In a new leak roundup that almost no one expected to see, especially now, we’re learning that AMD is considering significantly lowering the specs of its RDNA 3 series. We looked at recent leaks about both NVIDIA and AMD’s next-gen GPUs just a few hours ago, and how they are reportedly neck-and-neck in terms of computer power and clock speeds, but it now appears to be outdated.

For the past year or two, all leaks and rumours have pointed to one thing: Navi 31, AMD’s rDNA 3 lineup’s top GPU, will have 15,360 Stream Processors. This was the source of today’s leak, which claimed Navi 31 had 92 TFLOPs of FP32 compute capability. Now that we have more recent evidence that contradicts our previous assumptions, things have changed significantly.

AMD has cut the Stream Processor count of at least two of the three RDNA 3 GPUs, according to numerous leakers including Greymon55, Redfire, and 3DCenter.org. Not only that, but it appears that the recent tone-down has had an impact on the number of WorkGroup Processors. According to new sources, the Navi 31 GPU will have just 12,288 Stream Processors instead of 15,360, and only 48 WorkGroup Processors instead of 60.

That’s a 20% reduction in core count, which is significant enough in and of itself, but it also lowers the GPU’s theoretical computer performance, or TFLOP count, which was just revealed. But, as I previously stated, this circumstance involves more than one GPU, which leads us to our second candidate: Navi 32. According to recent information, the Navi 32 will only have 8,192 Stream Processors, not the 10,240 previously speculated.

It’s worth noting that Navi 33, which has reportedly been reduced from 5,192 Stream Processors to merely 4,096, had already been the subject of conflicting rumours about its core count. While most leaks pointed to the former, Greymon55 was adamant about the latter, even though it already had 4096 cores.

https://twitter.com/Redfire75369/status/1521052789996400640?ref_src=twsrc%5Etfw%7Ctwcamp%5Etweetembed%7Ctwterm%5E1521052789996400640%7Ctwgr%5E%7Ctwcon%5Es1_&ref_url=https%3A%2F%2Fappuals.com%2Frdna-3-revised-specs-rumors%2F

this decrease in the core count will result in a decrease in the number of teraflops of computer power available to the GPUs. 3DCenter.org kindly did the math for us, demonstrating how a 20% reduction in core count reduces the prior 92 TFLOPs claim to just 73 TFLOPs, roughly 30 TFLOPs slower than NVIDIA’s next-gen Ada Lovelace flagship GPU.

We don’t have any leaks on the clock speeds of the Navi 32 or Navi 33, so we can’t calculate their computer power right now, but the 3GHz clock speed, as seen in the tweet above, was leaked for the Navi 31, so we can calculate that. Furthermore, this confirms that the target frequencies of these GPUs have not changed, which is a positive indicator.

Navi 31 old → new spec
60 WGP, 15'360 FP32 → 48 WGP, 12'288 FP32
on 3 GHz = 73 TFlops

Navi 32 old → new spec
40 WGP, 10'240 FP32 → 32 WGP, 8'192 FP32

Navi 33 remains the same (16 WGP, 4096 FP32).

Thanks @greymon55 https://t.co/fdsrJT2aew
— 3DCenter.org (@3DCenter_org) May 2, 2022

As is common in these situations, you should always take leaks and assertions with a grain of salt, as they are, after all, rumours. Right now, the GPU market is about to experience an inflow, which means leaks are coming in thick and fast. Just knowing how quickly rumours may shift in a matter of hours should be enough to make you suspicious of these tales.

AMD and NVIDIA are still months away from releasing their respective products—both are aiming for Q3 2022—which means that more accurate leaks will begin to emerge closer to the release date than they are right now. The sources listed above, however, are among the most reliable in the industry, and they kept their word for nearly a year before today’s abrupt change.