PS4 Pro Official Specifications (Codename NEO)

[Scott_Arm]

@iroboto

From http://timothylottes.blogspot.ca/2016/09/thinking-clearly-about-4k.html

RAW DATA
========

Using "Google Supplied" numbers for Tflops,

360 -> .24 Tflops [Xbox 360    ]
XB1 -> 1.2 Tflops [Xbox One    ]
PS4 -> 1.8 Tflops [PS4         ]
PRO -> 4.2 Tflops [PS4 Pro     ]
SCO -> 6.0 Tflops [Xbox Scorpio]

And looking at flop/pix in units of 1000,

                                    360   XB1   PS4   PRO   SCO
================================= ===== ===== ===== ===== =====
 960 x  540 @  30 Hz =  16 Mpix/s   15    77   116   270   386
1280 x  720 @  30 Hz =  28 Mpix/s    9    43    65   152   217
 960 x  540 @  60 Hz =  31 Mpix/s    8  __39_   58   135   193
1280 x  720 @  60 Hz =  55 Mpix/s    4    22    33    76   109
 960 x  540 @ 120 Hz =  62 Mpix/s    4    19    29    68    96
1920 x 1080 @  30 Hz =  62 Mpix/s    4    19  __29_   68    96
1280 x  720 @ 120 Hz = 111 Mpix/s    2    11    16    38    54
1920 x 1080 @  60 Hz = 124 Mpix/s    2    10    14  __34_ __48_
1920 x 1080 @ 120 Hz = 249 Mpix/s    1     5     7    17    24
3840 x 2160 @  30 Hz = 249 Mpix/s    1     5     7    17    24
3840 x 2160 @  60 Hz = 498 Mpix/s    0     2     4     8    12
3840 x 2160 @ 120 Hz = 995 Mpix/s    0     1     2     4     6

Wow, that did not paste well, but you can see Scorpio will be 24,000 flops/pixel at 4k30, where PS4 was 29,000 flops/pixel at 1080p30. So Scorpio at native 4k has less flops/pixel than PS4 does at 1080p. Probably not a good choice.

You can also see PS4 Pro at 1080p60 has more flops/pixel than PS4 does at 1080p30, so either 60Hz is a good target, or something like double 1080p, which is half of 4k (good for reconstruction).

Also, I saw this link because it was retweeted by Tiago Sousa, lead rendering programmer at id software. He's retweeted a number of people who have said that 4k is a poor choice and that HDR > resolution. Unlikely much of the industry sees things differently.

 

[Scott_Arm]

@BRiT Code tag? Who knew? Other than you.

 

[BRiT]

@BRiT Code tag? Who knew? Other than you.

Yeah, although sometimes the CODE tag will show abbreviated like the quote tag and need to be expanded to see the entire contents.It seems to be the closest thing to using fixed width fonts which is a must for ASCII tables.

The data is interesting to look at from the perspective of rendering power available per pixel.

Though there might be other equalizers such as bandwidth available or cpu available per pixel that could place the Scorpio at 4K@30Hz having 24 Kflop/pix closer to hitting the PS4 perf/pix quality standard pf 29 Kflop/pix.

 

[iroboto]

@iroboto

From http://timothylottes.blogspot.ca/2016/09/thinking-clearly-about-4k.html

RAW DATA
========

Using "Google Supplied" numbers for Tflops,

360 -> .24 Tflops [Xbox 360    ]
XB1 -> 1.2 Tflops [Xbox One    ]
PS4 -> 1.8 Tflops [PS4         ]
PRO -> 4.2 Tflops [PS4 Pro     ]
SCO -> 6.0 Tflops [Xbox Scorpio]

And looking at flop/pix in units of 1000,

                                    360   XB1   PS4   PRO   SCO
================================= ===== ===== ===== ===== =====
 960 x  540 @  30 Hz =  16 Mpix/s   15    77   116   270   386
1280 x  720 @  30 Hz =  28 Mpix/s    9    43    65   152   217
 960 x  540 @  60 Hz =  31 Mpix/s    8  __39_   58   135   193
1280 x  720 @  60 Hz =  55 Mpix/s    4    22    33    76   109
 960 x  540 @ 120 Hz =  62 Mpix/s    4    19    29    68    96
1920 x 1080 @  30 Hz =  62 Mpix/s    4    19  __29_   68    96
1280 x  720 @ 120 Hz = 111 Mpix/s    2    11    16    38    54
1920 x 1080 @  60 Hz = 124 Mpix/s    2    10    14  __34_ __48_
1920 x 1080 @ 120 Hz = 249 Mpix/s    1     5     7    17    24
3840 x 2160 @  30 Hz = 249 Mpix/s    1     5     7    17    24
3840 x 2160 @  60 Hz = 498 Mpix/s    0     2     4     8    12
3840 x 2160 @ 120 Hz = 995 Mpix/s    0     1     2     4     6

Wow, that did not paste well, but you can see Scorpio will be 24,000 flops/pixel at 4k30, where PS4 was 29,000 flops/pixel at 1080p30. So Scorpio at native 4k has less flops/pixel than PS4 does at 1080p. Probably not a good choice.

