RTX 5090/5080: New leak claims to have all the details on the upcoming Nvidia GPUs – be prepared for high electricity costs and disappointment

Price and release date remain unknown.

The launch of the Nvidia Geforce RTX 5090 is probably getting closer: a new leak reveals possible specifications for the GPU flagship for the first time. There are also new details about the RTX 5080 that do not bode well.

Why this is important:Previous rumors have mentioned exact numbers for memory or power consumption, but for the GB202 and GB203 chip, only the approximate full configuration could be determined.

The new leaks now show in which format the Blackwell chips for RTX 5090 and 5080 could be installed, thus providing a much more accurate insight into performance – under the usual It’s just rumors-reservation.

In detail: The well-known and reliable leaker “kopite7kimi” has published a new post on X/Twitter listing the exact specs of the RTX 5090 and RTX 5080.

• Accordingly, the RTX 5090 with the GB202-300 chip should have over 21,760 CUDA cores
• The VRAM is called 32 GB GDDR7 memory with a 512-bit connection
• The power consumption is estimated at a whopping 600 watts Nevertheless, Nvidia is said to be using a 2-slot cooler for the RTX 5090, which, given the high TDP, indicates a revised cooling system.
  

  GeForce RTX 5090
  PG144/145-SKU30
  GB202-300-A1
  21760FP32
  512-bit GDDR7 32G
  600W

  — kopite7kimi (@kopite7kimi) September 26, 2024

  For the RTX 5080, kopite7kimi cites the following data sheet:

  • The GB203 chip is limited to 10,752 CUDA cores The TDP should be around 400 watts
  • On the memory side, 16 GByte GDDR7 memory with 256-bit interface are mentioned.

  GeForce RTX 5080
  PG144/147-SKU45
  GB203-400-A1
  10752FP32
  256-bit GDDR7 16G
  400W

  — kopite7kimi (@kopite7kimi) September 26, 2024

  The latter point in particular is causing discussion: some people simply expected more memory from the “second-best” RTX 5000 GPU.

  • In some ways, this brings back memories of the launch of the original RTX 4080, which only had 12 GB of VRAM.
  • After facing significant headwinds, Nvidia withdrew this GPU release to re-market it as the RTX 4070 Ti at a later date. You can read more about this in the article linked below.
  • At least according to the leak, RTX 4080 Super owners see no reason to switch to the successor ; the theoretical performance increase with the same amount of memory is too small.

  However, an important aspect is missing in both posts: prices were not mentioned for either the RTX 5090 or the RTX 5080.

  At least the data sheet with the most important ones compared to its direct predecessors is now emerging:

  	RTX 5090	RTX 4090	RTX 5080	RTX 4080
  Chip	GB202-300-A1	AD102-300-A1	GB203-400-A1	AD103-300-A1
  CUDA cores	21,760	16,384	10,752	9,728
  Memory size/type	32 GB GDDR7	24 GB GDDR6X	16 GB GDDR7	16 GByte GDDR6X
  memory interface	512 Bit	384 Bit	256 Bit	256 Bit
  Bandwidth*	At 28 Gbps: 1,792 GByte/s At 32 Gbps: 2,048 GB/s	1,008 GB/s	At 28 Gbps: 896 GB/s At 32 Gbps: 1,024 GB/s	717 GB/s
  Power consumption	600 watts	450 Watt	400 Watt	320 Watt
  MSRP at release	–	1.949 Euro	–	1,469 euros

  A brief note on bandwidth: the speed of the memory modules is still unknown. The only thing that is certain is that they will be manufactured by Samsung, which has presented the GDDR7 memory in two variants.

  In addition to the lack of price information, a release period is also still unclear. However, based on the exact details provided, it could be expected that official information will indeed follow before the end of this year.

  Now we want to know your opinion: What do you think of the leaked specs of the RTX 5090 and RTX 5080? At what price point would you consider buying one? Let us know in the comments!