Leaked: Final specs for Nvidia’s next-gen RTX 5090, 5080, and 5070 GPUs
Last month, the leaker known as kopite7kimi — known for his accurate predictions — revealed on X (formerly Twitter) the likely final specifications for Nvidia’s next-gen GeForce RTX 5080 and 5090 GPUs. More recently, Wccftech claims to have picked up details on the RTX 5070.
The launch for the new 5000 series has also been scheduled, with Nvidia tipped to unveil the graphics cards in just a few months at CES in January 2025. The first GeForce RTX 5080 and 5090 graphics cards are expected to be available for purchase in January, while the RTX 5070 is expected to launch in February (and the RTX 5060 in March).
The most important change compared to what we previously knew about the new top-of-the-line RTX 5090 model is an upgraded memory interface for a whopping 32GB of video memory, which is now 512 bits wide (instead of 448 bits). Nvidia has also increased the number of shader units to an incredible 21,760.
In fact, Nvidia can equip the GB202-300-A1 graphics processor with 24,576 shader units in full expansion, but the full computing power of the Blackwell chip is probably reserved for professional cards (which Nvidia can sell at a significantly higher price point).
However, it’s also possible that Nvidia will add an RTX 5090 Ti or Titan towards the end of the 5000-series product cycle — if the competitive situation demands it, which isn’t too likely for PC graphics cards.
RTX 4090 vs. RTX 5090, compared
| GeForce RTX 4090 | GeForce RTX 5090 (according to kopite7kimi’s leak) | |
| GPU | AD102-300-A1 / AD102-301-A1 | GB202-300-A1 |
| Codename | Ada Lovelace | David Blackwell |
| Shader units | 16,384 | 21,760 |
| Streaming processors | 128 | 170 |
| Memory bus | 448 bits | 512 bits |
| Memory type | GDDR6 | GDDR7 |
| Graphics memory | 24GB | 32GB |
| TGP | 450W | 600W |
| MSRP | $1,599 | ? |
Along with the enormous increase in computing power, the Total Graphics Power (TGP) — i.e., the maximum graphics card power consumption at maximum utilization — has unfortunately also increased from 450W to a crazy 600W in the new RTX 5090.
The technical specs of the Nvidia RTX 5090 promise a massive increase in performance compared to the already-very-fast RTX 4090, which is why we assume that the RTX 5090 will be an expensive graphics card.
The RTX 4090 FE launched with an MSRP of $1,599 but is pretty much impossible to find at that price anymore. You’re looking to spend upwards of $2,000 or more for an RTX 4090 these days, so there’s no reason to expect the faster and better-equipped RTX 5090 to debut with an MSRP below $2,000.
Since Nvidia has basically monopolized the graphics card market with a market share of 88 percent and is technically far ahead of their last remaining competitor in AMD, Nvidia can virtually price the 5090 however they want. Prepare yourself for a hefty price increase, given that it’s going to be the “fastest graphics card in the world.”
RTX 4080 vs. RTX 5080, compared
| GeForce RTX 4080 | GeForce RTX 5080 (according to kopite7kimi’s leak) | |
| GPU | AD103-300-A1 / AD103-301-A1 | GB203-400-A1 |
| Codename | Ada Lovelace | David Blackwell |
| Shader units | 9,728 | 10,752 |
| Streaming processors | 76 | 84 |
| Memory bus | 256 bits | 256 bits |
| Memory type | GDDR6 | GDDR7 |
| Graphics memory | 16GB | 16GB |
| TGP | 320W | 400W |
| MSRP | $1,199 | ? |
kopite7kimi also announced the expected technical specs of the RTX 5080, which comes with a smaller leap in performance over its predecessor. The number of streaming processors is going up by around 10 percent from 76 to 84, and the memory connection is remaining the same (although the 5080 is getting the faster GDDR7 video memory).
That said, it’s difficult to estimate how powerful the RTX 5080 will really be. In purely technical terms, it has 10 percent more shader units, but the new graphics chip architecture should also be more efficient and the faster video memory should provide extra performance. All in all, we expect an overall performance bump of between 15 and 25 percent.
Compared to its predecessor, the power consumption of the RTX 5080 is only being bumped up by a moderate 50 watts. As for the expected price of RTX 5080 cards? That’s still up in the air.
What’s most noteworthy here is the chasm between the RTX 5080 and 5090, which offers almost twice as much computing power. Nvidia has never had such a big difference between an 80 and 90-series chip before, which is reason to treat the leaker’s figures with caution.
RTX 4070 vs. RTX 5070, compared
| GeForce RTX 4070 | GeForce RTX 5070 (according to Wccftech) | |
| GPU | AD104-250-A1 | GB205 |
| Codename | Ada Lovelace | David Blackwell |
| Shader units | 5,888 | TBD (6,400?) |
| Streaming processors | 46 | TBD (50?) |
| Memory bus | 192 bits | 192 bits |
| Memory type | GDDR6 | GDDR7 |
| Graphics memory | 12GB | 12GB |
| TGP | 200W | 250W |
| MSRP | $599 | ? |
As expected with the RTX 5070, Nvidia is stingier with video memory, which — like the RTX 4070 — sits at 12GB and is connected with a 192-bit narrow interface. The main difference is a switch from GDDR6 to GDDR7, which increases the memory bandwidth by around a third (up to 672GB/s). The TGP will also increase from 200 to 250 watts.
The exact chip configuration of the RTX 5070 hasn’t been leaked yet, but a total of up to 50 streaming processors with up to 6,400 shader units should be available on the GB205 chip.
As neither the number of shader units nor the GPU clock rate is known, it isn’t possible to estimate the performance of the RTX 5070 yet. However, compared to an RTX 4070 with 5,888 shader units, we don’t expect the performance increase to be too great.
Further reading: The best graphics cards for every budget