Samsung unveils 24Gb GDDR7 memory — up to 42.5 Gbps modules with 30% higher efficiency
Mass production is planned for early next year.
Samsung has introduced the "Industry's First" 24Gb (3GB) GDDR7 memory, offering exceptionally high speeds of up to 42.5 Gbps to power next-generation GPUs. Owing to many refinements and updates, Samsung asserts a 50% density improvement alongside a 30% boost in efficiency versus the previous generation. Validation of these 24Gb GDDR7 modules is ongoing with "major" GPU manufacturers and mass production is planned for early next year.
Last July, Samsung unveiled the world's first GDDR7 memory, rated at a modest 32 Gbps and with capacities of 16Gb (2GB) per module. The new ICs net Samsung a 25% improvement in bandwidth against its already-announced GDDR7 modules, and a 2.36x improvement against its last-generation 18 Gbps GDDR6 offerings. In terms of manufacturing, Samsung's 24Gb GDDR7 memory utilizes the company's 5th generation 10nm node, resulting in a 50% density increase at the same package size.
To achieve these never-before-seen heights, Samsung has incorporated PAM3 signaling which makes use of 3 signals (-1, 0, +1) to transmit data. These new memory modules can easily hit 40 Gbps, though, under specific conditions, 42.5 Gbps is also possible, almost 80% faster than what GDDR6X delivers (24 Gbps).
With great power comes great responsibility and to address the concerns of efficiency, Samsung touts their new GDDR7 memory as 30% more efficient than last-generation thanks to technologies such as clock control management and a dual VDD design. Furthermore, the issue of current leakage is resolved by power gating; shutting down parts of the chip that aren't in use.
All that jargon aside, it is important we address the elephant in the room. Last-generation GDDR6/X memory featured 16Gb (2GB) memory modules. The fact that Samsung's GDDR7 offerings have 1.5x more memory per module should result in higher VRAM capacities for next-generation GPUs. Here is a brief overview of what you can expect if GPU manufacturers employ 42.5 Gbps 24Gb (3GB) modules at different bus widths:
- 128-bit: 12GB of VRAM at 680 GB/s
- 192-bit: 18GB of VRAM at 1,020 GB/s
- 256-bit: 24GB of VRAM at 1,360 GB/s
- 320-bit: 30GB of VRAM at 1,700 GB/s
- 512-bit: 48GB of VRAM at 2,720 GB/s
This should come as great news for gamers, as the last generation of GPUs was held back by a lack of VRAM. Even 1080p nowadays can require more than 8GB of memory, so we hope that GPU manufacturers don't cheap out on the memory capacity next generation. Not to sound like a wet blanket, but don't keep your hopes up since entry-level GPUs next generation might stick with readily available 16Gb GDDR6 modules as they would be cheaper to produce, but that's just speculation.
Stay On the Cutting Edge: Get the Tom's Hardware Newsletter
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.
Hassam Nasir is a die-hard hardware enthusiast with years of experience as a tech editor and writer, focusing on detailed CPU comparisons and general hardware news. When he’s not working, you’ll find him bending tubes for his ever-evolving custom water-loop gaming rig or benchmarking the latest CPUs and GPUs just for fun.
-
RUSerious
I didn't see anything about volume production. Makes me wonder if all the leakers are wrong and the 5080 will have 24GiB GDDR7 and the 5090 will have 48GiB. Certainly would be a nice step up for the 5080 compared to the 4080.Pierce2623 said:Well there’s your 24GB 5080ti for holiday season ‘25. -
Notton says mass production next year.Reply
I doubt we'll see any cards using it before March/April. -
Mama Changa
No, there will never be desktop 48GB 5090. That would be for Quadro. The 5080 will be 16GB, Nvidia isn't going to rush out and use Samsung, they just told them they won't even use their HBM3e. We won't see 24GB 5080 Ti until Micron or SK Hynix have 3GB dies IMO.RUSerious said:I didn't see anything about volume production. Makes me wonder if all the leakers are wrong and the 5080 will have 24GiB GDDR7 and the 5090 will have 48GiB. Certainly would be a nice step up for the 5080 compared to the 4080. -
jp7189
You can be pretty sure the consumer cards won't increase vram emplacements or bit width. So, 5090 with 24GB and 5080 with 16GB, and after 3GB module hit volume production, 36GB and 24GB ti versions as that requires them to do exactly 0 rework of the cards.RUSerious said:I didn't see anything about volume production. Makes me wonder if all the leakers are wrong and the 5080 will have 24GiB GDDR7 and the 5090 will have 48GiB. Certainly would be a nice step up for the 5080 compared to the 4080. -
RUSerious
Doh! Thanks.Notton said:says mass production next year.
I doubt we'll see any cards using it before March/April. -
kerberos_20 hynix started production already (july 30), but it offers "just" 32gbps now and 40gbps later, still better than gddr6xReply
so its possible to see 50 series sooner with hynix and later some ti/super variants with samsung, rummors shows Q1 25 for RTX 50xx, which doesnt give much time frame for some QA with new samsung chips to hit some volume targets -
Pierce2623
I read on another site that mass production begins q1 ‘25. That’s why I said it’s a 5080ti instead of a the 5080.RUSerious said:I didn't see anything about volume production. Makes me wonder if all the leakers are wrong and the 5080 will have 24GiB GDDR7 and the 5090 will have 48GiB. Certainly would be a nice step up for the 5080 compared to the 4080. -
RUSerious
Yeah, I was overly optimistic. So, as you and others have pointed out - there will be a Ti or Super a year later with possible memory upgrades.Pierce2623 said:I read on another site that mass production begins q1 ‘25. That’s why I said it’s a 5080ti instead of a the 5080.