Indepth Look a Xbox One's 'GPU'

[Pemalite]

disolitude said:

ethomaz said:

Like the previous article it wrote a lot of errors... the most notable the GPU/eSRAM clock... MS already confirmed 853Mhz so why he wrote 831Mhz? To fit his maths?

"There is obviously the Radeon GPU, something that sources say is somewhere between an HD6000 and HD7000 in capabilities and a fraction behind what Sony has in the PS4."

Why he is downplaying the GPU? It is obvious the GPU is far way better than HD6000... it is close to a HD7790 (Bonaire) with enhancements.

Really doubt its better than 6950 or 6970... but 6870 sound about right.

It wouldn't even beat a 6870 in theoretical performance, maybe on the same level as a Radeon 5770/6770.
Heck, the Playstation 4's GPU has less theoretical performance than the Radeon 3870 X2 released 5 years ago!

However, that's all theoretical.

The problem with prior generations is that VLIW5 (Very Long Instruction Word) and to a lesser extent VLIW4, was that they were heavily reliant on real-time compilers in order to achieve high levels of utilisation of the 5-wide/4-wide (6 and 5 respectively with a branch) SIMD's to achieve optimal throughput, which was fine, most games heavily used the Direct X 9 render path anyway which suited VLIW5 well.

However, skipping along over the years games started to use Direct X 10 and 11 as the PC's started to leave consoles behind in image quality, which then left AMD's VLIW5 architectures underutilised, usually a single unit out of 5 wasn't being used, so, AMD cut back the VLIW5 design to a VLIW4 design but made each unit in the SIMD more flexible.
Of course some units still weren't being used with the VLIW4 design, but not to an extreme degree, hence why the Radeon 6970 despite having less shaders than the 5870 was still faster in compute, it had less shaders but more shader groups that were being more heavily utilised resulting in higher throughput.

GCN found in the consoles however, took it to the next level, it's a compute focused architecture and it shows, it achieves much higher levels of shader utilisation without as many fancy tricks in order to do so.

That's not even worrying about other parts of the GPU like the ROPS, TMU's or the Geometry engines, there is a stupidly large difference in geometry capability between the 6000 series and 7000 series, I do expect Geometry to be a big focus this generation along with partical effects.

That's also why measuring in flops or theoretical performance is pointless unless you are comparing GPU's of the same Architecture, so many other factors comes into play which can tip that balance.