• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

AMD Confirms Sony PlayStation Neo Based on 14nm CPU and Polaris?

What would be wrong with an ARM CPU?

As has already been said, it completely fucks up/over BC again, which I don't think they want to do again. And as for why they didn't want to do it now, I don't think it's actually been proven that ARM technology could scale up to the level that they wanted/needed for consoles.
 

jdstorm

Banned
The Wii U released November 2012 so it has not been out 5 years and this year it has almost nothing. So pretty much 3 years of support. PS4 will be supported for another 2-3 years worst, but I guess you are of the mindset that your PS4 is obsolete the moment PS4.5 drops?

This year I've personally bought or wanted to buy Xenoblade X, Fast Racing Neo, Star Fox and before the NX mandated delay Zelda was due at X mass. So that's 4 games in the last 12 months. The Wii U has had solid Indy support and I assume that will continue, since it's apparently easy to port Unity games to. In general despite a few game droughts I've been very happy with the WiiU.

I definitely don't think the PS4 will be obsolete after Neo Drops, Iam worried about Bad Ports, given the change in CPU cores in the Neo to 14nm Puma cores compared to the PS4s Jaguar cores in addition to the GPU upgrade. The PS4 already has overly long load times on some of the more demanding titles and this only seems to make it worse. For Optimised exclusives I will gladly make an exception and use a PS4 but any 3rd party game that will be less optimised due to having to launch on multiple platforms. I'm no longer interested
 
About that cheaping out, I'm not sure how much performance bump that, say, a $100 BOM extra will buy with the available tech and TDP circa 2013.

It would certainly have bought a much better CPU, which would have resolved a lot of PS4's issues that are caused by having to offload CPU work to the GPU. As for the Xbone, well - they basically totally cheaped out because TVKINECTTVLOLTVTVTV Don Mattrick and all that stuff.
 

geordiemp

Member
I definitely don't think the PS4 will be obsolete after Neo Drops, Iam worried about Bad Ports, given the change in CPU cores in the Neo to 14nm Puma cores compared to the PS4s Jaguar cores in addition to the GPU upgrade.

Ps4 is the market leader, your concern for bad ports, ports from what exactly ?

Games made and optimised for Ps4 is like having PC options with 2 settings as opposed to 100, it really seems that simple to me.

From your game interest you should be more worried about what Nintendo will do in the console space, although I think they will make bank on mobile so will be fine whatever. Console, handheld and mobile is also stretching resources, does not compute unless the system plays them all.........I am dying to see what the gimmick is, mystery is so much fun.

Sony already seem comfortable this gen and Neo / VR OPTIONS will just cement that lead imo.

My concern for Neo is the CPU power and even more so the minimal bandwidth increase, thought they would go GDDR5x....
 

E-Cat

Member
It would certainly have bought a much better CPU, which would have resolved a lot of PS4's issues that are caused by having to offload CPU work to the GPU. As for the Xbone, well - they basically totally cheaped out because TVKINECTTVLOLTVTVTV Don Mattrick and all that stuff.
The most tragic part is that the Xbox One main SoC is actually HUGE @ 363 mm² and ~5B transistors. For comparison, the PS4 APU is 347.7 mm². MS really focused on the wrong points with their design.
 

pastrami

Member
It would certainly have bought a much better CPU, which would have resolved a lot of PS4's issues that are caused by having to offload CPU work to the GPU. As for the Xbone, well - they basically totally cheaped out because TVKINECTTVLOLTVTVTV Don Mattrick and all that stuff.

And what CPU is that?
 

shark sandwich

tenuously links anime, pedophile and incels
As has already been said, it completely fucks up/over BC again, which I don't think they want to do again. And as for why they didn't want to do it now, I don't think it's actually been proven that ARM technology could scale up to the level that they wanted/needed for consoles.
You are right about BC but wrong about ARM being able to scale. High-end ARM cores (Cortex A72, Qualcomm Kyro, Apple's custom CPUs) are already far more powerful than the Jaguar CPUs found in PS4. Even 2012's Cortex A15 was core-for-core on par with Jaguar.

