fusion Worldwide, we have been tracking these dynamics closely. In our previous article, “Memory Supply Tightens: Why Forecasts Matter More Than Inventory,” we explored the industry-wide transition toward forecast-driven procurement. What we are seeing now with DDR4 is a textbook example of that shift in action and its ripple effects are only beginning.

What’s Causing the DDR4 Price Surge?
1. The Phase Out
Leading memory manufacturers including Samsung, Micron, and SK Hynix are shifting their focus toward next-generation technologies such as DDR5, LPDDR5X, and high-bandwidth memory (HBM). As production lines are reallocated, DDR4 output has decreased more rapidly than many anticipated. With shrinking manufacturing capacity for DDR4, supply has begun to dry up faster than many anticipated.

2. Tariff Pressure and Demand Buildup
The U.S.–China trade environment added fuel to the fire. With new tariffs looming, many buyers rushed to stockpile memory before the grace period ended. This short-term demand spike compounded an already tightening supply, pushing prices sharply upward in a matter of weeks.

3. China’s Pullback and End of Oversupply
Previously, Chinese manufacturers like CXMT and Fujian Jinhua had flooded the market with low-priced DDR4, significantly undercutting global competitors and creating a period of oversupply. However, the Chinese government recently instructed these firms to halt DDR4 production. This abrupt shift has effectively ended the oversupply and immediately tightened global supply of DDR4.

[safar1980]

NVIDIA’s Next-Gen Rubin GPUs Have Reportedly Entered Production, Also Secures HBM4 Samples From All Major DRAM Manufacturers
NVIDIA's next-generation Rubin GPUs have entered production, and the company has also secured samples of HBM4 memory from all major suppliers.

NVIDIA's Rubin GPUs Will Be The Most Advanced AI Solution In 2026, Already In Production With HBM4 Samples In Hands
A few weeks ago, NVIDIA's CEO, Jensen Huang, showcased the next-gen Vera Rubin Superchip for the first time at GTC 2025 in Washington. We got to see two super massive GPUs stacked together with the next-generation Vera CPU, and loads of LPDDR memory on the outskirts. The Vera Rubin Superchip will lay the framework for the next wave of AI computing in data centers, and it looks like there are some good reports regarding the production timeline.

Related Story Hot Take: The True AI Chip Challenge for NVIDIA Isn’t from AMD or Intel — It’s Google’s TPUs Heating Up the Race
According to UDN, it is reported that during a recent visit to Taiwan, where Jensen Huang visited TSMC, he stated that the next-generation Rubin GPU is already entering the production line. This is big because just a few days back, Jensen had stated that they had received the first Rubin GPUs in the labs, so going from that to entering production in just a few days is groundbreaking.

[AdobeWriter]

Well now that Nvidia will force board partners to source their GDDR themselves, we're officially way into shit hitting the fan territory.

Man I was really hoping to build a new PC soon, but this RAM price explosion will really stop me dead in my tracks.

[safar1980]

SK Hynix Forecasts Tight Memory Supply Lasting Through 2028

SK Hynix held an internal company meeting, where the company reportedly presented some tough pills to swallow for many gamers. The company now forecasts the tight memory supply to last through 2028 for the commodity DRAM, which includes DDR5/DDR4, GDDR6/GDDR7, and LPDDR5x/LPDDR6. All the aforementioned DRAM variants are essential for PC and console components, making millions of gamers at risk of absorbing a massive price increase as a result. As memory supplier inventories deplete, production capacity will not increase to meet demand as it has in the past. This marks a departure from the usual response by memory manufacturers, who typically ramp up capacity in response to demand.

However, an interesting thing that SK Hynix has noted is that this situation will exclude the company's most advanced memory solutions-HBM and SOCAMM. These products are expected to get an additional capacity expansion, without tight supply impacting this part of the DRAM supply chain. These products are even in high demand as they are mostly consumed in higher volume than the regular commodity DRAM, due to their integration into AI products such as GPUs and servers.

[FP91G]

GIGABYTE Finally Drops Leaking Thermal Gel from New GPU Designs
GIGABYTE has confirmed that it has stopped using its controversial thermal gel on newer graphics card models. The material previously appeared on several GPUs and was intended to replace traditional thermal pads, but user reports showed that it could leak or shift under prolonged heat exposure. The issue became more visible when GPUs were mounted vertically, allowing the softened gel to move over time. The idea behind the thermal gel was to improve contact between components like VRMs or memory chips and the heatsink. In theory, this could result in better thermal transfer. In practice, some users noticed residue forming around the edges of the heatsink and PCB after months of use. While this did not immediately damage hardware, it raised concerns about long-term cooling consistency and general system cleanliness.
Link

Regular thermal pads are better, I always knew that. Liquid pads cause video memory overheating, GPU and GDDR7 failure.

[safar1980]

ASUS has officially denied recent rumors claiming the company is preparing to manufacture its own DRAM or build a dedicated memory fabrication facility. The company confirmed that it has no plans to invest in memory production, putting an end to speculation that spread rapidly across tech media in recent days. The reports originated from a regional technology website citing unnamed sources, suggesting ASUS was exploring vertical integration as a response to the ongoing global memory shortage. The story gained traction as it was echoed by multiple international outlets, largely due to heightened sensitivity around DRAM supply constraints affecting PCs, servers, and AI hardware.