Rumor: 24 GB 5080 coming from MSI

V

vjhawk

Gawd
2FA
Joined
Sep 2, 2016
Messages
671
Interesting that this new MSI 24 GB 5080 has been spotted.

Unless there's a CUDA count increase, I still don't see it beating the 4090.

This generation's 5080 is too core starved to even beat last generation's flagship.
Article:
4080 24 GB Leak.

I expect they are just using the 3GB memory modules instead of the 2GB modules to achieve this RAM increase and there won't be a bus width increase on this edition of the 5080 either.

Image:
1742877474975.png


My other theory is Nvidia originally intended the 5080 to be a 24 GB card when it first briefed its AIBs but later decided to cut down on the final VRAM to save money and to avoid using 3GB modules or avoid a bigger bus width.
 
3GB modules could have been simply not ready on time, making that decision easy.... it is not like you can increase the memory controller to accomodate that larger bus. That MSI memory table could always be simply a typo... (was quickly changed)
 
Last edited:
LukeTbk said:
3GB modules could have been simply not ready on time, making that decision easy.... it is not like you can increase the memory controller to accomodate that larger bus. That MSI memory table could always be simply a typo... (was quickly changed)
Click to expand...
This could easily be the case. Supposedly they weren't available at all until early this year. They're likely only available in limited quantities. So far NV is only using them on the laptop 5090 to give it 24GB. It's the same chip as the desktop 5080.

I bet we will see a lot more of the 3GB GDDR7 chips at the refresh. 24GB 5080 Super and 18GB 5070 Super. Maybe also a 12GB 5060 and/or 5060Ti.
 
vjhawk said:
Interesting that this new MSI 24 GB 5080 has been spotted.

Unless there's a CUDA count increase, I still don't see it beating the 4090.

This generation's 5080 is too core starved to even beat last generation's flagship.
Article:
4080 24 GB Leak.

I expect they are just using the 3GB memory modules instead of the 2GB modules to achieve this RAM increase and there won't be a bus width increase on this edition of the 5080 either.

Image:View attachment 719153

My other theory is Nvidia originally intended the 5080 to be a 24 GB card when it first briefed its AIBs but later decided to cut down on the final VRAM to save money and to avoid using 3GB modules or avoid a bigger bus width.
Click to expand...
Yes just a RAM bump for extra sales cost rewarding MSI for joining the Nvidia only club.
Of course it will add next to nothing for gamers but will garner MSI some flagship prestige. Welcome your new EVGA?
 
funkydmunky said:
Yes just a RAM bump for extra sales cost rewarding MSI for joining the Nvidia only club.
Of course it will add next to nothing for gamers but will garner MSI some flagship prestige. Welcome your new EVGA?
Click to expand...
MSI has a step up program and the best GPU warranty service in the US?

If not, they're not even close to what EVGA was.
 
arestavo said:
MSI has a step up program and the best GPU warranty service in the US?

If not, they're not even close to what EVGA was.
Click to expand...
Plus their own bustling forums with regularly-posting community reps? :)
 

“Newer GeForce RTX 5080 With 24 GB VRAM Rumoredly Hitting Shelves in Q1 2027​

As of now, we have only heard about the coming of a new variant, which will supposedly bring higher VRAM capacity, but no specific launch date has been confirmed. Even the discontinuation of the RTX 5090 and RTX 5080 has not been officially declared by NVIDIA. So, take this information with a grain of salt. As per Benchlife, the new RTX 5080 with 24 GB VRAM will arrive in Q1 2027.

On the other hand, NVIDIA plans to launch the GeForce RTX 5080 24GB GDDR7 graphics card in the first quarter of 2027, with the aim of replacing the current GeForce RTX 5080 16GB GDDR7
- Benchlife
Click to expand...
The publication didn't quote a public source and might have received this information from one of its own sources. According to the publication, NVIDIA has already informed its board partners about the new RTX 5080 24 GB GPU, but as far as the development information is concerned, board partners apparently don't know much about it. We don't know if NVIDIA will make any other changes to the card except for the memory configuration, but that's what we expect.


Since the current RTX 5080 variant features 16 GB of memory capacity using 8x 2 GB GDDR7 modules, the new configuration likely includes the new 3 GB GDDR7 memory chips, enabling 24 GB of VRAM capacity. It's possible that the new variant NVIDIA is launching might have been planned as an RTX 5080 "SUPER" previously, but since memory allocation was a big challenge, NVIDIA decided to cancel the Super series.

Nonetheless, expect the new RTX 5080 24 GB to cost more, as securing 3 GB GDDR7 modules costs more, and with the current challenges, the GPU could easily sell for around $2,000 when entering the market.

News Source: Benchlife“

Source: https://wccftech.com/nvidia-rtx-5080-24-gb-gpu-expected-to-arrive-in-q1-2027/
 
The problem is how they'll price the MSRP. The MSRP of the RTX 5090 32GB was $1,999.00, but is being sold at retailers for ~$8,999.99 (Newegg).

