Samsung basically announces 24GB 5080 GPU version

BlackTron

Member
yBmmjjJ.gif
 

Makoto-Yuki

Banned
Hopefully it has a decent improvement over 5080. Some rumors suggest it won't be. I'll probably just go for the 5090 to be safe.
 

Lorianus

Member
So this is a announcement that those chips are existing but 50xx is set to launch in early 2025 no ?, so the boards for the cards would already be long in production by now, maybe with a super revision down the line i think.
 
Last edited:

GHG

Gold Member
It changes a lot for the prospects of buying a 5080. It is far more future proofed with the extra memory and the lack thereof made Nvidia's first version a dubious investment.
The lack of memory on their mid-lower tier cards is the first bottleneck people encounter which forces them to upgrade, especially so with how far technologies like frame gen and AI upscaling have come, and they know this.
 

Bojji

Member
It changes a lot for the prospects of buying a 5080. It is far more future proofed with the extra memory and the lack thereof made Nvidia's first version a dubious investment.

Yeah, more future proof fot sure but it will be also more expensive. We don't know what will come first, lack of Vram of 5080 16GB or lack of power of 5080 24GB.
 

BlackTron

Member
Yeah, more future proof fot sure but it will be also more expensive. We don't know what will come first, lack of Vram of 5080 16GB or lack of power of 5080 24GB.

Nvidia has set things up so if you want 24GB, you need the most insane card they have a 5090. Go down one tier and you have to deal with 16. They're playing a game with gamers.

I can see a lot of people willing to pay for the memory without needing the immense horsepower of 5090.
 

CuNi

Member
Nvidia has set things up so if you want 24GB, you need the most insane card they have a 5090. Go down one tier and you have to deal with 16. They're playing a game with gamers.

I can see a lot of people willing to pay for the memory without needing the immense horsepower of 5090.

Most leaks and rumors point for the 5090 to be 32GB tho afaik.
 

BlackTron

Member
Most leaks and rumors point for the 5090 to be 32GB tho afaik.

You're right I was just thinking wait wasn't it really DOUBLE? 24 is really probably just the natural sweet spot for a 5080. 5090 parity isn't required but we don't need an artificial nerf either.
 
This is extremely ignorant of me to say, but isn’t 16 GB enough? I imagine not many games are RAM bound especially at those numbers.
 

StereoVsn

Gold Member
Cute that people think this is going on their gaming GPU's.

It's for their AI products.
It will go to their gaming GPUs, but they will wait for Super or Ti revisions a year (give or take a few months) after initial launch.
 

john2gr

Member
What a clickbait. Samsung didn't announce anything related to the 5080. There is simply a new GDDR7 memory chip. What makes you think that it will be used in the RTX 5080? If you've known NVIDIA, you'd already know that they might use it for a 5080Ti or 5080 SUPER (and, once again, that's a guess and nothing more).
 

SF Kosmo

Banned
24GB 5080 not gonna happen.
Maybe a year later the 5080ti, or a two tiered 5090.
Well I think the implication here is that this could end up in third party versions of the card, not the reference versions (which are already set), because it can essentially slot into existing designs and it makes an easy differentiators for these cards which are generally all close to identical other than cooling and tweaks to clock speed.
 
Last edited:

hlm666

Member
Cute that people think this is going on their gaming GPU's.

It's for their AI products.
Don't they use hbm for those? I remember the specs for gb200 said something about 8TB/s of memory using hbm3e memory.

edit: quick check and they have been using hbm variants since hopper (h100), this memory isn't for them bandwidth probably too low.
edit 2: the tech specs for GB200 say up to 16TB/s hbm memory bandwidth and that article said the 5080 would have 1024GB/s bandwidth with that memory for comparison.
 
Last edited:

Three

Member
Don't they use hbm for those? I remember the specs for gb200 said something about 8TB/s of memory using hbm3e memory.

edit: quick check and they have been using hbm variants since hopper (h100), this memory isn't for them bandwidth probably too low.
edit 2: the tech specs for GB200 say up to 16TB/s hbm memory bandwidth and that article said the 5080 would have 1024GB/s bandwidth with that memory for comparison.

"With its high capacity and powerful performance, the 24Gb GDDR7 will be widely utilized in various fields that require high-performance memory solutions, such as data centers and AI workstations, extending beyond the traditional applications of graphics DRAM in graphics cards, gaming consoles and autonomous driving."
 
Last edited:

Magic Carpet

Gold Member
From the article-
"We’re not expecting to see memory speeds as high as 42.5Gb/s, but recent rumors have put the RTX 5080 VRAM speed at 32Gbps, which is over 52% quicker than the 21Gbps GDDR6X VRAM used in the RTX 4090. If true, this would mean the RTX 5080 has a total memory bandwidth of 1,024GB/s, despite it being rumored to only have a relatively narrow 256-bit interface."

I think I have enough saved up for a 5080, But I'm going for the stretch goal of 5090.
 

hlm666

Member

"With its high capacity and powerful performance, the 24Gb GDDR7 will be widely utilized in various fields that require high-performance memory solutions, such as data centers and AI workstations, extending beyond the traditional applications of graphics DRAM in graphics cards, gaming consoles and autonomous driving."
So tell me what hardware this is going in, none of nvidia's use gddr. Last gen (hopper) and this gen (blackwell) use HBM. How are you going to sell AI hardware to compete with nvidia if your going to use gddr vs their hbm solutions??? Secondly my response was to a comment saying nvidia were putting this in their AI solutions and thus we wont see it in game gpus.
 
Top Bottom