PlayStation 4K - Codename Neo - Technical analysis

[Shortbread]

In addition, PC CPUs are generally faster than the consoles at present.

If PS4 Neo is based AMD Polaris, then we'll see speeds (hopefully) closer to the PC. From all indications the current samples can be passively cooled, which does leave alot of room for higher clock speeds with proper cooling of course. The 900MHz mentioned in the leaks could just be minimum target specs as of now, since there is no finalized PS4 Neo SDK/hardware.

http://wccftech.com/amd-polaris-10-gpu-pictured/

The Polaris 10 GPU was demoed in the latest Hitman which utilizes DirectX 12 tech. The GPU offered 60+ FPS on 2560×1440 resolution when showcased at Capsaicin. The demo was short but a really interesting story from Ryan Shrout at PCPerspective reports that he got to see a demo of the Polaris 11 GPU playing back 4K VR content while being passively cooled. The Polaris 11 GPU is the smaller of the two siblings but it’s quite a hard task to run a 4K VR content at stable rate while being passively cooled, looks like the Polaris 11 GPU was able to achieve this feat with no problem at all.

The Polaris 10 GPU was featured in a tiny Cooler Master Elite case that comes in ITX form factor. That case has limited airflow and its quite impressive that the Polaris 10 GPU maintained more than 60 FPS in the small demo. We know from the Hitman benchmarks published last week that only the top-end cards such as the Radeon R9 Fury X are able to maintain more than 60 FPS at 1440p resolution.

 

[Michellstar]

I can't see it. Not when paired with that amount of memory bandwidth.

Yes, they have al least to use tonga level color compression, in order to save bandwidth.

 

[Michellstar]

If PS4 Neo is based AMD Polaris, then we'll see speeds (hopefully) closer to the PC. From all indications the current samples can be passively cooled, which does leave alot of room for higher clock speeds with proper cooling of course. The 900MHz mentioned in the leaks could just be minimum target specs as of now, since there is no finalized PS4 Neo SDK/hardware.

http://wccftech.com/amd-polaris-10-gpu-pictured/

Yes, but those are discrete parts, not paired with an 8 core cpu.

Moreover, polaris 11 it´s the low level performance polaris, around 16 CUs

 

[Shortbread]

Yes, but those are discrete parts, not paired with an 8 core cpu.

Moreover, polaris 11 it´s the low level performance polaris, around 16 CUs

Yes. However, the Jaguar (current PS4) TDP is fairly low (15-25W). As well, the PS4 Neo CPU TDP could be much smaller. My point was more so that when combined with proper cooling, we should see better/higher clocks for PS4 Neo (closer to its PC counterparts), than PS4 could achieve.

Polaris 10 should offer better TDP than PS4's current GPU.

 

[msia2k75]

PS4 Neo could be paired with a Polaris GPU? No compatibility issues?

 

[Akumajou]

I like that line of thinking...PS4K is gonna be a higher tier platform with die shrunk chip processes and additional unknown benefits of perhaps offering a bit more hardware.

In other words the PS4 was already a knockout at delivering a compromise between TDP, power consumption and thermals versus clock speeds.

Now although we might get some GameCube to Wii or Wii to Wii U type of hardware enhancements (which isn't the case) the new chips are gonna be too efficient if kept to that paper spec.

Basically there's no reason to obey that paper spec because the chips hand handle much more performance/efficiency so besides architecture revisions and benefits there should be some higher clocks which will still run either the same or cooler than current PS4.

Although this is gonna be yet another APU under a heatspreader the shrink is significant.

 

[mpg1]

huh rumors have Polaris 10 as pretty powerful: http://www.game-debate.com/news/?ne...ly Offers Near 980 Ti Performance For 300 USD

Wondering now if indeed PS4.5 is using Polaris will the performance be closer to something like the 390 than the 380X....

 

[pjbliverpool]

It's going to come down to clock speed. I understand Polaris 10 is rated at a max TDP of 175w but is expected to operate at much lower than that for most of the time. The big question is how close is the rumoured 911Mhz to Polaris 10's final clock speed, both in full form (40 CU's) and in cut down form (36 CU's).

