Renaissance GPUs could offer up to 128 cores per compute unit for AMD RDNA 5

Further details on AMD RDNA 5 GPUs suggest 128 cores per compute unit and a total of over 12K cores, possibly bringing the largest leap yet for Radeon gaming.

Well, if you think that AMD’s RDNA 4 is a strong improvement, hold onto your hat, AMD RDNA 5 GPUs have even bigger plans on the way. New leak claims the next-gen Radeon graphics cards can offer 128 cores per compute unit (CU) and scale all the way up to 12,288 cores on the flagship die. That is not merely a generational incremental bump; it is perhaps the kind of generational leap that could catapult AMD back into ‘serious competition’ territory with NVIDIA at the high end ultra compute mega GPU segment.

Core Counts for AMD RDNA 5 GPUs Exposed

New reports suggest that AMD is only prepping four different GPU dies under the RDNA 5 GPUs, from entry-level to flagship designs. Well, if the leaks are true, this is what we are in for:

  • Flagship Die → 12,288 cores | 96 CUs
  • Mid Tier Die: 40 CUs / 5,120 cores
  • we have Low-End Die: 24 CUs / 3,072 cores.
  • Base-Level Die: 12 CUs / 1536 cores

At 128 cores per CU, that number will be double that of the current RDNA 4 GPUs, which feature 64 cores per CU. Basically, the RDNA 5 flagship could be three times higher core than AMD’s current flagship Navi 48 GPU.

What Makes RDNA 5 GPUs Different From RDNA 4

Taking a different strategy, AMD’s RDNA 4 architecture makes use of two monolithic dies, providing greater efficiency and more cost-effective than the chiplet design of RDNA 3. However, it limited scalability. As for RDNA 5 GPUs, AMD appears to be following from RDNA 2, which hit the market back in late-2020 also at the same time with multiple dies. That wider spread may give gamers a lot more bang for their buck on various performance tiers.

Competitors Outside of NVIDIA, & Intel too

During the RDNA 2 generation, with the RX 6900 XT, AMD enjoyed a lot of success at the top end against AMD, particularly versus NVIDIA’s RTX 3090. NVIDIA has owned the ultra-enthusiast space for some years now, while AMD aimed its RDNA 3 and RDNA 4 GPUs in the RTX 80-class — not 90 — space.

However, AMD RDNA 5 GPUs could deliver more than 12K cores, improved ray tracing (RT) and AI improvements, so perhaps AMD is ready to take on NVIDIA’s top products once again. Assuming the performance targets are met, RDNA 5 might even find itself closing the gap against the next-generation NVIDIA Blackwell SUPER GPUs and Intel’s Arc Battlemage G31 series as well.

How Important AMD RDNA 5 GPUs Could End Up Being

That translates to better parallel performance for gamers with higher frame rates at higher levels in more demanding titles while also being able to tackle advanced workloads like ray tracing and AI capability more efficiently. Well executed RDNA 5 GPUs may give Radeon back its place at ultra-enthusiast-market competitiveness, hopefully giving buyers a true NVIDIA alternative.

Expected Memory & Specs

The specs also tease plenty of memory for AMD RDNA 5 GPUs:

  • Flagship: 24–32 GiB VRAM, up to 512-bit bus
  • Mid-Range: 12–24GB VRAM, 384-192 bit bus
  • Bottom range: 8–16 GB VRAM, 256–128 bit bus
  • Entry-Level: 8–16 GB VRAM, 128–64–bit bus

This indicates AMD is not only shooting for higher core counts but bandwidth and VRAM capability as well for the next wave of gaming demands at higher resolutions.

FAQ

How many cores will it have the AMD RDNA 5 flagship gpu?

A high-end AMD RDNA 5 GPU is said to pack 12,288 cores across 96 CU.

What is the difference between RDNA 5 and RDNA 4?

RDNA 4 GPUs had 64 cores per CU, but RDNA 5 GPUs may sport 128 cores per CU, which would be effectively double the core density.

Will AMD RDNA 5 GPUs challenge NVIDIA?

If AMD has the actual performance to match the leaks, RDNA 5 GPUs could well continue to close the gap to NVIDIA RTX Blackwell releases, returning AMD to high-end GPU territory.

Join Telegram

Join Now

2 thoughts on “Renaissance GPUs could offer up to 128 cores per compute unit for AMD RDNA 5”

Leave a Comment