PS4 Pro Speculation (PS4K NEO Kaio-Ken-Kutaragi-Kaz Neo-san)

Status
Not open for further replies.
Trying to make this as much of an apples to apples comparison as possible:

Radeon 7970 - Late 2011 - 3.79TFLOPS - 3DMark 10,690 - Max TDP 250W - die size 352mm²
Radeon 390x - Mid 2015 - 5.9TFLOPS - 3DMark 18,670 - Max TDP 275W - die size 438mm²

I don't call that improving greatly.

Edit:Actually, lest I be accused of cherry-picking, let's look at Nvidia also and add some 3DMark scores since we all know not all flops are created equal:

GeForce GTX 680 - Early 2012 - 3.09 TFLOPS - 3DMark 10,660 - Max TDP 195W - die size 294mm²
GeForce GTX 980 - Late 2014 - 4.6 TFLOPS - 3DMark 18,540 - Max TDP 165W - die size 398mm²

I guess it depends on how you define improving greatly.

If you look at 3DMark/Watt, AMD improved by 50% and Nvidia almost doubled. In terms of performance unit area, the jump for AMD was very similar. I call that pretty significant.
 
So I'm confused. Sony has two different console designs and they can pick either one? Did they send out two different dev kits to developers?

Perhaps they have not decided what is a generation yet themselves or how to market it or built flexibility to react to the market after controlled leaks.

If neo will always just be a PS4 with slight better gfx but never have games target it as base spec. Higher clocked jaguar or equivalent fits right in to try and get to 60 to soak up some flops. PS5 when launched becomes base spec and PS4/neo development phases out.

Now if neo is designed to become base spec when the following revision comes out then a better cpu may be on the cards but that will cost more and Sony have to justify this now without alienating original PS4 owners by talking about neo becoming a base spec.

Who knows what Sony is planing post neo but I would think any reveal with a semi close launch needs a complete story defining and transitioning generations so people understand the product.

Given they are down in the numbers game perhaps they are waiting to roll out a stunning exclusive example of what neo will bring. As said before in the thread make it about games somehow.
 
I guess it depends on how you define improving greatly.

If you look at 3DMark/Watt, AMD improved by 50% and Nvidia almost doubled. In terms of performance unit area, the jump for AMD was very similar. I call that pretty significant.

It took between 2.5 and 3.5 years to get there. That allows plenty of time for another refresh at this cadence without those gamers who are willing to pay the premium having to sacrificing the larger relative performance increase they will be able to get in 8 or 18 months thanks to 14nm.
 
Last edited:
They are just going to adopt what we already have in the PC and mobile space where the platform owners will support hardware until the hardware isn't performant or where the majority of the user base has moved on. Developers will simply cater to the percentage of the user base that drives the most profit.
 
So the latest scoop from another insider on Gaf makes the option of a 5.5tf Neo less likely but apparently with some secret sauce? I wish Sony would an end to this fast.
 
Some early test shows that RX480 consumes 137 W @ 1266 MHz. Considering reasonable power consumption of PS4K, its GPU should't be faster than 1.1 GHz. In fact 1085 MHz is high enough because PS4K can have 5TF which is already useful for PR.
 
The smart way forward for neo is not to battle with scorpio on specs but to release it as soon as possible and at a price of no more than $399. If they do that they will have a year head start and at least $100 price advantage.

Lets remember that not only was Xbox One was weaker than PS4 it also launched at the same time and cost $100 more. So it was bad for a number of reasons. Sony can compete by being first to market by considerable amount and on price.
 
Last edited:
Similarly, XB launched more powerful and more expensive and later than PS2, while 360 launched before PS3 and, in the minds of the public, less powerful. Early and less powerful can often be a win. Although I kinda feel these new boxes will be selling to the current brand owners.If that's true, the important thing is making a box your current owners want to upgrade to. If Neo isn't enough of an advance that PS4 owners don't upgrade, it's market is probably limited, while if Scorpio somehow appeals to all XB1 owners so they all upgrade, that's 20 million units.

So actually, launching early might not be a great advantage for sales after all.

We literally know nothing about this new console paradigm!
 
Honestly the only "secret sauce" that would get me excited would be a huge array of the latest tensilica cores. Or an advanced implementation of reprojection into a ASIC module using depth map and geometry hints. At least something clever related to efficiency, instead of brute force general processing.

Upping the frame rate with a high-quality interpolation for any existing game would be nice.
 
Some early test shows that RX480 consumes 137 W @ 1266 MHz. Considering reasonable power consumption of PS4K, its GPU should't be faster than 1.1 GHz. In fact 1085 MHz is high enough because PS4K can have 5TF which is already useful for PR.
I think that 1266 was the max boost clock, it is not representative.
 
Some early test shows that RX480 consumes 137 W @ 1266 MHz. Considering reasonable power consumption of PS4K, its GPU should't be faster than 1.1 GHz. In fact 1085 MHz is high enough because PS4K can have 5TF which is already useful for PR.

