News AMD Integrated Radeon 780M 25% Faster Than RDNA 2 Predecessor

[Admin]

AMD’s iGPUs continue to make progress with the first RDNA 3 architecture models delivering about a 25% generational performance uplift in 3DMark Time Spy tests.

AMD Integrated Radeon 780M 25% Faster Than RDNA 2 Predecessor : Read more

 

[jackt]

News about the smaller desktop gpus ?

 

[w_barath]

I'm a bit disappointed with the 7040 RDNA3 APU performance. 25% performance increase could easily be attributed to the DDR5 memory interface alone. Where's a sing of the RDNA 3 architectural improvements? Ah, this is AMD. The real performance is hiding behind a 1-1.5 year driver development lag. It'll probably be 30% faster than now same time next year.

 

[prtskg]

I'm a bit disappointed with the 7040 RDNA3 APU performance. 25% performance increase could easily be attributed to the DDR5 memory interface alone. Where's a sing of the RDNA 3 architectural improvements? Ah, this is AMD. The real performance is hiding behind a 1-1.5 year driver development lag. It'll probably be 30% faster than now same time next year.

Ryzen 6000 series apus use DDR5. Of course, in last one year, I expect improvement in ram speed but your comment gives me idea that you were thinking about DDR4.

 

[watzupken]

Ryzen 6000 series apus use DDR5. Of course, in last one year, I expect improvement in ram speed but your comment gives me idea that you were thinking about DDR4.

The current 680M generally runs on laptop, and DDR5 SODIMMs runs at 4800. Compared to this where the DDR5 is running at 5600, the increase bandwidth will naturally benefit performance. I suspect there is also improvement in the hardware to get a 25% performance bump, i.e. in the form of high CU count. The increase in memory bandwidth itself is not likely to contribute to this big a performance improvement.

 

[Karl_Schwarzschild]

I'm really confused here. 780m is supposed to have 12 CUs. In RDNA3 there are 128 shaders/CU (or 64 shaders with 2 FP Units) which translates into 1536 Shaders vs 768 for the 680m (only 64FPs/CU in RDNA2).
+Higher Clocks +DDR5
Shouldn't that be more like +125-150%?
The 7900XTX with 96CUs also behaves like a 12288 Shader GPU, although it has 6144 (double) Shaders on paper.
Early leaks were also hinting 3600M like performance for the 780M. 2050Mish is pretty disappointing now.
And that's not only me. Here they also list it as 8,9 FP32 TFLOPs vs 3,4 for the 680M.