Nvidia RTX 5070 could launch at CES 2025, and rumored spec – including 12GB of VRAM – means everyone hates this GPU already

October 10

Nvidia’s RTX 5070 is apparently going to be revealed at CES 2025, alongside the rumored RTX 5080 and 5090 next-gen graphics cards.

This is a notable change in the way the rumor winds are blowing, as up until now, we’ve been hearing that it’s just the RTX 5080 and 5090 that’ll be the initial Blackwell GPUs to pitch up from Nvidia.

Wccftech has made this claim, and also provides some spec revelations for the RTX 5070, after speaking to sources (who yesterday delivered a leak on the RTX 5080 getting speedier VRAM). Furthermore, two leakers on X have chimed in, not disagreeing with the tech site, and providing their own take on the spec of the RTX 5070.

Wccftech believes that the RTX 5070 will use the GB205 chip, but doesn’t give us a core count. We’re also told that the graphics card will run with a 192-bit memory bus and 12GB of VRAM (GDDR7). This will be the slower flavor of GDDR7, 28Gbps, which will also be on board the RTX 5090 (but the 5080 will get the 32Gbps memory, if Wccftech’s sources are correct, that is).

For the RTX 5070, then, we’re looking at a total memory bandwidth of 672GB/s, quite a bit less than the RTX 5080’s supposed 1024GB/s, but a good chunk more than the RTX 4070 (a third more, in fact).

However, we’re told power usage will also creep up for the RTX 5070, and that it’ll chug 250W (compared to 200W for the RTX 4070).

Leaker Harukaze5719 backs up these specs entirely, while adding a purported core count, namely 6,400. Another regular leaker on X, Kopite7kimi, backs up the 250W TDP theory, too (and the 12GB of VRAM assertion, for that matter).

GB202: 24,576 RTX 5090: 21,760 / GB202-300-A1 / 512-bit / 32GB GDDR7, 28Gbps, ~1.8TB/s / 600W TBP / 14-Layer FE PCBGB203: 10,752 RTX 5080: 10,752 / GB203-400-A1 / 256-bit / 16GB GDDR7, 32Gbps, 1TB/s, 400W TBPGB205: 6,400 RTX 5070: TBD / GB205 Based / 192-bit / 12GB GDDR7,… https://t.co/KTO6QLdKCkOctober 9, 2024

As to the arrival timeframe, remember that this is referring to the initial unveiling of these three Blackwell graphics cards together at CES 2025 (in theory). When it comes to the on-sale dates, the RTX 5090 and 5080 are supposedly going to be close together, but we’d imagine the RTX 5070 is likely to be further away given its very sudden appearance in the Blackwell launch rumors.

data:image/gif;base64,R0lGODlhAQABAIAAAAAAAP///ywAAAAAAQABAAACAUwAOw==

(Image credit: Future / John Loeffler)

Analysis: Nvidia’s RTX 5070 – a sheep in wolf’s clothing GPU?

Okay, deep breaths. As you can imagine, there’s been some displeasure at these RTX 5070 specs, and notably with the memory (and the drop to the GB205 chip – the RTX 5080 will use GB203).

You can guess the kind of stuff: ‘Nvidia is scamming PC gamers again!’ ‘The RTX 5070 should have 16GB of VRAM, the RTX 5060 should have 12GB – this is an RTX 5060 being sold as a 5070!’ And so on…

Well, these people have a point, certainly – but we can’t get ahead of ourselves here. We’re still in rumor-land, firstly, and also, we don’t know the pricing attached to these GPUs. Whatever the respective specs, Nvidia will hopefully ensure that these graphics cards make sense in terms of performance per dollar, and overall value.

Although we guess the trouble is that most folks find it hard to believe Nvidia will strike a reasonable value proposition, and that includes us. This would mean a fundamental change in attitude for Team Green, and there would have to be a reason for that. The only possibility is RDNA 4’s performance and value balance – but remember, AMD will top out at the mid-range. Meaning it might have some influence on the RTX 5070, perhaps, but for the higher-end Nvidia Blackwell cards, there won’t be any rivals, not remotely.

So, in short, we fully understand the worries here, but let’s not condemn these graphics cards based on rumors before they even arrive, however tempting that might be.

Still, if it is true that the RTX 5070 will have 12GB, and RTX 5080 16GB, you’ve got to ask yourself: why? This really doesn’t feel like enough VRAM at this point in time for PC gamers, let alone a remotely future-proofed loadout as the years roll on (and we’ll have a good while until RTX 6000, of course).

The cynical will, of course, suggest that these video RAM configurations give Nvidia room to maneuver down the line with ‘Super’ or ‘Ti’ versions. There is rumored to be a 24GB variant of the RTX 5080 coming later, which might be the RTX 5080 Super, and this allows Nvidia to charge more for what most gamers would see as the correct amount of VRAM for this level of graphics card in the current gaming climate. Perhaps it’ll be ditto for a 5070 packing 16GB of video memory, all of which feels like cynical tactics to maximize sales and profits.

This raises another specter – when we do get the RTX 5060 arriving, presumably that’s going to have 8GB, then? As if there weren’t enough complaints about sticking 8GB on the current RTX 4060. Yeah, we’re kind of getting sucked into the negative spiral of worries here, but we must keep reminding ourselves – these are just rumors. And Nvidia must ensure these graphics cards make some kind of sense, in terms of their overall performance and price tag.

However, it does feel like Nvidia is pushing gamers who play at 4K to buy their priciest GPUs, and you can bet that those top-end Blackwell offerings are going to be seriously expensive.

On a final, more positive note, Wccftech talks about Nvidia revealing “next-generation AI technologies” for the PC gamer which will be a big surprise at CES 2025. Color us intrigued. A way for DLSS 4 to make up for lean VRAM loadouts, perhaps?

Via VideoCardz