You can also see PS4 Pro at 1080p60 has more flops/pixel than PS4 does at 1080p30, so either 60Hz is a good target, or something like double 1080p, which is half of 4k (good for reconstruction).

Also, I saw this link because it was retweeted by Tiago Sousa, lead rendering programmer at id software. He's retweeted a number of people who have said that 4k is a poor choice and that HDR > resolution. Unlikely much of the industry sees things differently.

This is pretty sweet. and a great way to put things into perspective.

The numbers that I'm taking big interest in are:
1920x1080x60 for XB1 vs 3840 x 2160x60 Scorpio
1920x1080x30 for XB1 vs 3840 x 2160x30 Scorpio
Both of which Scorpio does have more flops per pixel.

When xbox games dip into 900p to 720p range, I think that's where we will see Scorpio also need to dip. MS has been pushing this dynamic resolution to help with that, it's probably going to be a big factor for games go forward. If XBO were to drop to 720p during dynamic resolution, we could probably expect the same with Scorpio dropping to 1440p of the sorts.

If both Pro and Scorpio are truly meant to be nothing more than just high res versions of this generations games, then next gen would ultimately be the device everyone is looking to take 4K to the next level, or something to that effect. But this gen, these numbers seem to line up well; XBO to XBS.

edit: overall I enjoyed that blog post. If you are following along the same lines of it, I don't disagree. What's right for developers or gamers, or what not. I do agree with the blogpost, but I think I'll see reality turn out slightly different. The marketing is so strong, resolution just a simple number...

 

[Scott_Arm]

I don't know ... do you really want to use Xbox One performance at 1080p as a baseline? PS4 is the real standard at 1080p on console, and with stuff like Battlefield it's at 900p. Put it this way, xbox one runs games mostly at 900p or 720p for a reason.

 

[iroboto]

I don't know ... do you really want to use Xbox One performance at 1080p as a baseline? PS4 is the real standard at 1080p on console, and with stuff like Battlefield it's at 900p. Put it this way, xbox one runs games mostly at 900p or 720p for a reason.

Obviously not my preference, as many have stated here in B3D, I think we'd all prefer a proper and amazing looking 1080p.

But there are limits to that as well, the more expensive the pipeline gets, the more pressure for the assets to look good. The costs would be higher on developers I imagine.

 

[Esrever]

1440p upscaled to 4k will look a lot better than 720p upscaled to 4k, which is what you'd have if you were using an xbox one vs scopio on a 4k tv. Dynamic resolution and maybe even foveated rendering can be used to make the most of the hardware.

 

[Wynix]

The number of the XB1(OG) is wrong, it's 1.31Tflops, the 1.2 number is from pre clock increase.

The numbers listed are very interesting, but I think the difference will be minimal as architectural improvements will offset most of the difference and then they could end up making it more powerful than promised.
I could easily see that 6Tflop number being a conservative amount used to prevent themselves from a sticky situation, maybe the real target is something like 6.5Tflops?

I think they both should have just focused on making a FHD-60Hz-HDR console, 4K just isn't a good option now and even next year it's only a maybe.
I'm starting to get anxious that the next generation will not be good enough to absorb the impact of both 4K and the expected graphics increase, although perhaps HBM bandwidth will help alleviate that problem.

 

[Love_In_Rio]

http://www.4gamer.net/games/990/G999024/20160910006/
In this 4gamer article Zenji Nishikawa talked with developers at SIE studios about developing games for PS4 and PS4 Pro.

• They don't use different 3D models or textures across PS4 and PS4 Pro due to the memory size limit. Both versions share the same assets.
• One developer said their PS4 Pro version chooses a mipmap level that is more detailed than the PS4 version at runtime by fragment shaders.
• As for LoD on PS4 Pro, they don't use more detailed models or don't do tessellation to increase polygon count. Some said due to the cost to reduce popping it's not realistic to do LoD optimization for PS4 Pro, others said optimizing it for 4K output makes it look optimally on both platforms since less detailed LoD models for far objects are not rendered in sufficient resolution on PS4 anyway.
• Upres of shadow maps or increasing shadow-map cascades for PS4 Pro is not done either due to the memory size limit.
• According to SIE, Lossless Delta Color Compression (of GCN 1.2) was implemented already in the original PS4. Which generation of it is present in PS4 Pro was not disclosed.
• All developers use sRGB textures and HDR light sources.
• The frame buffer format they use is 64bit float (16bit for each RGBA) that is converted to 10bit for each YUV as the HDR10 format.
• The developer of Farpoint for PSVR said for PS4 Pro the graphics is rendered in the resolution higher than 1080p then oversampled into 1080p to show it on PSVR. The frame rate is identical across PS4 and PS4 Pro. Better mipmaps give more details, oversampling gives AA and less pixel shimmer.

