Samsung, SK Hynix, and Micron Compete for GDDR7 Dominance

[Nomad76]

Competition among Samsung, SK Hynix, and Micron is intensifying, with a focus on enhancing processing speed and efficiency in graphics DRAM (GDDR) for AI accelerators and cryptocurrency mining. Compared with High Bandwidth Memory (HBM), the GDDR7 has a faster data processing speed and a relatively low price. Since Nvidia is expected to use next-generation GDDR7 with its GeForce RTX50 Blackwell GPUs, competition will likely be as strong as the demand. We can see that by looking, for example, at the pace of new GDDR7 releases from the past two years.

In July 2022, Samsung Electronics developed the industry's first 32 Gbps GDDR7 DRAM, capable of processing up to 1.5 TB of data per second, a 1.4 times speed increase and 20% better energy efficiency compared to GDDR6. In February 2023, Samsung demonstrated its first GDDR7 DRAM with a pin rate of 37 Gbps. On June 4, Micron launched its new GDDR7 at Computex 2024, with speeds up to 32 Gbps, a 60% increase in bandwidth, and a 50% improvement in energy efficiency over the previous generation. Shortly after, SK Hynix introduced a 40 Gbps GDDR7, showcased again at Computex 2024, doubling the previous generation's bandwidth to 128 GB per second and improving energy efficiency by 40%.





GDDR vs HBM memory

GPUs with GDDR memory:

• More common and easier to find
• Cheaper because the memory is attached to the circuit board, not the chip
• Fast enough for most regular uses
• Use more power and are less efficient

GPUs with HBM memory:

• Less common and harder to find
• Expensive, usually in top-end products like the H100
• Used for special tasks that need a lot of data processing power
• More efficient and can handle more data at once

According to research made by Omdia, it is expected that graphics DRAM will account for 15% of the entire DRAM market by the end of this year, from 7% in 2022. Whether we like it or not, the AI boom and companies' rush towards these new technologies will largely dictate the evolution of the market for the next two or three years.

View at TechPowerUp Main Site | Source

 

[bonehead123]

Whether we like it or not, the AI boom and companies' rush towards these new technologies will largely dictate the evolution of the market for the next two or three years.

Well.... D.U.H. !

Hello Omdia, this is 2022 calling, and we want ALL our outdated, AI-generated PR's back, like, yesterday

 

[Denver]

SK Hynix dominates the HBM market.
Samsung dominates the memory market as a whole.
Micron dominates the litigation market:

 

[Nomad76]

SK Hynix dominates the HBM market.
Samsung dominates the memory market as a whole.
Micron dominates the litigation market:

AI wins with GDDR7 help

Well.... D.U.H. !

Hello Omdia, this is 2022 calling, and we want ALL our outdated, AI-generated PR's back, like, yesterday

2023 was a bit dull imho

 

[AnarchoPrimitiv]

"32 Gbps GDDR7 DRAM"

What is the 32 Gbps measuring? 32 Gbps per what? Per IC? Per Pin/connection? Am I the only one that feels that when it comes to DRAM specifically the manufacturers use very ambiguous and vague terms....just like with their manufacturing nodes....they call them 1β for example, what's that about? How am I supposed to compare that to other manufacturers' processes?

Back to the 32Gbps...I feel like these speeds are thrown around a lot, especially when discussing GPUs, but they never specific to what they refer....maybe I'm just the only one

 

[persondb]

"32 Gbps GDDR7 DRAM"

What is the 32 Gbps measuring? 32 Gbps per what? Per IC? Per Pin/connection? Am I the only one that feels that when it comes to DRAM specifically the manufacturers use very ambiguous and vague terms....just like with their manufacturing nodes....they call them 1β for example, what's that about? How am I supposed to compare that to other manufacturers' processes?

Back to the 32Gbps...I feel like these speeds are thrown around a lot, especially when discussing GPUs, but they never specific to what they refer....maybe I'm just the only one

It's per pin, so a 256-bit bus would have around 1 TB/s of bandwidth.

So essentially it's the signaling rate, since those devices can be configured to have any arbitary bus width.

You aren't supposed to compare it to other manufacturer nodes too, those are just purely marketing name. The technical details to compare manufacturing nodes for memory would probably go over the head of most people(myself included).

 

[bonehead123]

How am I supposed to compare that to other manufacturers' processes

Apparently, you missed that memo, the one that says: "Absolutely ZERO comparisons allowed, we are 100% authorized to confuse and commiserate consumers, and there aint nuthin you can do about it" /s

 

[AnarchoPrimitiv]

Apparently, you missed that memo, the one that says: "Absolutely ZERO comparisons allowed, we are 100% authorized to confuse and commiserate consumers, and there aint nuthin you can do about it" /s

Hahaha!

It's per pin, so a 256-bit bus would have around 1 TB/s of bandwidth.

So essentially it's the signaling rate, since those devices can be configured to have any arbitary bus width.

You aren't supposed to compare it to other manufacturer nodes too, those are just purely marketing name. The technical details to compare manufacturing nodes for memory would probably go over the head of most people(myself included).

Thanks, would it be so hard to include the "per pin" part? I'd like to know the difference between nodes...I see enthusiasts talking about hynix vs samsung vs micron ICs all the time and I feel like it's all just a bunch of alchemy/psuedo-science without actually detailing what makes them different...either way, thanks for the answer!

 

[N/A]

I can't stand AI generated images when there is nothing real in them.
also boasting 40Gbps for prestige, only 28 at first. 60 series for the real thing.