First of all, if he was talking about a 50% increase in CPU cores then
he would be correct. However, GPUs are a highly parallel environment.
CPUs are regarded generally as the serial processing unit and GPUs as
the parallel processing unit.
In the case of 12 CUs & 768 Shaders vs 18 CUs & 1152 Shaders on a
GPU, more is always better. While I admit that in itself will not
create a 50% power advantage; it is a considerable advantage.
He also neglects to mention another important point; the PS4 GPU has 32
ROPS & 72 Texture Units vs the Xbox One GPU's 16 ROPS & 48
Texture Units; this in itself is another considerable advantage that
will cause the PS4 to age much better.
The Xbox One GPU has an 853MHz vs 800MHz advantage in clock speed over
the PS4 GPU. I'm going for the PS4 but any bumps in tech specs on any
system is always a plus for the industry. However, 12 Compute Units @
853MHz vs 18 Compute Units @ 800MHz(?) on a GPU and the latter still
wins out easily.
MS/Xbox continue to add the bandwidth of separate pipes together in
their pr statements to try and fool the uninformed when in reality it
does not work like that. It is NOT 272GB/s!
That 204GB/s peak theoretical read/write is limited to a 32MB chunk of
low latency eSRAM but the main DDR3 2166MHz RAM is unified pool limited
to 68GB/s of peak theoretical bandwidth. Around 20GB/s of that is
inherently available to the CPU. The Xbox advantage here is a 30GB/s
CPU-to-GPU HSA link.
On the PS4 you have a single, simple unified pool of GDDR5 5500MHz RAM
which acts as a unified address space with 176GB/s of peak theoretical
bandwidth across the board. Again, around 20GB/s of which is inherently
available to the CPU. The PS4 has a 20GB/s CPU-to-GPU HSA link.
...and before anyone brings up latency advantages, in a traditional PC
arena the DDR3 has a clear edge and the GDDR5 suffers (though the gap
isn't quite as pronounced in the first place as some would say),
however, hUMA/HSA implement a special memory controller configuration
which neautralises much of this issue; and even without this, the better
choice and compromise in a GPU-centric gaming console would be GDDR5.
So...the Xbox Advantages? A CPU with a marginally higher clock (though
PS4 CPU clock is yet to be confirmed) and a notable amount of additional
bandwidth coming in directly from the GPU. It's likely the SHAPE Audio
chip is stronger than the Audio chip in the PS4 so there will be a
slight ease in CPU load vs the PS4. A GPU with 53MHz clock advantage.
And...the PS4 advantages? A GPU with 29% additional computational power,
a 50% increase in Compute Units/Shaders, a 50% increase in texture
units, a 100% increase in fillrate and a 700% increase in compute queue
granularity. A simpler, faster pool of GDDR5 RAM that acts as a unified
address space.
Sony also have the stronger tools this time around, have had a better
policy from the outset, they've put together a console with more raw
power, likely with a chip die that is ironically smaller and
subsequently cooler, in a smaller, more iconic, cheaper box with an
internal psu that is selling in more countries earlier.
Multiplatform games will for the most part be developed to the lowest
common denominator; and the fact is that overall that is the Xbox One.
So we're only likely to see a slightly stronger Image Quality &
Performance on PS4 for most multiplats.
The first party PS4 titles though will pull away considerably though by
the second and third generation of titles; and the lack of
fillrate/texture units will age the Xbox One more by then.
