Console hardware: what Sony/MS went with versus what they should have

[soccerballtux]

I think it's fine, developers are lazy and can kiss my shiny FX-8310. I am a C programmer. everything should be written in C

 

[Cerb]

They should have specified them some form of higher clock speed operation, using fewer game cores. That would have helped. Aside from that, the only other way would have been to repeat past generations, and take huge losses until it could all fit on one die; but losing possible performance in the process from needing buses on the PCB. They could have done a huge die, and accepted big losses until a shrink, of course, but they were going for only small early losses.

Instead of an 8 core, I'd have rather had a faster Quadcore (or a Big/Little combination with a Quadcore x86+ARM Processor for OS or something like that).

Where do you fit all that? Your preference would take as much size as their custom APUs all over again just for the CPU, meaning either the CPU or GPU would need to be far away from the memory controller; and hat's if it were even doable; or they'd have a big server-CPU-sized chip that would be horribly expensive, until shrunk. A single SR module would take around as much room as they did for a doubled up Jaguar set, if not more, and use more power for its actual performance offered (I agree with Tuna-Fish that if MS or Sony wanted a fat synth core, they'd have bent over backwards to make it happen).

They had other options, if they wanted $700+ BOM costs at launch, this time around. Neither of them did.

Also keep in mind that MS and Sony have both made bad blood with nVidia, and MS is far from keen on licensing external Intel CPUs. For both companies, that they could do the hardware in-house was a big deal, leaving few options, even if they had been willing to wait for updated usable fat AMD CPUs. And make no mistake: there are, even today, no licensable ARM or Power CPUs that can come even close to Jaguar, in anything but SIMD; and Jaguar is, IMO, especially impressive in that area, for the little narrow CPU it is (from what I've read, console devs like it quite a lot for vector arithmetic, as well).

ARM is the future of consoles, no one wanted to buy a 700$ PS3 and it almost bankrupted Sony. If the APU is 100$ like suggested I'm sure Sony feels it still costs too much.

Lots of people wanted to buy those PS3s, and did, though often at far lower than $700. But, Sony mismanaged the development, had people in technical positions behind it that were like the Itanium guys, and what didn't sell was the games, in relative terms, until the platform matured. I knew more people that bought PS3s because other early BD players were steaming piles than that were gamers. Sony did not count on that, and so lost a lot of money on the console's early sales. Were they not the juggernaut that they are, it could have been a Sega or Atari death knell release, due to their hubris*.

Developers will eventually figure it out. This will force them to figure out how to properly multithread.

For the most part, they already know, and knew years ago. The problem is that some games are more complicated than your entire Windows OS. Tweaking and rewriting so much of that code has taken ages, and some of it was fruitless without DX10+ improvements, and other OS improvements, on the Windows side (which the XBoxes basically are). Now, using engines that scale out fairly well, they still have to be able to break problems up, and do so without lots of copying (memory may be cheap, but bandwidth is not, and using an Erlang-like, or Haskell-like, way to manage shared data will chew up on-CPU and/or RAM bandwidth in a heartbeat), and without creating a traditional mire of locks (a bad habit largely born from SMP database processing, and memory being expensive). That is not a trivial task for any programmers, even the genius-level ones. It takes time and money, both of which are always in short supply.

2GB is not enough for high rez textures today.

Agreed. I use more VRAM than that with games that came out on the prior consoles (I've used up to 2.6GB in Skyrim, and I'm still not done tweaking my mod choices for my new install; and 1GB was already a limitation for me by the end of 2010, w/ FO3, using not so high res textures).

* I can respect Atari's demise, though, as they misread the market, but made some innovative stuff while doing so. It sucks, and looks bad in hindsight, but there's no shame in failing that way, IMO.

 

[Enigmoid]

Something quite subtle that people may be forgetting is that the xbone/PS4 APU is NOT a true 8 core chip, rather it is two quad core jaguar clusters on die connected by a bus. While L2 access on the same module is a minimum 26 cycles it increases to a whopping ~190 cycles when accessing cache on the other module (RAM latency is ~220 cycles).

This is a terrible barrier to effectively multithreading game code; for performance purposes code needed both by the two 'odd duck' cores and the four game cores must pretty much be present in both caches for prompt execution. Thus it seems quite likely that two of the 6 console cores are going to be used for relatively remote tasks that do not depend much on the main game code.

 

[ultimatebob]

Intel i5 with 2.5+ teraflops GPU.

If you could have got that in a $400 system in late 2013, it would have been awesome

Seriously, though, I'll bet that Nvidia could have made a custom ARM APU that would have blown the doors off of AMD's offering. Instead, they invested that money and development effort into the failed Shield product line. Sad, really.