The screenshots of the overclocking options for Polaris give a set of voltage and clock points at the apparent defaults.
The one at roughly 910 MHz matches the rumored Neo clocks, while bracketing the 5TF mark can put things between 1075 and 1145 where the rumored base clock would be.
The GPU chip itself might draw between 55% or 90% more power at the two higher points versus where Neo is supposedly clocked.
If the non-turbo clocks can be managed in the 120W range and are coupled with mildly lower memory clock, Neo running all-out could draw somewhat less power than what was measured for the original PS4 under heavy load.
If tasked with running a workload more appropriate for the original PS4, it might be possible to draw power as if Neo were a hypothetical slim model.

Perhaps with some creative math, it might be possible to fit two of these GPUs in the rough neighborhood of the launch PS4's power draw. That is, if one wants to add an extra bit of spice to the secret sauce speculation about having hybrid graphics as a fallback.
It would take an extra interface, higher silicon cost, and losing the straightforward memory space, however.

Speaking of special, one point of curiosity I have is what type of Polaris might be used in Neo. The PS4's GPU had some tweaks added to it, some of which do not have clear equivalents in the main line or Durango.

The one Cerny pointed out was the volatile flag for its cache lines. I have not seen direct discussion of how heavily leveraged that was so far. Assuming older titles have used it, would Polaris need to be re-customized to keep it, or does it have some other means to provide the same result without losing performance?
Polaris was listed as making changes to the L2, although it would be funny to find out that what's new for it is what's old for the PS4.
Cerny mentioned a tweaked geometry culling process with paired compute and vertex shaders, which would help discard geometry. Polaris has a "new" label for that process, too.
 
@3dilettante

Could you give a breakdown of wattage's you're thinking of here?

Assuming Sony are going for power use as close to OG PS4 as possible I would guesstimate that the APU will target ~80-90W leaving ~40-50W for 8GB GDDR5, HDD, secondary chip+512MB DDR3, Bluetooth+WiFi Blu Ray drive etc.

Assume 85% PSU efficiency and we'd be looking at ~140-150W at the wall. Maybe they'll go a bit higher but I doubt it especially if the Neo is even smaller.
 
@3dilettante

Could you give a breakdown of wattage's you're thinking of here?

Assuming Sony are going for power use as close to OG PS4 as possible I would guesstimate that the APU will target ~80-90W leaving ~40-50W for 8GB GDDR5, HDD, secondary chip+512MB DDR3, Bluetooth+WiFi Blu Ray drive etc.

Assume 85% PSU efficiency and we'd be looking at ~140-150W at the wall. Maybe they'll go a bit higher but I doubt it especially if the Neo is even smaller.

I was going with some old assumptions of 100-110W for the APU itself.
My recollection of a test that turned out to be higher than 150W was measuring something else, so upon further review the consumption figures would be closer.

The level of uncertainty for the RX480's power draw is very high, currently, but if I give the GPU itself about 110-120W around its rumored base clock, the 910 MHz point could be roughly between 53% and 64% of the desktop set points.
That could put ~60-70W, and I'm going with 27W for the GDDR5 memory (Anandtech's Fury X review put the 290X's 512-bit bus at 37-50, which I put to 40, cut that in half, then times 6.7/5.
Between 90 and 100 for GPU+RAM at full load. If Neo does not go for clamshell mode, the RAM consumption might be a watt or two lower. The CPU block, if Jaguar/Puma-level cores at 14nm, could probably drop to sub-15W. Possibly 100-110W for the APU plus GDDR5.

Depending on how much time the leaked tests for RX480 ran at the significantly less efficient clocks at 1200MHz and above, there could be more leeway. The turbo point looks about 2.5x worse than the 911 MHz point.
 
Tests I see put the ps4 around a 140w under game play. So I would assume 150w is the most they want to hit. The 4 gig card leaks point to 120w. I think that would be putting it real close unless they make a larger console or the ps4 can cool a more power hungry system
 
The alleged RX480 numbers also give a factor of 4 range in the voltage and clock product between the lowest power state and turbo, and 2.5-2.6 for the power state equivalent to Neo's rumored clock.
At an admittedly idealized level, the V^2xC product would give a rough picture of how much more power each state consumes than the previous, with the question of where in the range of leaked power states the GPU was operating in. The greater the portion that the leaked power numbers come from the least-efficient power states, the more Neo could improve by operating in the 911 MHz range.
Neo's rumored RAM speed is also modestly lower than the RX480's, which gives some additional wiggle room for the APU+DRAM power budget, which would be the largest contributor.

Since Neo is also supposedly the premium option, it might be able to adjust the investment in the cooler in order to allow for peak draw that is close but not equal to the PS4 at launch.
 
Status
Not open for further replies.
Back
Top