Monday, February 21st 2022
AMD Radeon RDNA2 680M iGPU Beats NVIDIA MX450 Discrete GPU
The recently announced AMD Ryzen 6000 series mobile Zen 3+ processors feature a significant graphics improvement with up to 12 RDNA2 Compute Units available. These new graphics solutions have recently been tested and compared by an engineer working for Lenovo in China. The Radeon 680M and Radeon 660M feature 12 and 6 RDNA2 Compute Units respectively and have been tested against the NVIDIA MX450 and the Intel Iris Xe Graphics G7. The Radeon 680M represents an 85% performance improvement over the Radeon RX Vega 8 and is 24% faster than the discrete NVIDIA MX450 mobile GPU in 3DMark. This lead narrows in real-world tests where the 680M is only 1.1% faster than the MX450 and the 660M is 37% slower.
The mid-range Radeon 660M is still significantly faster than the Intel Iris Xe Graphics G7 (96EU) found in the i7-12700H beating it by 9% in 1080P gaming. The review also looks at power efficiency for the Radeon 660M & 680M showing that in their highest power configurations performance increases by 10% for the 660M and 42% for the 680M. The Radeon 680M remains behind the NVIDIA GTX 1650 Max-Q which holds a 25% lead. The Ryzen 6000 mobile series will be available in laptops starting from next month.
Source:
Zhihu (via VideoCardz)
The mid-range Radeon 660M is still significantly faster than the Intel Iris Xe Graphics G7 (96EU) found in the i7-12700H beating it by 9% in 1080P gaming. The review also looks at power efficiency for the Radeon 660M & 680M showing that in their highest power configurations performance increases by 10% for the 660M and 42% for the 680M. The Radeon 680M remains behind the NVIDIA GTX 1650 Max-Q which holds a 25% lead. The Ryzen 6000 mobile series will be available in laptops starting from next month.
18 Comments on AMD Radeon RDNA2 680M iGPU Beats NVIDIA MX450 Discrete GPU
Though personally I want AMD a create separate product line for CPUs with powerful iGPU, unrestricted by power and thermal constraints like mobile ones. Basically $200 budget graphics card replacement. Dreaming that for years. Even one SKU would be fine in my books. Just think about it. Tech sites will list a CPU as best budget GPU. Fucken hilarious.
memory latency isn't good on lpddr but bandwidth is better so should give some good uplift, tdp down to 35 w "tdp" so 45 boost would be my dream laptop :)
AMD could make a 256bit LPDDR5 chip with 24+ compute units that would be better than most low end GPU and could even compete at the mid range ala Apple M1 Pro/MAX. it wouldn't be supported on AM4/5 but on laptop, that do not really matter.
But i suspect that neither of them want that, they prefer to sell dedicated GPU on the side and keep iGPU as low end.
Integrated only works because there is zero additional system cost to use it (128-bit bus max); as for why Zen 4 is including it, I'd say it's because they finally have the space (and what else are you going to use all that DDR5 memory bandwidth for? Alder Lake is wasting that bandwidth on the same 32 eus as Rocket Lake!)
The things is Apple have no interest into making a dGPU. So it make sense for them to build huge APU with a heavy iGPU. it do not compete with their own product.
Since most of the time they re-use the same die as desktop for mobile GPU, they would have to spend more to develop a new chip and it would compete for other product for fab capacity. It's the main reason. AMD prefer to re-use as much as possible their own chip and it do not make sense for them to make a new one just for that specific case.
As for bandwidth, getting into mid range GPU market, the chip would need more bandwidth than 128 bit DDR could offer.
By the time you spends all that money on memory bandwidth, you cold have simply included the discrete card!Thanks to ARM superiority, Apple now lives in it's own world; but that doesn't mean they suddeny licked the big iGPU scaling problem:
wccftech.com/m1-max-slower-than-laptop-rtx-3080-in-gaming-benchmarks/
As for adding a discrete GPU, it would still cost less to add another 128 bit bus on top of the actual one than add another chip + other coldplate/heat pipes + another set for VRM + another memory bus.
So it would still be cheaper. And since you wouldn't have to mess up with the dGPU <-> iGPU switching or routing, it would be a simpler and more stable product.
As for ARM superiority, this is not a thing. It's more a mix of Apple design superiority with a closed source OS that they manage and better process (5nm).