SK Hynix Targets Q1 2025 for GDDR7 Memory Mass Production

[AleksandarK]

The race is on for memory manufacturers to bring the next generation GDDR7 graphics memory into mass production. While rivals Samsung and Micron are aiming to have GDDR7 chips available in Q4 of 2024, South Korean semiconductor giant SK Hynix revealed at Computex 2024 that it won't kick off mass production until the first quarter of 2025. GDDR7 is the upcoming JEDEC standard for high-performance graphics memory, succeeding the current GDDR6 and GDDR6X specifications. The new tech promises significantly increased bandwidth and capacities to feed the appetites of next-wave GPUs and AI accelerators. At its Computex booth, SK Hynix showed off engineering samples of its forthcoming GDDR7 chips, with plans for both 16 Gb and 24 Gb densities.

The company is targeting blazing-fast 40 Gbps data transfer rates with its GDDR7 offerings, outpacing the 32 Gbps rates its competitors are starting with on 16 Gb parts. If realized, higher speeds could give SK Hynix an edge, at least initially. While trailing a quarter or two behind Micron and Samsung isn't ideal, SK Hynix claims having working samples now validates its design and allows partners to begin testing and qualification. Mass production timing for standardized memories also doesn't necessarily indicate a company is "late" - it simply means another vendor secured an earlier production window with a specific customer. The GDDR7 transition is critical for SK Hynix and others, given the insatiable demand for high-bandwidth memory to power AI, graphics, and other data-intensive workloads. Hitting its stated Q1 2025 mass production target could ensure SK Hynix doesn't fall too far behind in the high-stakes GDDR7 race, with faster and higher-density chips to potentially follow shortly after volume ramp.



View at TechPowerUp Main Site | Source

 

[Chaitanya]

So thats when we will see next gen "flagship".

 

[Wirko]

Eight of these GDDR7 chips will achieve the same bandwidth as one HBM3e stack, or roughly one-and-a-half HBM3 stack in the MI300, or two HBM3 stacks in the H100 or H20.

Why are some people still dreaming about HBM in consumer products?

 

[R0H1T]

You're talking about a 512bit wide bus? Also HBM as the name implies is higher "bandwidth" & stackable, it's also more efficient IIRC so there's that.

 

[Philaphlous]

Eight of these GDDR7 chips will achieve the same bandwidth as one HBM3e stack, or roughly one-and-a-half HBM3 stack in the MI300, or two HBM3 stacks in the H100 or H20.

Why are some people still dreaming about HBM in consumer products?

Motherboard room is more and more valuable in laptops. I see HBM as a valuable product in the mobile market mostly... Maybe not so much on dGPU's for desktops but certainly for laptops... I'd much rather have 1 more m.2 SSD slot with HBM in a laptop than just a single m.2 slot and GDDR7 for the same surface area...but that may come in...2030...

 

[Denver]

Now that's the kind of marketing material you'd expect a memory company to show.

It seems that the availability of GDDR7 will put a brake on the train of new GPU launches.

 

[robert3892]

Then I assume NVIDIA will look to Micron or Samsung for GDDR7 memory for their next-gen high-end consumer GPUs. This could leave SK Hynix with having to supply GDDR7 from mid to low-range GPUs from NVIDIA.

 

[Random_User]

I wonder if the market for gaming video cards would still be exist, by the time the GDDR7 will enter mass production.

 

[wNotyarD]

It seems that the availability of GDDR7 will put a brake on the train of new GPU launches.

Not so much, no. Samsung and Micron are on schedule with NVIDIA.

 

[Tomorrow]

Eight of these GDDR7 chips will achieve the same bandwidth as one HBM3e stack, or roughly one-and-a-half HBM3 stack in the MI300, or two HBM3 stacks in the H100 or H20.

Why are some people still dreaming about HBM in consumer products?

At the same power? Unlikely.

More likely is that one stack of HBM3e consumes much less power than eighth GDDR7 chips while taking up significantly less space.
To say nothing of power to move data across the PCB and a wide enough bus for those GDDR7 chips that is not cheap.

People take examples of a few cards released nearly a decade ago on not so great architecture as examples why HBM is no good. Not realizing that it took ~2,5 years, card that costs more than double with 52% higher board TDP to equal (not surpass) the last consumer HBM2 cards bandwidth.

And you ask why people keep dreaming of HBM on consumer cards? Because we could have ~2027 bandwidth today with twice the capacity and half the TDP.
At least on enthusiast cards that already cost a lot of money.

 

[TumbleGeorge]

You're talking about a 512bit wide bus? Also HBM as the name implies is higher "bandwidth" & stackable, it's also more efficient IIRC so there's that.

Eight chips, so he means the 5080 with its 256 bus width.

 

[Minus Infinity]

So thats when we will see next gen "flagship".

No, 5080 and 5090 are releasing this year, so Nvidia must be using say Micron or Samsung as others are pointing out

 

[10tothemin9volts]

Personally, next to the +50% bandwidth improvement, I'm interested in the new 3GB densities (if the 3GB densities don't consume more power than the 2GB ones). AFAIK every chip consumes a few Watt of power. Say 32GB VRAM or more in consumer cards will happen at some point, that would need 16 2GB chips. Powering 16 chips may lead to higher idle and overall power consumption. This and the space on the PCB are something like the lowest common denominators.
Other than that: It may also mean avoiding a more expensive VRAM interface of 448 bit (the 28GB rumor) or 512 bit (16 2GB chips * 32bit per chip), and instead 384 bit (12 3GB chips * 32bit per chip) could be used, although the ~+50% bandwidth would be satisfied, I don't think the +50% increase in VRAM (36GB) in consumer cards in the upcoming gen is going to be a thing.
That said, 3GB densities would indeed allow for consumer ~"AI Extra High VRAM" (TM) card products (no clamshell).