I am convinced MS and Sony's decision to use AMD x86-based SoCs was based on a) being able to get a good integrated GPU and b) AMD gave it to them for incredibly cheap. Cost and size were clearly far, far higher priorities than performance.
 
And what CPU is that?

They could have at the very least gone with Puma cores and/or accepted worse yields and went for higher clockspeed. The current Jaguar CPU itself can easily go up as high as 2.1Ghz I believe and Puma is around 20% more efficient. The 1.7GHz Jaguar was a terrible choice, especially for a console released for the age of async compute. There's also the fact that despite its higher clockspeed the Xbone CPU has terrible memory bandwidth so there's the cheaping out on RAM issue as wellç
 

pastrami

Member
They could have at the very least gone with Puma cores and/or accepted worse yields and went for higher clockspeed. The current Jaguar CPU itself can easily go up as high as 2.1Ghz I believe and Puma is around 20% more efficient. The 1.7GHz Jaguar was a terrible choice, especially for a console released for the age of async compute. There's also the fact that despite its higher clockspeed the Xbone CPU has terrible memory bandwidth so there's the cheaping out on RAM issue as wellç

Wikipedia is telling me Puma was only in production from mid-2014. Not sure how that would help a 2013 launch. So, what else? I mean, yeah, faster clock speeds would have been nice, but they have to balance clock speeds with the heat the chip produces. I don't think they had chips capable of 2.0GHz and just decided to cap it at 1.6GHz. I just don't know if the increase in price would have been worth a marginal increase in CPU speed.
 

Marlenus

Member
edit for slightly underestimated r9 270 performance thanks to amd's insanely convoluted, everchanging naming schemes

2.2x r9 270 performance barely gets you r9 390 performance , let alone fury x

If you add an additional 10 percent to the r9 270 (to account for 212mm² die vs 230mm² for polaris 10) then you still don't get near r9 fury x performance, maybe about r9 390x.

2.2x is perf/mm.

2.5x is perf/watt.

Here is the most recent TPU review. The 270X is about 5% faster than the 270 so based on the 1440p results the 270 would be about 106% of the tested 950.

Multiply by 2.5 for perf/watt and you get 265% which is a few % behind Fury X. Add 10% and multiply by 2.2 and the performance is between Fury and Fury X.

At 1080p the 270 would be at 100% so 2.5x is 250% which is ahead of Fury X. Do the perf/mm calc and you get 242% which is also ahead of Fury X.

So ballpark figure of around Fury X seems right based on scaling from a similar sized, similar TDP, same market segment card.

It might be a bit less but that is still > 390X perf. The idea that the top version will only be 390 perf is pretty daft IMO.
 
RE: DX11 I think you mean that PS4 games have to run unmodified on Neo so it's a DX 11. Maybe maybe not.

Remember it's a semi-custom and since it's using GDDR5 not using HBM or DDR4, and it's a APU, the ARM IP will still be in Southbridge. Blocks, not the entire GPU, of already designed Polaris FinFET dGPU will be fit into the smallest space possible on the die. It will follow the Launch PS4 GPU design in more ACE controllers which is what Polaris does for VR, beyond that it can be custom.

Leaks from developers have this:

PlayStation 4/Neo Game Compatibility

Sony is very keen on developers supporting both systems simultaneously, and is motivating developers to ensure that there is Neo support in all PS4 titles from Q4 onwards this year. It's also firm on unifying the platforms with little or no exceptions:

Neo-only or PS4-only games are not permitted (remember that Neo can still run unenhanced titles - developers are simply prohibited from locking out audiences of either console).

All games you purchase, whether via disc or from the PlayStation Store, should offer both PS4 and Neo functionality with no extra costs associated in running titles on a different console.

All new titles with Neo support use unified packages that run on both platforms. The CPU binary is identical, while three GPU binaries (shared, PS4-specific and Neo-specific) are all contained in the same package.

All DLC and additional content is entirely cross-platform. Unified downloads are used for both this, and basic game patches.

Neo support for old games is allowed via 'forward compatibility' patching - but this will not be allowed for new titles.