 

[Erenhardt]

Something quite subtle that people may be forgetting is that the xbone/PS4 APU is NOT a true 8 core chip, rather it is two quad core jaguar clusters on die connected by a bus. While L2 access on the same module is a minimum 26 cycles it increases to a whopping ~190 cycles when accessing cache on the other module (RAM latency is ~220 cycles).

edit. nvm, i googled it for myself

 

[itsmydamnation]

Anything?

or you expect us to take your word as granted?

no thats correct, but the point is who cares, its not like you don't have control of your threads.

 

[escrow4]

An Intel i3 + custom Iris Pro variant should have been the choice. And before you mention price, Intel is flipping Atom's for nothing, I'd bet they would be flexible in price given they have put so much work into their iGPUs.

 

[frozentundra123456]

An Intel i3 + custom Iris Pro variant should have been the choice. And before you mention price, Intel is flipping Atom's for nothing, I'd bet they would be flexible in price given they have put so much work into their iGPUs.

Iris pro is far from an atom. They are subsidizing atoms in some markets, because they want to get into mobile desperately. I definitely think they would not make the same concessions with a custom iris pro, especially to get into the console market.

 

[Idontcare]

Iris pro is far from an atom. They are subsidizing atoms in some markets, because they want to get into mobile desperately. I definitely think they would not make the same concessions with a custom iris pro, especially to get into the console market.

IMO the Iris pro SKUs are about as relevant as the Devil's Canyon SKU when it comes to topic of APU performance. It is a niche product created to sate a niche market segment for reasons (marketing halo) that extend beyond the 5k-ft perspective.

One thing that Intel does well, and you have to step back to appreciate this, is they shotgun as many niche markets as possible, all at once.

Not only is this self-evident in their diverse product offerings but also in the very existence of their Intel Capital group.

 

[seitur]

2gb is more than enough for 1080p gaming. The bloated memory demand of newer games is somewhat due to current gem console unified memory.

Previous gen AAA games were consiciously designed in order to reduce their RAM/VRAM usage because of how little memory was available of previous gen consoles & how weak their GPUs were.

You are giving too much weight to screen resolution itself.

One other example : once SSDs will get cheap and widespread enough to be included in games minimum requirements. (propably in few years from now with next Xbox/PS5) then you will see another VRAM requirement bump, regardless of what resolution will be 'standard' then.

 

[Fjodor2001]

Just to summarize, from what I can tell so far nobody here has been able to present an alternative hardware spec that would be better than the AMD APU used in the PS4/XBONE, assuming the TDP and price of those units should be the same as currently. Is that correct?

 

[stuff_me_good]

The diesizes are nowhere near 550mm2. They are 348mm2 and 363mm2.

Like you said, the core on PS4 was nowhere near 550mm2. Personally I was disappointed that sony went so conservative route this time. They could have easily order just slightly bigger chip which would include 7870 equivalent GPU making the chip no bigger than 400mm2.

From the cooling side the enclosure was well designed. But they could have just slightly bump the size of the enclosure(which was already so small compared to xbox) making it able to cool down the 400mm2 chip. Also some CPU overclocking could have been possible due to bigger enclosure and there fore better cooling capacity.


Given the time and cost constraints, no point arguing what they should have done, because it's no brainer that they already made right choices. If only sony would have made the reasonable and well doable choice that I presented.

 

[ShintaiDK]

Something quite subtle that people may be forgetting is that the xbone/PS4 APU is NOT a true 8 core chip, rather it is two quad core jaguar clusters on die connected by a bus. While L2 access on the same module is a minimum 26 cycles it increases to a whopping ~190 cycles when accessing cache on the other module (RAM latency is ~220 cycles).

This is a terrible barrier to effectively multithreading game code; for performance purposes code needed both by the two 'odd duck' cores and the four game cores must pretty much be present in both caches for prompt execution. Thus it seems quite likely that two of the 6 console cores are going to be used for relatively remote tasks that do not depend much on the main game code.

Not surprising really, considering they had to bolt them together. And it puts another nail in the coffin. 4 threads (or less) will be the sweetspot for a long time.

 

[seitur]

Just to summarize, from what I can tell so far nobody here has been able to present an alternative hardware spec that would be better than the AMD APU used in the PS4/XBONE, assuming the TDP and price of those units should be the same as currently. Is that correct?

Not just BOM price, but also R&D, but overally answer is: Yes.

Diffrent specs would need diffrent release time and/or price and/or R&D expense and/or TDP, etc

 

[Nachtmaer]

Back when the rumors and leaks were going full throttle before the new consoles were released, I saw this Japanese document talking about Sony's hardware choice. I don't know whether it was fake or not, but it mentioned both Jaguar and Steamroller, so perhaps they were actually considering SR at some point. I guess they opted out of that for the reasons mentioned in this thread; die size, TDP, reserving cores for background tasks (and possibly SR not being ready at the time). Maybe still being stuck on 28nm played a role in that as well.

