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

Please note that this post is coded as one Rumor.

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

Why leaks share their information on the same day is a very puzzling thing. It should be noted that both Kopite7kimi and Gryemon55 have good sources, and their leaks have repeatedly been proven to be true.

Does the NVIDIA RTX Ada GA102 GPU get 21 Gbps of memory?

Cobit contains new information about the next gen RTX GPU flagship. Interestingly, the upcoming “Ada” GPU series is not really called the RTX 40. The leaker now refers to them as ** 90 / ** 80, if they are called, for example, the RTX 50 series.

However, Kopite7kimi has updated its message posted in April Fool, as noted at the time, it has not been confirmed and probably not yet. The new message is that a 4080 class GPU with an SKU number of 330 can actually be used for the 4090 series.

Most importantly, the memory specifications were reduced from 24 Gbps to 21 Gbps, which is actually the same speed as the RTX 3090 Ti. This upcoming SKU will have a TDP of 600W. This may be the reason why some rumors have suggested that the now released RTX 3090 Ti will be the test vehicle for the next gen GPU series.

It should be noted that by sharing NVIDIA board numbers PG137, PG139 or PG141, this leak gains considerable reliability. This is something we can check later as the release actually approaches.

AMD Navi31 with 7 chiplets

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

After attempts to compete in the high-end GPU space with sophisticated and expensive HBM GPU designs such as the Fiji 10 and Vega 10/12, AMD turned to monolithic designs with the Navi 21 GPU.

It is unknown at this time what he will do after leaving the post. So, Infiniti may be where the cache or memory controllers are. It is not entirely certain whether those MCD chiplets will be located on top of GDC (Graphics Complex Dice) or elsewhere in the package.

In total, the Navi 31 GPU is expected to consist of 7 chipsets manufactured using 6nm or 5nm TSMC process technology.

INTEL ARC B-Series (Formage)

Some time ago, the company accidentally released an Intel test driver, and we learned that the company is working on the next gen Xe-HPG framework, called Battlemage, which is locally referred to as Elasti. One of the configurations listed is referred to as ‘ELG_X4’, which represents the 4 tile configuration of the next gen graphics processor.

At this time we do not know Battlemage’s specifications or how it will be stacked against AMD and NVIDIA, but the company has officially confirmed that it will target the top “Enthusiast” line in the GPU market, and then rose to the “ultra enthusiast” segment reaching “Celestial”. Battlemage was not launched until 2023/2024, but it is likely to be available as the RTX 4090 and RX 7900XT are still on the market.

According to Redfire’s available data analysis, Battlemage’s potential 4 tile configuration may have a total of 40 Xe-cores and 320 Xe-vector engines per tile. This means that the 4-tile configuration will have 1280 vector engines (also known as execution units), which basically represents 20480 FP32 cores. Obviously, we do not know if “ELG_X4” actually refers to the tile configuration or if such an area is planned. However, for speculation, this could actually lead to some interesting competition in the top spot from Intel.

Next Generation Top GPU Comparison (Rumor) GeForce RTX 4090 – Class Radeon RX 7900 – Class Arc B9XX – Class
Fabrication tip TSMC N5 TSMC N5 / N6 TBC
Architecture Nvidia is available AMD RDNA3 Intel Xe2-HPG “Formage”
GPU package Monolith Multi-chip-module (MCM) Multi-chip-module (MCM)
Estimated GPU size ~ 600 mm 800 mm TBC
The graphics are dying 1 2 GCD + 4 MCD + 1 IOD 4 tiles
GPU mega clusters 12 Graphics Processing Clusters (GPC) 2 × 3 shader engines 10 render slices
GPU Super Clusters 72 Texture Processing Clusters (TPC) 2 × 30 RDNA Task Forces (WGP) 40 Xe-cores
GPU clusters 144 Streaming Multiprocessors (SM) 1280 Xe Vector Engines
FP32 cores 18432 CUDA colors 15360 Stream Processors 20480 FP32 ALUs
GPU clock 2.2 GHz 2.5 GHz TBC
Memory type 24GB GDDR6X TBC GB GDTR6 TBC
Memory & Bus 21 Gbps 384-bit TBC GPS 256-bit TBC
Temporary storage 96MB (L2 cache) 256 or 512MB Infinity cache TBC
Power consumption 600W (TDP) 500W (TBP) TBC
Release date Q3 / Q4 2022 Q3 / Q4 2022 2023-2024
Rumored performance 2x GA102 2.5x NAVI 21 TBC

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

Leave a Comment