If all we, not 4K TV users, are going to get from more than 2 times the flops in the Pro GPU is downsampling what a great meh, a pointless investment.
A PC with a RX 470 allows at least to choose if you want to run your game at 720p120 1080p60 4K at 10 fps at high 4k at 25 at super mega low settings...
But this, seems to use the easy way, with the excuse of not enough memory, up the res for 4k and downsample for the rest.

 

[Clukos]

I'm starting to get anxious that the next generation will not be good enough to absorb the impact of both 4K and the expected graphics increase, although perhaps HBM bandwidth will help alleviate that problem.

The hardware is already here, the Titan X pascal is a 4k 60hz capable GPU

And does not even use HBM. Although, it is 12-14 tflop gpu (depending on clocks) with 480 gb/s mem b/w, these numbers shouldn't be that far off by 2019/2020 for console level hardware.

 

[Rodéric]

MIP map bias is not a good option.
PS4 Pro offering higher framerate is good enough already, at last 1080p@60Hz !

 

[Rodéric]

The hardware is already here, the Titan X pascal is a 4k 60hz capable GPU

And does not even use HBM. Although, it is 12-14 tflop gpu (depending on clocks) with 480 gb/s mem b/w, these numbers shouldn't be that far off by 2019/2020 for console level hardware.

I don't think any company will release a high-end desktop PC with a pad next gen either.

 

[goonergaz]

You need to look at the massive bandwidth increase and the memory increase that Scorpio has. Neither of which the 4Pro has.

lol, so we have some sayig the jump to pro is marginal/very small (etc) and now a 50% increase is massive!? BTW PS4 to Pro b/w is increased about half that of Pro to Scorpio so b/w will still be an issue.

When xbox games dip into 900p to 720p range, I think that's where we will see Scorpio also need to dip. MS has been pushing this dynamic resolution to help with that, it's probably going to be a big factor for games go forward.

"fully uncompressed pixels (10% of the time)"

 

[sebbbi]

MIP map bias is not a good option.

Depends. If they use temporal supersampling, they need negative mip bias. Real 4x supersampling (render at 2x2 higher resolution) fetches one mip level more detailed texture data (as UV gradients are halved). -1 mip level bias is thus recommended for temporal supersampling that mimics 4x supersampling. Mimicking 16x supersampling would need -2 mip bias. As a bonus, temporal jittering + negative mip provides temporal anisotropic filtering. The filter kernel shape is actually better than hardware aniso (which is usually just a line, not a trapezoid).

Negative mip bias if obviously a bad idea if temporal supersampling is not used, since it causes shimmering.

 

[goonergaz]

Back OT, the Pro is apparently using the Tiger (not Jaguar) CPU - is this a missinformation on the webs? If not does anyone know if that will mean much (ie anything over a standard overclock - maybe cooler running?)

 

[Shifty Geezer]

All these makes me think that PS4 Pro was pointless and they would have been better off waiting for a better upgrade.

What's the point of 4Pro? To make Sony money. If it enables them to make more money this gen than without 4Pro, selling to a subset of their audience and increasing market penetration through hand-me-downs and trade-ins (and cheaper 2nd consoles), it'll have served its point nicely.

 

[Shifty Geezer]

Back OT, the Pro is apparently using the Tiger (not Jaguar) CPU - is this a missinformation on the webs? If not does anyone know if that will mean much (ie anything over a standard overclock - maybe cooler running?)

Never even heard of Tiger cores.

 

[goonergaz]

Seen it mentioned on a couple of sites now, have tried a bit of digging but I'm useless at that sort of thing lol

"PS4 Pro CPU core of, but the power-saving CPU core "Jaguar" series of the same AMD and PS4, is now the most recent version of the "Tiger" core. Tiger is an improved micro-architecture, it is optimized core to FinFET process. AMD's Jaguar system CPU core, but synthesizer Bull (synthesizable) core, as well as the Seisesaizaburu of ARM core, is optimized for each process. Optimized Tiger core FinFET process, PS4 Pro can operate the CPU at a high frequency. Performance improvement in the architecture surface do not know yet."

 

[Rikimaru]

All Jaguar modifications are targeted at leakage and area reduction. It did not have microarch modifications.

 

[I.S.T.]

So prolly just a shrink then. Helps facilitate the much higher clocks.

Question: What clocks are Scorpio running at? I know the standard Xbox One runs at 1.75 GHZ, but is that different in Scorpio?