⋮  

Rumor: AMD’s MCM Based RDNA3 RX 7000 GPUs Will Be More Power Efficient Than NVIDIA RTX 4000 GPUs And Could Even Offer Higher Performance

Submit

An analysis by Moores Law is Dead (via HardwareTimes) suggests that AMD's next-generation RTX 7000 series (which have been confirmed to be based on the MCM approach) could out-perform NVIDIA's RTX 4000 series GPUs. It goes without saying that this is a huge rumor and should be taken with buckets of salt till further verification is received. That said, this is an inference by MLiD based on information provided by his sources (and corroborated by at least one independent leak given below) and if those are considered to be credible enough - then this inference could very well turn out to be true as well.

Rumor: AMD's RDNA3 based MCM-GPUs will blow away NVIDIA RTX 4000 series GPUs in power efficiency and could even take the performance crown

A quick recap for our readers before we dig into this. A leaked LinkedIn description of an AMD Principal Member of Technical Staff more or less confirmed that the upcoming RX 7000 series is based on RDNA3 and has at least 1 SKU with two different processes involved (TSMC 5nm and 6nm). The only reason this would be the case is, of course, if AMD was going with an MCM approach as far as its GPUs go.

AMD Engineer confirms 5nm and 6nm process nodes for next-gen Radeon & Instinct GPUs. (Image Credits: LinkedIn)

AMD has already turned the tides in the CPU market thanks to its MCM-based Ryzen processors and it looks like it wants to repeat that trick on the GPU side as well. While this is where the confirmed news ends, another rumor by Twitter user Olrak suggests that the RDNA3 GPUs could have as many as 15,360 cores  - which would be quite the step up from the 5120 cores of the RX 6900.

This is a frugal, smart lineup from AMD that should just completely blow Lovelace out of the water when it comes to efficiency at the very least and while doing that, maybe even take top performance. And this is not like RDNA2 vs Ampere where generally speaking AMD was more efficient most of the time. This time I think NVIDIA has got a real, true efficiency problem coming up here.

.... Lovelace's performance, dont get me wrong again, Lovelace is going to be a huge performance increase, bigger than Ampere over Turing, but at the same time, the initial targets guys, and I can't stop saying this, was 66% to about 80% increase over Ampere.

-Moore's Law is Dead

We are pretty early in the leak cycle for the next-generation GPUs from both AMD and NVIDIA, but based on the data we have - MLiD's statement does not seem that far-fetched. It hinges on the fact that AMD is working on an MCM approach to GPUs. If this part truly comes to pass (that LinkedIn description could have been of a planned roadmap - and as we know, roadmaps change all the time) and AMD goes with an MCM approach, it should very easily be able to eat NVIDIA's lunch in terms of power efficiency. Whether it can beat NVIDIA in terms of absolute performance is another matter altogether.

Ever since Lisa took over, AMD has focused more on carving out a niche for itself in the GPU side of things that is focused on the mainstream segment rather than the high-end gaming segment - but all this could change with RDNA3. With some key limitations of the RDNA2 architecture gone, a node shrink, and the MCM design philosophy finally unleashed with GPUs - it could be the first check from AMD to NVIDIA.

Copy of AMD RDNA3 / RX 7000 Series GPU Rumored SKUs [Unconfirmed]

Radeon LineupRadeon RX 5000Radeon RX 6000Radeon RX 7000
GPU ArchitectureRDNA 1RDNA 2RDNA 3 / RDNA 2
Process Node7nm7nm5nm/6nm?
GPU FamilyNavi 1XNavi 2XNavi 3X
Flagship GPUN/ANavi 21 (5120 SPs)Navi 31 (15360 SPs)
High-End GPUNavi 10 (2560 SPs)Navi 22 (2560 SPs)Navi 32 (10240 SPs)
Mid-Tier GPUNavi 12 (2560 SPs)Navi 23 (2048 SPs)Navi 33 (5120 SPs)
Entry-Tier GPUNavi 14 (1536 SPs)Navi 24 (1024 SPs)Navi 34 (2560 SPs)
Submit