News Nvidia may release the RTX 5080 and 5070 Super with boosted memory configurations according to leaker

[hannibal]

Nvidia needs to worry about releasing the 5090, 5080and 5070 regular versions before they release the super versions? When are those coming out?

Somewhere 2026…

 

[usertests]

Nvidia needs to worry about releasing the 5090, 5080and 5070 regular versions before they release the super versions? When are those coming out?

Some of these cards, like the 5070 and 5060 Ti 8GB, are not exactly flying off the shelves. For example, the 5070 was available below MSRP in the UK, while it's still a bit above MSRP in the US.

If we take this as a leak and not idle speculation, these releases are still likely going to be months away.

 

[acadia11]

Some of these cards, like the 5070 and 5060 Ti 8GB, are not exactly flying off the shelves. For example, the 5070 was available below MSRP in the UK, while it's still a bit above MSRP in the US.

If we take this as a leak and not idle speculation, these releases are still likely going to be months away.

It was sarcasm … stating I’m still waiting for thr 5090 release …

 

[usertests]

It was sarcasm … stating I’m still waiting for thr 5090 release …

I heard your sarcasm and steamrolled over it. Supply issues don't matter for cards that don't even sell.

 

[acadia11]

I heard your sarcasm and steamrolled over it. Supply issues don't matter for cards that don't even sell.

I see your steamrolled and raise you a “sold out” sign.

 

[usertests]

I see your steamrolled and raise you a “sold out” sign.

I have airdropped a pallet of 8 GB graphics cards to your location. Enjoy the paperweights.

 

[YSCCC]

Some of these cards, like the 5070 and 5060 Ti 8GB, are not exactly flying off the shelves. For example, the 5070 was available below MSRP in the UK, while it's still a bit above MSRP in the US.

If we take this as a leak and not idle speculation, these releases are still likely going to be months away.

That was exactly what I am hoping to see, and the card that fly off the shelves are... AMD 9070XT. So hope them finally slash off the stupid 70%+ margin and finally get something accetably mid-high end capable for what a gaming GPU should do... gaming

 

[Notton]

I wouldn't be so sure. They might just decide to "gift" AI users with a 48 GB Super, but we'll see.

I doubt 48GB for AI would be a thing.
From what I'm seeing, people who want to run a LLM are going big on VRAM, and they have the option of a Ryzen AI Max+ 395 which can be configured for 96GB VRAM with the 128GB model.

From a pricing perspective, Nvidia will gladly sell you a 96GB RTX 6000 Pro or 3x 32GB 5090.
But a 48GB 5090 would have to be priced at $3000 MSRP ($5000 street price) for it to make any profiteering sense.
And if it's going to cost $10k to get 96GB, you might as well get a 96GB RTX 6000 Pro so you don't have to deal with a multiple GPU setup.

 

[thestryker]

I think it's worth pointing out that (AFAIK), they never sold RTX 4080 with GA102 dies, which suggests that yields on the "4N" node were quite good.

They sold RTX 4070 Ti Supers with AD102.

 

[beyondlogic]

Shocked. No one expected this. Absolutely no one.


Except every tech youtuber in the world.

lol no one needed a crystal ball for that it was no accident that the first generation were awful.

 

[bit_user]

I doubt 48GB for AI would be a thing.

They sold Titan RTX, which was basically just a slightly up-spec'd RTX 2080 Ti with double the RAM. The cost was also doubled, but it was cheaper than their workstation cards. So, maybe they could do another Titan?

From what I'm seeing, people who want to run a LLM are going big on VRAM, and they have the option of a Ryzen AI Max+ 395 which can be configured for 96GB VRAM with the 128GB model.

You could run two 48 GB cards, and people would do it if they were cheaper than one 96 GB card. Two cards could be faster than one, depending on how well the model is split between them.

 

[acadia11]

I have airdropped a pallet of 8 GB graphics cards to your location. Enjoy the paperweights.

Received, appreciated, and now running Crysis!

 

[Notton]

They sold Titan RTX, which was basically just a slightly up-spec'd RTX 2080 Ti with double the RAM. The cost was also doubled, but it was cheaper than their workstation cards. So, maybe they could do another Titan?


You could run two 48 GB cards, and people would do it if they were cheaper than one 96 GB card. Two cards could be faster than one, depending on how well the model is split between them.

If they could reuse the same PCB for the RTX Pro 6000 96GB, it would result in a RTX 5090Ti 64GB with 32x2GB chips.

 

[RUSerious]

A 5080 with 24 GB of VRAM has my attention

Same, mainly because my 4070 Ti has run out of VRAM on a couple of games forcing to drop graphics settings lower than I'd like (probably poorly coded games). Of course, only if their are models with decent prices and NV fixes the drivers.

Otherwise, I'll wait till NV's and AMD's next generation in a couple of years.

 

[bit_user]

If they could reuse the same PCB for the RTX Pro 6000 96GB, it would result in a RTX 5090Ti 64GB with 32x2GB chips.

The pro cards usually have the same PCB as the reference/Founders cards. For partner cards based on that reference design, you will typically see unpopulated pads for DRAM on the backside of the PCB.

One downside of doubling up the DRAM chips is that memory sometimes has to run at a lower frequency. Another is added assembly and parts cost. Whereas if they simply swapped in 24 Gb chips in the RTX 5090 FE card, they don't have to change the cooler, probably the DRAM frequency, and assembly doesn't get any more expensive. Plus, they maintain a comfortable 2x gap vs. the workstation card.

So, that's why I think the highest they'd go is 48 GB.