Developers cannot supply exclusive gameplay features for Neo owners. If the game has a split-screen mode, it must be available on both systems. However, modes can be enhanced - so a two-player split-screen mode on PS4 could be expanded to allow a four-player variant on Neo.

Developers cannot add exclusive content to either PS4 or Neo systems.
If there's a bug in Neo-specific code, developers are not allowed to divert Neo owners to the PS4 codepath. The game must be fixed.
Beyond some way to run PS4 launch games it can have DX 12 features..we don't know anything yet. Southbridge is getting an update as NEO can support 4 separate screens/players while the Launch PS4 can support only two.
 

Renekton

Member
They could have at the very least gone with Puma cores and/or accepted worse yields and went for higher clockspeed. The current Jaguar CPU itself can easily go up as high as 2.1Ghz I believe and Puma is around 20% more efficient. The 1.7GHz Jaguar was a terrible choice, especially for a console released for the age of async compute. There's also the fact that despite its higher clockspeed the Xbone CPU has terrible memory bandwidth so there's the cheaping out on RAM issue as wellç
Puma is Jaguar with slight tuning, which I don't think was available in that 2012-2013 period? Plus 2.1ghz is only feasible now with Neo on 14nm. Current PS4 is a hair dryer in UC4, so they hit the thermal wall on AMD tech.
 

MuchoMalo

Banned
Polaris 10 is supposed to be around 36CU (2304 shader units) for 130W~. That's your 2.5x perf/watt. R7 370 is 110W for 16CU (1024 shader units). R9 390 is 40CU (2560 shader units) for 250W.
So at the same clock, let's say 1ghz, Polaris 10 is 4.6Tflops while R9 390 is 5.2Tflops. Although, with the smaller process node, people expect clock bumps. Let's say 1.3Ghz for Polaris 10. It would be 6Tflops, for a 130~150W GPU. So that would be a GPU 20% faster than R9 390 for nearly twice less power consumption. For a rumoured 299 dollars.

THe 36CU version is not the fully enabled chip. The full one should be 40CU.
 

Snail

Banned
Puma is Jaguar with slight tuning, which I don't think was available in that 2012-2013 period? Plus 2.1ghz is only feasible now with Neo on 14nm. Current PS4 is a hair dryer in UC4, so they hit the thermal wall on AMD tech.

That's right. Puma was not an option for these consoles.

Also, Jaguar might very well go up to 2.1GHz but the power consumption (and so the heat) exponentially increases over 1,6GHz.

So again, Mark Cerny and the Sony hardware engineers are not stupid. They've made the right decisions from the cost and what was available.
 
As has already been said, it completely fucks up/over BC again, which I don't think they want to do again. And as for why they didn't want to do it now, I don't think it's actually been proven that ARM technology could scale up to the level that they wanted/needed for consoles.
Consoles have trash CPUs though, I'm pretty sure that Apple has better per core performance than a Jaguar.
 
2.2x is perf/mm.

2.5x is perf/watt.

Here is the most recent TPU review. The 270X is about 5% faster than the 270 so based on the 1440p results the 270 would be about 106% of the tested 950.

Multiply by 2.5 for perf/watt and you get 265% which is a few % behind Fury X. Add 10% and multiply by 2.2 and the performance is between Fury and Fury X.

At 1080p the 270 would be at 100% so 2.5x is 250% which is ahead of Fury X. Do the perf/mm calc and you get 242% which is also ahead of Fury X.

So ballpark figure of around Fury X seems right based on scaling from a similar sized, similar TDP, same market segment card.

It might be a bit less but that is still > 390X perf. The idea that the top version will only be 390 perf is pretty daft IMO.

Why are you multiplying both by perf/watt and perf/mm²

You should ONLY be multiplying by performance/mm² since we are comparing the performance of a 212mm² die to a 230mm² one,not by perf/watt
which is 2.2 (and x 1.1 to account for the minor die size difference)

Again: you barely arrive at 390x performance from doing this.

