Samsung Announces Industry's First GDDR7 Memory Development, 32 Gbps Speeds

[btarunr]

Samsung Electronics, a world leader in advanced semiconductor technology, today announced that it has completed development of the industry's first Graphics Double Data Rate 7 (GDDR7) DRAM. It will first be installed in next-generation systems of key customers for verification this year, driving future growth of the graphics market and further consolidating Samsung's technological leadership in the field.

Following Samsung's development of the industry's first 24 Gbps GDDR6 DRAM in 2022, the company's 16-gigabit (Gb) GDDR7 offering will deliver the industry's highest speed yet. Innovations in integrated circuit (IC) design and packaging provide added stability despite high-speed operations. "Our GDDR7 DRAM will help elevate user experiences in areas that require outstanding graphics performance, such as workstations, PCs and game consoles, and is expected to expand into future applications such as AI, high-performance computing (HPC) and automotive vehicles," said Yongcheol Bae, Executive Vice President of Memory Product Planning Team at Samsung Electronics. "The next-generation graphics DRAM will be brought to market in line with industry demand and we plan on continuing our leadership in the space."



Samsung's GDDR7 achieves an impressive bandwidth of 1.5-terabytes-per-second (TBps), which is 1.4 times that of GDDR6's 1.1TBps and features a boosted speed per pin of up to 32 Gbps. The enhancements are made possible by the Pulse Amplitude Modulation (PAM3) signaling method adopted for the new memory standard instead of the Non Return to Zero (NRZ) from previous generations. PAM3 allows 50% more data to be transmitted than NRZ within the same signaling cycle.

Significantly, in comparison with GDDR6, the latest design is 20% more energy efficient with power-saving design technology optimized for high-speed operations. For applications especially mindful of power usage, such as laptops, Samsung offers a low-operating voltage option.

To minimize heat generation, an epoxy molding compound (EMC) with high thermal conductivity is used for the packaging material in addition to IC architecture optimization. These improvements dramatically reduce thermal resistance by 70% in comparison to GDDR6, aiding in stable product performance even in conditions with high-speed operations.

View at TechPowerUp Main Site

 

[btarunr]

inb4 RTX 5060 Ti with 64-bit memory.

 

[timliu1996]

inb4 RTX 5060 Ti with 64-bit memory.

Nah, 5060Ti will still be on GDDR6X 128 bit, only 5080 and 5090 will have the "premium" GDDR7 memory

 

[N/A]

Lower than the initial 36 GBps for compatibility. this means the entire series is using it this time.
256 bit in 5080, 192 bit 768GBs in 5070 and 128 bit 512GBs in 5060 kind of.

 

[dj-electric]

Im just here to be thankful we won't have to rely on Micron for exclusive speedy GDDR6X and be locked to a single GPU vendor for it

 

[TumbleGeorge]

Lower than the initial 36 GBps for compatibility. this means the entire series is using it this time.
256 bit in 5080, 192 bit 768GBs in 5070 and 128 bit 512GBs in 5060 kind of.

I haven't read of anyone discontinuing development of faster and higher capacity chips. The article under which we are writing is an announcement by Samsung that the development of 32GT/s GDDR7 chips has ended. It seems to me that the rest of this listing is still speculation.

 

[csendesmark]

Insane, can't wait to get a puppy with GDDR7 RAMs

 

[Wirko]

DDR RAM manufacturers realised they won't be able to bring 32 Gbit chips to the market soon enough, so they introduced 24 Gbit chips ... now I'm wondering if we're going to see a similar development in GDDR RAM too.

 

[kondamin]

nice put 4 of them on package with an exynos soc

 

[AnarchoPrimitiv]

Dumb question, but when they say "16Gb GDDR7", what is the "16Gb" referring to? Storage capacity? And if it's storage capacity, could you explain a bit further?

 

[bonehead123]

So to summarize:

GPU and next-gen DDR prices will continue to rise even more as soon as these chips are available

Go figure !

 

[ixi]

So to summarize:

GPU and next-gen DDR prices will continue to rise even more as soon as these chips are available

Go figure !

If we would see x2 of performance, then why not to pay additional as it would ne huge leap.


If we see 5% then f no and totally agree with you.

 

[TranceHead]

Dumb question, but when they say "16Gb GDDR7", what is the "16Gb" referring to? Storage capacity? And if it's storage capacity, could you explain a bit further?

Memory size, there's 8 bits (b) to a byte (B) so each chip can store and buffer up to 2GB each.

 

[Wirko]

Dumb question, but when they say "16Gb GDDR7", what is the "16Gb" referring to? Storage capacity? And if it's storage capacity, could you explain a bit further?

Yes, capacity. It's always stated in bits, not bytes, for individual RAM chips, and it's true for NAND chips as well.

 

[AnarchoPrimitiv]

Memory size, there's 8 bits (b) to a byte (B) so each chip can store and buffer up to 2GB each.

Yeah, but here's what I'm not getting. In the picture below we see a representation of a GPU with it's memory, let's say the GPU has 24GB of "16Gb GDDR7" memory spread across 12 chips like in the picture, so each one of those chips is 2GB each.....so what PHYSICAL part of one of these chips is the 16Gb in "16Gb GDDR7" referring to? You understand what I'm asking?

 

[TranceHead]

Yeah, but here's what I'm not getting. In the picture below we see a representation of a GPU with it's memory, let's say the GPU has 24GB of "16Gb GDDR7" memory spread across 12 chips like in the picture, so each one of those chips is 2GB each.....so what PHYSICAL part of one of these chips is the 16Gb in "16Gb GDDR7" referring to? You understand what I'm asking?

View attachment 305441

The silicon (or transistors within) inside the chip is 16Gb (2GB), I'm not sure what you mean.

 

[Wirko]

Yeah, but here's what I'm not getting. In the picture below we see a representation of a GPU with it's memory, let's say the GPU has 24GB of "16Gb GDDR7" memory spread across 12 chips like in the picture, so each one of those chips is 2GB each.....so what PHYSICAL part of one of these chips is the 16Gb in "16Gb GDDR7" referring to? You understand what I'm asking?

It's the total number of cells that store one bit each, of course they're organised in banks and rows and columns internally but these details don't matter from a bird's view. Each RAM chip is connected to the GPU chip through a 16-bits-wide bus.

 

[TumbleGeorge]

The silicon (or transistors within) inside the chip is 16Gb (2GB), I'm not sure what you mean.

I believe the colleague is asking how the GPU "sees" and works with VRAM. Like with a whole object, or is it subdivided by the driver into separate areas, which are the physical memory chips?

 

[TranceHead]

I believe the colleague is asking how the GPU "sees" and works with VRAM. Like with a whole object, or is it subdivided by the driver into separate areas, which are the physical memory chips?

A fair assumption.

 

[AnarchoPrimitiv]

It's the total number of cells that store one bit each, of course they're organised in banks and rows and columns internally but these details don't matter from a bird's view. Each RAM chip is connected to the GPU chip through a 16-bits-wide bus.

Thank You, that's what I was asking, basically what physical aspect/part/component of a GDDR7 IC the "16Gb" is describing, and if I'm reading this correctly, the "16" in "16Gb GDDR7" is describing the "total number of cells that store one bit each".

 

[Wirko]

A related note: the standardised symbol for the byte is B, and for the bit, both b and bit are correct; I choose to use "bit" wherever there's any chance of misunderstanding.