It's hard to come up with a what if situation as I am not a chip designer or an economist, but it would have been interesting if they decided to go with two or three PD/SR modules instead.

Overall I think they did a good job considering the price in this age of <300W high end GPUs. I don't even want to imagine how things would turn out if they crammed >200W hardware into a small box that gets locked away somewhere.

 

[Erenhardt]

Like you said, the core on PS4 was nowhere near 550mm2. Personally I was disappointed that sony went so conservative route this time. They could have easily order just slightly bigger chip which would include 7870 equivalent GPU making the chip no bigger than 400mm2.

But it is 7870. There are 20 CUs on ps4 die, but only 18 are enabled (redundancy).

http://www.anandtech.com/show/6972/xbox-one-hardware-compared-to-playstation-4/2
The clocks are lower obviously. Whole PS4 takes as much as 7870 alone, so they had to keep clocks lower to drop voltage a bit. Not much as they need to make sure there are as much usable chips as possible - there is no cut down version of ps4, so no die harvesting i possible.

 

[Rezist]

The only real option was ARM but beyond phones I'm not sure what they have in the high performance segment, most of there stuff is sub 5W. They needed 15-20W core(s) with a 80W GPU component SoC, I'm not sure even with A57 if they could come up with similar or better performance given the extra TDP.

 

[monstercameron]

The only real option was ARM but beyond phones I'm not sure what they have in the high performance segment, most of there stuff is sub 5W. They needed 15-20W core(s) with a 80W GPU component SoC, I'm not sure even with A57 if they could come up with similar or better performance given the extra TDP.

If a57 can clock as high as a15 then it would be better than jaguar and maybe more efficient. A57 does have higher ipc than jaguar...think project skybridge.

 

[Fjodor2001]

ARM A57 was not available in time for PS4/XBONE anyway.

 

[frozentundra123456]

If they had gone ARM, would the programming have been easier or more difficult? I think porting to PC would have been more difficult, but I don't think that would be a major factor to the devs.

 

[soccerballtux]

Thing is FX series is 32nm and GCN is 28nm.


2 GB is nowhere near enough. If you wanna employ certain techs and get rid of heavy compromises that were made in times of previous gen multiplaform AAA games then you need much more than that.

2gb is more than enough for 1080p gaming. The bloated memory demand of newer games is somewhat due to current gem console unified memory.

this x1000

2gb would be plenty of dedicated VRAM in a console aiming @ 1080p

wow. so none of you guys know what a framebuffer is. LOL

 

[Ventanni]

Honestly for the price, I think the hardware in this generation of consoles is about as perfect as you can get. I's not design constrained, it's not proprietary, it uses off the shelf parts, and both MS/Sony are making money right out the door this time around. Still, I prefer the hardware in the PS4 over the XBone, but they're both solid.

 

[Headfoot]

From the "AMD Claims It Will Power Another Gaming Device" Thread.

The problem is you think you know more than Sony and Microsoft combined.

 

[Techhog]

TL;DR Everyone wanted them to be high-end PCs for $400 in a small box.

Instead of an 8 core, I'd have rather had a faster Quadcore (or a Big/Little combination with a Quadcore x86+ARM Processor for OS or something like that).
Both using GDDR5 ram.
Stronger GPUs (equivalent to the Xbox360/Available GPU at the time of purchase of that console).
Better Kinect Support (Seriously this wasn't a bad featuer of hte Xbox One. It was poorly handled. On Madden, I should be able to make play calls with the Kinect. In NBA 2K, I should be able to call for a pick, on army games I should be able to tell the AI to flank, etc. the Kinect had a TON of potential and M$ just screwed it up. It should have been integrated into a TON of titles from development phase but instead it was tossed in and half baked and then FORCED onto people).

Better SmartGlass/Smartphone/Tablet support.
This was promised with Halo4 and was TERRIBLE. You were promised more advanced stats on your phone at the end of matches and it never worked well along with the other features they promised. I was very upset and it rarely worked as intended.

In the Xbox's case, I wish they had used the Sony Touchpad on their controller. Simply because when it made it's way over to PC, that controller would be a game changer. Even now the DS4 is a great controller for PC but it's not as reliable as the Xbox controller still.

That would have been either the 7970 or 290X... Yeah, I don't think I don't think that would have happened.

 

[ElFenix]

An Intel i3 + custom Iris Pro variant should have been the choice. And before you mention price, Intel is flipping Atom's for nothing, I'd bet they would be flexible in price given they have put so much work into their iGPUs.

intel wasn't flexible with the original xbox so burned that bridge. intel wasn't going to happen because MS wasn't going back to them, and sony saw what intel had done so sony wasn't going to them either.