Idk why you latch on to the perf/mm² figure (which is a best case scenario in all likelyhood, like all these marketing numbers are) when it's painting such a dire picture.
Wait for benchmarks and hope that they can clock it high enough to approach fury x performance and not just be a 390 performance gpu, that's all there is to do right now.
 

Panajev2001a

GAF's Pleasant Genius
That was the Swift core? I think it lost to Intel's Atom at the time.

Good then that it has 8 cores and not 2 and the console design is centred around shifting the bulk of the expensive workload over to the GPU ;).

The SoC they ended up with on both sides is not trash nor bad for the power consumption and price targets they had (nor were those targets low to begin with), but I guess it is a lot more fun to joke about Jaguar cores being slower than a 486DX :p.
 

Snail

Banned
Good then that it has 8 cores and not 2 and the console design is centred around shifting the bulk of the expensive workload over to the GPU ;).

The SoC they ended up with on both sides is not trash nor bad for the power consumption and price targets they had (nor were those targets low to begin with), but I guess it is a lot more fun to joke about Jaguar cores being slower than a 486DX :p.

Exactly.

And we should also mention that the console has a fixed hardware, so if a developer runs into CPU bottleneck that's entierly their fault.
They should've planned or designed differently, optimise more, or shift load towards the GPU.

People still think that the PS4 is a PC, just because it has X86 and GCN, but it's so far from being true...
 
So the way that this is positioned is as if Sony was looking to make an even smaller change for the PS4 and decided that it was cheaper to go with this new hardware that gives them what they want thus, the neo is born. Hmmmm 🤔🤔
 

c0de

Member
My concern for Neo is the CPU power and even more so the minimal bandwidth increase, thought they would go GDDR5x....

Especially considering the actual bandwidth when both CPU and GPU access the RAM at the same time, it's way less than the theoretical maximum. Perhaps they improved the busses, though.
 

platocplx

Member
This makes it sound like Neo exists mainly because just doing a new SoC ended up being a financially smarter move than a die shrink of the existing one.

yeah which i think would make sense for any company, especially if its easier to just adopt a newer chip vs having to spend a shit load of money to redevelop the chip they have now and make it smaller.
 

Marlenus

Member
Why are you multiplying both by perf/watt and perf/mm²

You should ONLY be multiplying by performance/mm² since we are comparing the performance of a 212mm² die to a 230mm² one,not by perf/watt
which is 2.2 (and x 1.1 to account for the minor die size difference)

Again: you barely arrive at 390x performance from doing this.

Idk why you latch on to the perf/mm² figure (which is a best case scenario in all likelyhood, like all these marketing numbers are) when it's painting such a dire picture.
Wait for benchmarks and hope that they can clock it high enough to approach fury x performance and not just be a 390 performance gpu, that's all there is to do right now.

1) my numbers clearly sit between Fury and Fury X at 1440p and exceed Fury X at 1080p. This is based on TPUs most recent benchmarks. Allowing for cherry picked numbers still gives > 390X performance. This also matches a simple Tflop check vs 390x as 2560 SPs @ 1.2 GHz > 390X or 2304 SPs @ 1.3Ghz = 390X.

Apple and NV both had 45% clock speed bumps going from 28nm to 14nm which makes those clockspeed rather conservative.

2) you do both to make sure your start point is reasonable as all these things are related to each other through power draw. More Watts = more perf/mm, up to a point.

If you include a Tflop comparison then you have 3 methods of predicting performance (well same or similar uarch performance) that all point to the same performance area which is around Fury X.

If the die size or power draw is different to what has been leaked then obviously my numbers will be off.
 

Schnozberry

Member
What does this have to do with PSVR?!

Nothing. It's pure speculation on the part of the author. In the comments below he's now saying that the CPU is now a relative of Zen rather than Jaguar/Puma based, which makes little sense as it would require extensive work not to break compatibility with existing software. The slides in the article are only tangentially related to the semi-custom business.
 

mckmas8808

Mckmaster uses MasterCard to buy Slave drives
Nothing. It's pure speculation on the part of the author. In the comments below he's now saying that the CPU is now a relative of Zen rather than Jaguar/Puma based, which makes little sense as it would require extensive work not to break compatibility with existing software. The slides in the article are only tangentially related to the semi-custom business.

