Fastest RDNA3-Mobile Based ‘RX 7900M’ GPU Rumoured to Be On Par With The RTX 3090
AMD will unveil the RDNA3 architecture on the 3rd of November. We have not seen a lot of performance leaks from team red this time. RDNA3 brings many architectural changes, tipped to offer a 50% efficiency increase over its predecessor. The next-generation of Infinity Cache along with a completely rearchitected Compute Unit shows that RDNA3 promises massive gains. Coming back to the topic, various leakers have showcased the rumoured performance levels of AMD’s highest-end RDNA3 mobile GPU.
A 70% Performance Uplift
Yesterday, Greymon in a vague manner highlighted RDNA3’s performance. The ‘laptop emoji’ signals towards a mobile GPU. This combined with the fact that the comparison has been made with an AMD GPU, confirms that Greymon is hinting at RDNA3’s performance in the mobile segment. The ‘unnamed’ RDNA3 mobile GPU is on-par with the 6950XT.
This in itself is a massive leap from last-generation. The Radeon RX 6850M XT is the highest end GPU we saw from AMD’s RX 6000 Mobile lineup. This SKU landed in the ballpark of an RX 6700XT / RTX 3060 Ti. The 6950XT is 70% faster than the 6700XT (courtesy of techpowerup). This means that AMD’s RDNA3 (Mobile) is at least 70% faster than RDNA2 (Mobile).
However, that’s not the complete leak. Kopite, while replying to the RTX 4000 mobile lineup leak from Harukaze stated that the RX7900M is equal to the RTX 3090. To recap, the RTX 3090 is almost equal to the 6950XT. So the un-named RDNA3 mobile GPU above is indeed the RX7900M. This GPU is possibly the highest offering (Mobile-GPU) from AMD this time around.
😮💨Battle with RX 7900M, it has performance like RTX 3090.
— kopite7kimi (@kopite7kimi) October 22, 2022
The next-gen offerings from AMD as stated above, do not have many ‘leaked’ benchmarks. Although, speculation wise, these GPUs can deliver a 70% uplift in performance at least based on today’s leak. Do bear in mind, Navi31 has 2.4x more Compute Units as compared to Navi21.
This is a very rough estimate, however. RDNA moves beyond the typical CU architecture and integrates the WorkGroup Processor for computing. Basically, we are using GCN (Pre-RDNA1) metrics to calculate RDNA3’s performance, which is prone to errors. The 2.4x number, for performance estimates at least, should be taken with a grain of salt.
NVIDIA’s AD102 (4090) is equipped with 50% more cores than the Ampere based GA102. That alone nets it a 60-70% uplift in raster performance. If AMD’s additional Shading Units (12288 vs 5120) can scale just like NVIDIA, expect a similar, if not a larger performance uplift. However, for many people the main concern is pricing. Provided that AMD can offer a respectable gen-on-gen performance increase with decent pricing, they can take a significant cut of the market share from NVIDIA.
RDNA3 Desktop (Leaked) Lineup
|SKU||Chip||CU (128 Shading Units/ 1 CU)||CU (64 Shading Units/ 1 CU)||FP32||Max Clock||Cache||Memory Bus||VRAM||Memroy Spec||Speed (Gbps)||TDP|
|RX 7970XT3D||Navi31||96||192||12288||3.2 GHz||192MB?||384-bit||24GB||GDDR6||24||450W|
|RX 7950XT||Navi31||84||168||10752||3.1 GHz||96MB||384-bit||24GB||GDDR6||24||420W|
|RX 7900XT||Navi31||70||140||8960||3.0 GHz||80MB||320-bit||20GB||GDDR6||20||340W|
|RX 7800XT||Navi32||60||120||7680||3.4 GHz||64MB||256-bit||16GB||GDDR6||18||330W|
|RX 7800||Navi32||56||112||7168||3.3 GHz||48MB?||256-bit||16GB||GDDR6||18||250W|
|RX 7700XT||Navi32||48||96||6144||3.2 GHz||48MB?||192-bit||12GB||GDDR6||18||200W|
|RX 7600XT||Navi33||32||64||4096||3.2 GHz||32MB||128-bit||8GB||GDDR6||18||150W|
|RX 7600||Navi33||28||56||3584||3.1 GHZ?||32MB?||128-bit||8GB||GDDR6||18||130W|