The MSRP of the RTX 5080 16GB was $999.99, so I would guess that the MSRP of a RTX 5080 24GB would be something like $1299.99. However, retailers would just end up selling them for $4,999.99 or more, depending on how much crazier the market gets.
 
MSRP means nothing at this point. Normal 5080 are going for at least $1600. The 24gb model will be $2600 to start.
 
Unfortunately I feel like a 24GB 5080 will do nothing but feed the AI craze for now. You don't really need more than 16GB of VRAM for games. I mean sure, there are edge cases with modding where it can be useful but really it isn't a thing where with a 16GB card you are normally saying "Man, I could turn up the quality if only I had more VRAM." However for LLMs, well 8 more GB matters quite a bit. So not only will the cost of RAM push the price on this up, but I think people buying it for LLMs will as well.

I work at a research university and we have researchers buying 5090s all the time because they want something with as much VRAM as possible, but don't want to pay the Quadro prices. With 5090s now costing like Quadros do (and possibly getting discontinued entirely if rumors are true) they'll jump on these. That kind of thing pushes prices up.

Really as gamers you almost want the opposite: A Card with a bigger core like a 5090, but only 16GB of VRAM so the LLM people leave it alone.
 
Sycraft said:
Unfortunately I feel like a 24GB 5080 will do nothing but feed the AI craze for now. You don't really need more than 16GB of VRAM for games. I mean sure, there are edge cases with modding where it can be useful but really it isn't a thing where with a 16GB card you are normally saying "Man, I could turn up the quality if only I had more VRAM." However for LLMs, well 8 more GB matters quite a bit. So not only will the cost of RAM push the price on this up, but I think people buying it for LLMs will as well.

I work at a research university and we have researchers buying 5090s all the time because they want something with as much VRAM as possible, but don't want to pay the Quadro prices. With 5090s now costing like Quadros do (and possibly getting discontinued entirely if rumors are true) they'll jump on these. That kind of thing pushes prices up.

Really as gamers you almost want the opposite: A Card with a bigger core like a 5090, but only 16GB of VRAM so the LLM people leave it alone.
Click to expand...

That won't stop them because they would just add more memory themselves like they have been doing. There are 4090's with 96GB VRAM.
 
We all know that NVIDIA plans to launch the GeForce RTX 5080 24GB to replace the current GeForce RTX 5080 16GB version, but the launch time is still unclear.

news that NVIDIA plans to launch GeForce RTX 5080 24GB in January 2027 actually has internal sounds coming out as early as September, but according to the current schedule, NVIDIA GeForce RTX 5080 24GB should not happen in January. As for when NVIDIA will launch GeForce RTX 5080 24GB, this depends on when AIC partners will receive the design information given by NVIDIA.

https://benchlife.info/nvidia-geforce-rtx-5080-24gb-might-not-launch-at-q1/
 
  • Like
Reactions: erek
like this

“Board Partners Reportedly Still Await NVIDIA GeForce RTX 5080 24 GB GPU Kits​

by AleksandarK Today, 05:00 Discuss (1 Comment)
NVIDIA's board partners reportedly still lack the supply of GeForce RTX 5080 24 GB design kits, making a nearby launch seemingly impossible. According to BenchLife, design kits have not yet been provided to NVIDIA's add-in card partners. These design kits typically include a GPU die and memory modules shipped directly from NVIDIA. NVIDIA negotiates GDDR7 module pricing and supply while also receiving GPU die shipments from TSMC. The company then packages these together and sends them to board partners for assembly on their custom PCBs and designs. Since this process takes time to validate and test new designs, board partners might be unable to meet the previously assumed launch in January.

Earlier rumors suggested that NVIDIA's GeForce RTX 5080 24 GB might arrive sooner than expected, with a post from Board Channels' ChannelGate account claiming NVIDIA is speeding up the launch, with availability now expected around January 2027. Apparently, NVIDIA has already cleared out its RTX 5080 16 GB GPU inventory, and the remaining stock should last around two months. However, since board partners aren't receiving those GPU kits yet, this timing might be impossible to meet, and the window would be around Q1 2027, but not in January.“
 
The supply will be very limited and since it is 24gb AI assholes will eat them up. Expect nothing less then $3k starting.
 
I still want to know what nvidia does with all the GB202 dies that don't cut it for 5090. Do they literally just bin them?

I can only assume they're going into data centres in some form, as a 5080ti with ~15,000 cores would make perfect sense than nothing at all.
 
spine said:
I still want to know what nvidia does with all the GB202 dies that don't cut it for 5090. Do they literally just bin them?
Click to expand...
5090 are already a bit cut down version of the GB202, TPU is a nice rapid way to see all the sku's for a die, you can order them by core counts:

https://www.techpowerup.com/gpu-specs/nvidia-gb202.g1072

24,576 possible cores, top of the lines RTX 6000 has virtually all of them enabled (24,064)

6000: 24,064
5090/5500: 21,760

For not good enough for a 5090-5500 you go down into the PRO 5000 (14,080) where they are cut in half or or the 6000D (19,968)
 
Last edited:
As an Amazon Associate, HardForum may earn from qualifying purchases.
You must log in or register to reply here.
Back
Top