Digital Foundry: "PS4 GPU 50% more powerful than Xbox One"

In terms of the GPU hardware, hard information was difficult to come by, but one of the engineers did let slip with a significant stat - 768 operations per clock. We know that both Xbox One and PlayStation 4 are based on Radeon GCN architecture and we also know that each compute unit is capable of 64 operations per clock. So, again through a process of extrapolation from the drip-feed of hard facts, the make-up of the One's GPU is confirmed - 12 compute units each capable of 64 ops/clock gives us the 768 total revealed by Microsoft and thus, by extension, the 1.2 teraflop graphics core. So that's another tick on the Durango leaked spec that has been transposed across to the final Xbox One architecture and the proof we need that PlayStation 4's 18 CU graphics core has 50 per cent more raw power than the GPU in the new Microsoft console. Now, bearing in mind that we fully expect PlayStation 4 and Xbox One to launch at similar price-points, how did this disparity come about?

The answer to that comes down to a specific gamble Sony made that Microsoft could not - the utilisation of a unified pool of GDDR5 memory. In the early days of PS4 development, only 2GB of this type of memory looked viable for a consumer-level device. As higher density modules became available, this was duly upgraded to 4GB. By the time of the reveal back in February, Sony had confidence that it could secure volume of 512MB modules and surprised everyone (even developers) by announcing that PS4 would ship with 8GB of unified GDDR5 RAM. The design of its surrounding architecture would not need to change throughout this process - one set of 16 GDDR5 chips would simply be swapped out for another.

Microsoft never had the luxury of this moving target. With multimedia such a core focus for its hardware, it set out to support 8GB of RAM from day one (at the time giving it a huge advantage over the early PS4 target RAM spec) and with serious volume of next-gen DDR4 unattainable in the time window, it zeroed in on supporting DDR3 and doing whatever was necessary to make that work on a console. The result is a complex architecture - 32MB of ESRAM is added to the processor die, along with "data move engines" to courier information around the system as quickly as possible with bespoke encode/decode hardware to alleviate common bottlenecks. Bottom line: if you're wondering why Xbox One has a weaker GPU than PlayStation 4, it's because both platform holders have similar silicon budgets for the main processor - Sony has used the die-space for additional compute units and ROPs (32 vs. 16 in One), while Microsoft has budgeted for ESRAM and data move engines instead. From the Xbox perspective, it's just unfortunate for Microsoft that Sony's gamble paid off - right up until the wire, it was confident of shipping with twice the amount of RAM as PlayStation 4.