 

[mpg1]

It's going to come down to clock speed. I understand Polaris 10 is rated at a max TDP of 175w but is expected to operate at much lower than that for most of the time. The big question is how close is the rumoured 911Mhz to Polaris 10's final clock speed, both in full form (40 CU's) and in cut down form (36 CU's).

True. 36CU Polaris 10 card will probably end up running at 1000MHz..

 

[mpg1]

Either way if the performance leap rumors by Polaris 10 in that article are true then I would not be shocked if the PS4.5 gpu comes close to a 390. (if indeed it is using Polaris).

I mean they are basically saying that the top Polaris 10 card is going to almost match 980ti/Fury X...

 

[MrFox]

There's a lot of talk about performance per watt on AMD slides... But what about performance per transistor count? Is this a metric that is improving or is the performance per watt only tied to new nodes and conservative clocking?

 

[pjbliverpool]

Either way if the performance leap rumors by Polaris 10 in that article are true then I would not be shocked if the PS4.5 gpu comes close to a 390. (if indeed it is using Polaris).

I mean they are basically saying that the top Polaris 10 card is going to almost match 980ti/Fury X...

But it's possible that Polaris 10 will be running at a much higher clock speed. The Pascal announcements suggest that speeds approaching 1.5Ghz should be possible on this node. GCN with 40 CU's at 1.5Ghz would certainly be competitive with Fury whether it's Polaris or not, at least as long as it isn't memory bandwidth constrained.

 

[mpg1]

But it's possible that Polaris 10 will be running at a much higher clock speed. The Pascal announcements suggest that speeds approaching 1.5Ghz should be possible on this node. GCN with 40 CU's at 1.5Ghz would certainly be competitive with Fury whether it's Polaris or not, at least as long as it isn't memory bandwidth constrained.

Point taken. That 911MHz could also get a bump in final specs...

 

[mpg1]

BASICALLY WHAT I AM SPECULATING IS THAT THE PS4.5 GPU WILL HAVE THE SAME AMOUNT OF JIGGAFLOPS AS A 980Ti. YOUR WELCOME.

 

[BRiT]

Speculation should be in the other threads, not this thread.

 

[mrcorbo]

Speculation should be in the other threads, not this thread.

For the purposes of discussion in this thread, it makes sense that we accept the conceit that the leaked specs are solid. If they aren't solid then there's no firm base on which to *reasonably* speculate. At a certain point when you start building speculation on top of speculation, the discussion becomes pointless.

 

[mrcorbo]

PS4 Neo could be paired with a Polaris GPU? No compatibility issues?

It's almost certainly a different GCN revision to the one used for PS4's APU, so if there are issues they appear to be manageable.

 

[Jawed]

APUs tend to have graphics IP that lags discrete GPUs. A change in process node doesn't necessitate a change in graphics IP.

 

[pjbliverpool]

On the other hand though, it's in AMD's interest to get their latest IP into the console world because that means more direct optimisation for their future product line in cross platform games. It's also in Sony's interest since Polaris will no doubt be a lot more efficient than Tonga even after you account for the process shrink, and so within the power constraints of a console, that means more performance. I'd also guess that from a competitiveness point of view in the PC market, AMD would be keen to move to Polaris in it's APU's as quickly as possible, and leveraging Sony R&D funding would be a great way to do that. Finally, the XBO and PS4 APU's both featured GCN1.1 level IP only a short time before it launched on the PC (and I think before it appeared in PC APU's).

All that combined with Polaris already being a 36 CU 14nm part makes me pretty confident PS4K will be Polaris based. I'll be pretty surprised if it's not in any case.

 

[Michellstar]

APUs tend to have graphics IP that lags discrete GPUs. A change in process node doesn't necessitate a change in graphics IP.

Not last time with current gen consoles, both were developed in parallel

Correct me if i´m wrong, but GCN 1.1 (Sea islands ip) was introduced with Bonaire in march 2013 when both Orbis and Durango specs were set in stone. It could happen again, we will get polaris around summer, and PS4k for the end of the year.