Micron GDDR6X Increases Bandwidth and Capacity

[TheLostSwede]

Micron Technology, Inc., today announced the volume production of its new 16Gb GDDR6X memory, which is now shipping in the NVIDIA GeForce RTX 3090 Ti graphics card. The latest GDDR6X memory, available only from Micron, features twice the capacity and up to 15% higher performance over the previous 8Gb version. These additional capabilities mean end users can experience razor-sharp visuals, higher frame rates and outstanding performance in memory-intensive applications like gaming and content creation. Today Micron extends its performance leadership by providing a new graphics memory with 16Gb capacity and running at an industry-best 21 Gb/s in the GeForce RTX 3090 Ti. With a performance roadmap up to 24 Gb/s, GDDR6X is ready for data-hungry applications of the future. Additionally, Micron's innovative use of PAM4 signal techniques in GDDR6X make it more power efficient than any of the GDDR6 products publicly available.

"The industry-leading capabilities of Micron's GDDR6X memory help bring new levels of realism and performance to the most demanding applications," said Mark Montierth, vice president and general manager of High-Performance Memory and Networking at Micron. "Micron is once again at the forefront of the memory innovation powering today's highest bandwidth solutions and built with the advanced process and interface technology to enable continued graphics performance leadership."




As games and graphic rendering techniques become even more sophisticated, graphics processing units require massive amounts of data to move at incredible speeds, which can challenge system graphics memory. Virtual reality gaming, 4K and 8K monitors with 240 Hz refresh rates, and intense 3D CAD applications stretch the limits of graphics systems. The unrivaled memory bandwidth of GDDR6X enables these high-resolution experiences with ultra-fast frame rates and near-instant rendering, providing users with lifelike effects. GDDR6X capabilities, integrated with the GeForce RTX 3090 Ti graphics card, improve real-time ray tracing and neural graphics processing to deliver the immersive, cinematic effects virtual worlds demand. The GeForce RTX 3090 Ti features 24 GB of GDDR6X enabling 40 teraflops of GPU performance.

The breakthrough bandwidth of GDDR6X is possible because of Micron's groundbreaking work with PAM4 signal transmission technology, which has revolutionized how memory moves data. Micron's PAM4 signaling technique means GDDR6X increases memory bandwidth and powers system bandwidth up to 1 TB/s, making it the optimal choice for graphics applications that require exceptionally fast, lower power memory. No other memory vendor offers this capability.

The addition of 16Gb GDDR6X memory marks an important milestone in the growth of Micron's Ultra-Bandwidth Solutions portfolio, which includes GDDR6 and HBM2E. Together, these solutions comprise the industry's most comprehensive portfolio of high-performance, high-bandwidth products designed to feed data-intensive applications.

View at TechPowerUp Main Site | Source

 

[wolf]

innovative use of PAM4 signal techniques in GDDR6X

When Ampere launched I thought the whole point is that this signalling technique could double bandwidth at the same speed, now I'm not sure what the point is.

 

[R-T-B]

When Ampere launched I thought the whole point is that this signalling technique could double bandwidth at the same speed, now I'm not sure what the point is.

I guess just that they could not reach these bandwidths without it...

 

[ModEl4]

Micron 16Gb GDDR6X memory for RTX 3090Ti enters volume production, 24GB/s on a roadmap - VideoCardz.com

Micron GDDR6X Increases Bandwidth and Capacity 16Gb memory supports industry-leading speeds up to 24Gb/s, ideal for gamers and content creators BOISE, Idaho, April 12, 2022 — Micron Technology, Inc., today announced the volume production of its new 16Gb GDDR6X memory, which is now shipping in...

videocardz.com

Since we have already the 3070/3070Ti comparison to draw a conclusion, the way i see it is that 24Gbps GDDR6X needs around 1.45X more power than a 16Gbps GDDR6, that's the way I'm reading the pJ/bit figures unless I'm wrong (probably I'm wrong and it's a little bit more energy efficient than GDDR6 but why this haven't materialize in 3070/3070Ti? maybe the memory controller is the culprit for the much higher consumption?). If that's the case then the only benefit can either be cost-related (surely not!) vs a equivalent speed GDDR6 or time to market (19Gbps GDDR6X vs 16Gbps GDDR6, 21Gbps GDDR6X vs 18Gbps GDDR6, 24Gbps GDDR6X vs 20Gbps GDDR6 etc)?
EDIT: Tom's Hardware had an interesting article:

Micron Reveals GDDR6X Details: The Future of Memory, or a Proprietary DRAM?

Micron discusses graphics memory: GDDR6X is here, PAM4 & PAM8 for HBM patented

www.tomshardware.com

So according to this GDDR6X costs higher than GDDR6 to manufacture but it offers 15% better power efficiency per device (device refers to the memory ICs?)
Still why 3070Ti needs so much more power vs 3070?

 

[watzupken]

More power efficient than GDDR6? I feel Micron is just comparing power requirement vs memory speed. It is true that the RTX 3070 got a significant memory bandwidth upgrade moving to GDDR6X in the form of RTX 3070 TI. But in real life performance, for that substantial amount of power required to get that kind of memory bandwidth, and end up with single digit % performance improvement is not efficient at all. The jump from 14Gbps to 19Gbps only sounded great on paper.

 

[mouacyk]

Just say 2GB chips. 16Gb is too much.

 

[R-T-B]

Just say 2GB chips. 16Gb is too much.

Chip industry tends to measure in Gb's. This is chip industry PR. So Gb's is to be expected.