Okay cool. I pre-ordered PSVR and I don't want to be surprised by anything.
 

pastrami

Member
Die shrinking being pricier than buying a new GPU? How is that even possible?

Where does it say that? I think the issue is that if you are spending that much money for a die shrink, you might as well spend a little more for a redesign. This gives you a stronger console, and a reason to keep a $399 price point.
 

Inuhanyou

Believes Dragon Quest is a franchise managed by Sony
So the Neo is a PS5 in everything but name only.
Combine that with a short lifecycle (3years for the ps4) and Next gen I will definitely be looking at another platform to game on. Nintendo + a Windows PC here I come

30% CPU bump and a 2x GPU upgrade is not a 'PS5', especially since this is not going to be used independently of the ps4 development platform.

by the time PS5 actually comes out, a Zen CPU upgrade, a RAM upgrade and a further GPU upgrade that allows devs to take advantage of that power full stop will be a real game changer.

I think some people are going to have to tone down the salt, Neo is coming whether you want it or not.
 

DjRalford

Member
, Neo is coming whether you want it or not.

NEO.jpg
 

Marlenus

Member
Die shrinking being pricier than buying a new GPU? How is that even possible?

Because finfet is fundamentally different from planar so there is no simple die shrink. It requires a redesign anyway so why would Sony/MS pay when AMD has already designed Polaris?
 

blu

Wants the largest console games publisher to avoid Nintendo's platforms.
They could have at the very least gone with Puma cores and/or accepted worse yields and went for higher clockspeed. The current Jaguar CPU itself can easily go up as high as 2.1Ghz I believe and Puma is around 20% more efficient. The 1.7GHz Jaguar was a terrible choice, especially for a console released for the age of async compute. There's also the fact that despite its higher clockspeed the Xbone CPU has terrible memory bandwidth so there's the cheaping out on RAM issue as wellç
Puma is the exact same uarch as Jaguar.
 

blu

Wants the largest console games publisher to avoid Nintendo's platforms.
The layout was tweaked to reduce leakage so it clocks higher at the same power draw.
That does not make an improved uarch. But yes, it can be clocked higher.
 
Die shrinking being pricier than buying a new GPU? How is that even possible?

The most important thing about reducing production costs is the amount of possible customers for such a die shrink.

Outside of Sony and Microsoft who would still buy an outdated GPU?
 

Marlenus

Member
That does not make an improved uarch. But yes, it can be clocked higher.

True, puma was just a design improvement. Finfet is different though so not sure if the changes that were made to jaguar would carry forward on the new node.
 

labx

Banned
Don't see the need of releasing NEO this year. Sony can milk their fanbase with the ps4 alone for another 6 months. Holidays will be PSVR. Spring NEO or they will be an announcement at E3... "ITS AVAILABLE NOW." 400 dollars with no VR. 850 the bundle with VR.
 

notaskwid

Member
Don't see the need of releasing NEO this year. Sony can milk their fanbase with the ps4 alone for another 6 months. Holidays will be PSVR. Spring NEO or they will be an announcement at E3... "ITS AVAILABLE NOW." 400 dollars with no VR. 850 the bundle with VR.
You can't 'it's avaliable now' a console or anything physical nowadays lol. We are not in the 5th generation anymore.
 

ZoyosJD

Member
True, puma was just a design improvement. Finfet is different though so not sure if the changes that were made to jaguar would carry forward on the new node.

I've mentioned this earlier. That assessment is pretty much spot on which is probably why we are seeing confusion from some leakers on which CPU architecture is being used.

Don't see the need of releasing NEO this year. Sony can milk their fanbase with the ps4 alone for another 6 months. Holidays will be PSVR. Spring NEO or they will be an announcement at E3... "ITS AVAILABLE NOW." 400 dollars with no VR. 850 the bundle with VR.

For financial reasons they likely won't be releasing outside of the holiday season unless yields and manufacturing force them to. Not even Apple tries to pull the "ITS AVAILABLE NOW" line with modern hardware.
 
Top Bottom