Next-Gen flagship GPU rumors: NVIDIA “Ada” GA102 with 21 Gbps memory, AMD’s “RDNA3” Navi 31 with 7 chiplets, Intel “Battlemage” with 20480 cores

Published: Apr 11th 2022, 07:37 GMT   Comments

Please note that this post is tagged as a rumor.

GeForce RTX 4090 vs Radeon RX 7900XT, or something else?

It’s quite intriguing why leakers tend to share their information on the same day. What is important to note that both Kopite7kimi and Gryemon55 clearly have good sources, as their leaks have proven to be true many times.

NVIDIA RTX Ada GA102 GPU gets 21 Gbps memory?

Kopite supposedly has new information on next-gen RTX GPU flagship. Interestingly, it is no longer a given that the upcoming “Ada” GPU series are, in fact, called RTX 40. The leaker is now referring to them by **90/**80 etc, just in case they are called, for instance, the RTX 50 series.

In any case, Kopite7kimi updated his message posted on April’s Fool, as noted at the time, it was not confirmed and probably still isn’t. What this new message says is though that 4080 class GPU with an SKU number of 330 might in fact be used for 4090 series.

More importantly, the memory specs were toned down from 24 Gbps to 21 Gbps, which is actually the same speed as RTX 3090 Ti. This upcoming SKU would have a TDP of 600W. This is likely why some rumors suggested that the just-released RTX 3090 Ti is a test vehicle for next-gen GPU series.

It is worth pointing out that by sharing NVIDIA board numbers PG137, PG139 or PG141, this leak significantly gains in credibility. This is actually something that we can later verify once we get closer to the launch.

AMD Navi 31 with 7 chiplets

It is now rumored that the upcoming Navi 31 GPU, the flagship processor of the RDNA3 and Radeon RX 7000 series, would feature a total of 7 chiplets.

AMD made a return to monolithic designs with Navi 21 GPU, after a bit less successful attempts to compete in high-end GPU space with complex and expensive HBM GPU designs such as Fiji 10 and Vega 10/12.

Neither Greymon nor any other person is sure what the MCD chiplet is for, however, it was long rumored to be a Memory Complex Die. Therefore, it might be where the Infinity Cache or memory controllers are located. It is not entirely confirmed if those MCD chiplets will be located on top of the GDC (Graphics Complex Dies) or located elsewhere on the package.

In total, the Navi 31 GPU is expected to have 7 chiplets manufactured using either 6nm or 5nm TSMC process technology.

INTEL ARC B-Series (Battlemage)

A while ago, thanks to an Intel test driver that the company accidentally published, we learned that the company is working on a next-gen Xe-HPG architecture called Battlemage, internally referred to as Elasti. One of the listed configurations mentioned ‘ELG_X4’, which would imply a 4 tile configuration of the next-gen graphics processor.

At this moment we do not know Battlemage’s specs or how it will it stack up against AMD and NVIDIA, but the company did officially confirm it will target the higher “Enthusiast” tier in GPU market, later to be followed by “Celestial” reaching as high as “Ultra Enthusiast” segment. Battlemage is not launching till 2023/2024, but there is a chance it will become available as RTX 4090 and RX 7900XT are still on the market.

According to Redfire’s analysis of the available data, a potential 4 tile configuration of Battlemage could have a total of 40 Xe-Cores and 320 Xe-Vector Engines per tile. This means that a 4-tile configuration would have 1280 Vector Engines (also known as Execution Units), which would basically mean 20480 FP32 cores. Obviously, we do not know whether “ELG_X4” really refers to the tile configuration, nor if such part is even planned. However, just for speculation sake, this would really put some interesting competition in the high-end space from Intel.

Next-gen Flagship GPU Comparison (RUMORED)
VideoCardz.comGeForce RTX 4090 – ClassRadeon RX 7900 – ClassArc B9XX – Class
Fabrication NodeTSMC N5TSMC N5/N6TBC
ArchitectureNVIDIA AdaAMD RDNA3Intel Xe2-HPG “Battlemage”
GPU PackageMonolithicMulti-Chip-Module (MCM)Multi-Chip-Module (MCM)
Estimated GPU Size~600mm²~800mm²TBC
Graphics Dies12 GCD + 4 MCD + 1 IOD4 Tiles
GPU Mega Clusters12 Graphics Processing Clusters (GPC)2×3 Shader Engines10 Render Slices
GPU Super Clusters72 Texture Processing Clusters (TPC)2×30 RDNA Workgroups (WGP)40 Xe-Cores
GPU Clusters144 Streaming Multiprocessors (SM)1280 Xe Vector Engines
FP32 Cores18432 CUDA Cores15360 Stream Processors20480 FP32 ALUs
GPU Clock~ 2.2 GHz~ 2.5 GHzTBC
Memory & Bus21 Gbps 384-bitTBC Gbps 256-bitTBC
Cache96MB (L2 Cache)256 or 512MB Infinity CacheTBC
Power Consumption600W (TDP)500W (TBP)TBC
Release DateQ3/Q4 2022Q3/Q4 20222023-2024
Rumored Performance2x GA1022.5x NAVI 21TBC

Source:  @kopite7kimi (NVIDIA), @greymon55 (AMD), @Redfire75369 (INTEL)

Comment Policy
  1. Comments must be written in English and should not exceed 1000 characters.
  2. Comments deemed to be spam or solely promotional in nature will be deleted. Including a link to relevant content is permitted, but comments should be relevant to the post topic. Discussions about politics are not allowed on this website.
  3. Comments and usernames containing language or concepts that could be deemed offensive will be deleted.
  4. Comments complaining about the post subject or its source will be removed.
  5. A failure to comply with these rules will result in a warning and, in extreme cases, a ban. In addition, please note that comments that attack or harass an individual directly will result in a ban without warning.
  6. VideoCardz has never been sponsored by AMD, Intel, or NVIDIA. Users claiming otherwise will be banned.
  7. VideoCardz Moderating Team reserves the right to edit or delete any comments submitted to the site without notice.
  8. If you have any questions about the commenting policy, please let us know through the Contact Page.
Hide Comment Policy