Xbox One G(Hz).
X1G. It's perfect. G for geforce. G for game. G for "G-Unit", like Fiddy's tag-alongs.
I was more thinking about this at a new node. If they're going to do any further work with the APU design, they can do it in one go at the next node rather than spending any further money on 16nmFF, and the timing would line up such that it's far enough away from the 1S launch anyway in terms of a market standpoint.
Could work, especially if you share development costs with the shrink of X1X. Potentially limited in the shrink by the 256-bit memory bus though, even if the interface is smaller for DDR3/4 than GDDR5. DDR5 insn't here till 2020+
Good thing about 16nm is that it's here now, it going to be really cheap once 10nm and 7nm are taken up, and MS already have the designs and the clocks figured. Frequency scaling on it is really good now. Well, for Nvidia anyway...
Forced AF may be more difficult. OneX can get away with it since you're looking at over 2x texture fillrate, and >4x main memory bandwidth (where the textures are being sampled) + 4x L2 cache.
On some X1X enhanced games like Halo 5 it's hitting 5X the resolution of the X1 version with enhanced texture filtering to boot (not 16x but still improved). I know aniso isn't free, but a huge boost to effective memory bandwidth and 20%+ jump in raw clocks should be more than enough to force apply a high level of aniso while also significantly increasing performance across the board.
[snip]...That being said we moved from 28nm -> 16nm and we got a boost of 853Mhz -> 914Mhz.
To be fair, the priority with X1S was dropping power consumption. And they nearly halved it. The small clock boost was to prevent HDR hurting performance. When MS targeted frequency, they hit 1.18 gHz, and on a chip with 3x the CUs.
I do wonder how much further the clock can go here on a smaller node. Better cooling would increase costs, so I don't see them going that route. if they can get up to 1200 Mhz though
lol, but without additional cooling I don't see this as possible.
1.2 wouldn't be a problem on 7nm, at least if TSMCs guideline figures area meaningful. Up to 40% higher frequency for the same power or something (I'm sure there are some caveats!). Limits of physically shrinking the chip, in the form of IO, might be a show stopper though. :/
I didn't say it wasn't true, it certainly is. Contention is certainly there, and I'm willing to bet under heavy load the contention will probably drop to 20-40 GB/s, which is 30% - 60% loss outright. Which is idealistically close to the bandwidth of it's available DMA channels.
Why does XBO need DDR3 at it's peak? It alone only provides 67GB/s at maximum output as a pure read or write function. With memory contention and lots of read/writes it's probably closer to 20GB/s - 40 GB/s.
I think this is where the advantage of DDR4 and DCC can be readily demonstrated. For a dynamically scaling game with a locked 60hz, a CPU boost wouldn't result in further BW erosion for the GPU. So any
increase in BW would mostly help the GPU.
Taking your 20 ~ 40 GB/s example figure for the GPU from the current DDR3 setup, you can see that a DDR4 2666 setup would offer something like 37 ~ 57 GB/s. That's a 40 ~ 80% increase.
Throw in DCC and you're looking a system that could show performance gains even greater than the percentage by which the SoC's clock was increased.
Even
wihtout higher clocks or DCC, such an X1 revision would run games like Wolfenstien 2 at a higher average resolution and higher average frame rate.[/quote][/QUOTE]
