By using this site, you agree to our Privacy Policy and our Terms of Use. Close
elektranine on 03 April 2016
Edit Thread
Soundwave said:
elektranine said:

That is just an arcticle based on a routine email for those on AMD's mailing list. I've already said that SCE went to AMD for a reference design but they took that reference design and used that to create a new custom APU chip orders of magnitude more powerful and supporting new technologies not found even in AMD's current APU chips.

How does the PS4 APU become 1200% (12x) more powerful than even most high end APUs released in 2013?

How is it that even AMD's newest announced APUs (not even released yet) are still outclassed by the PS4 by over 2x?

All this extra power & technologies (such as HSA, hUMA) are not found in AMD's retail lineup because AMD does not own the rights to manufacture them. SCE paid alot of money to license the reference design & then customize it. Nintendo would have to do the same. But they lack SCE's engineering background. SCE created a whole new architecture, the cell. Nintendo has never done anything quite to that magnitude so they would have to rely on AMD which would largely increase the R&D costs. So spending a few hundred million $ in a single year doesn't really support this happening. More likely Nintendo is taking an AMD retail APU increasing things like clock speed, memory, etc and ending up with something about as powerful as the xbone.

The GameCube was an amazing chip that outperformed Ken Kutaragi's ballyhooed PS2 for $100 less. The CELL was a design disaster which got Ken Kutaragi fired.

The Gamecube came out over 19 months after the PS2 launched. Besides that the PS2 was never designed to have high specs in the first place as it was SCE's first attempt at semi-custom chip design with another company. So its not surprising that the Gamecube is more powerfull in theory as it came out almost 2 years later and Nintendo went for high end specs. But that is only in theory as the PS2 had much higher raw polygon performance (over 2x) and was overall better recieved by developers.

The cell was a disaster? Ah yes, the cell was a complete disaster for SCE I remember now. It was only used in billions of devices, supported by dozens of multi-national multi-billion dollar corporations, used in everything from TV's to consoles to Set top boxes to Blu ray/HD-DVD players, heavily used in research & academia, even used by the military and some super computers. The Cell was SCE's first attempt at co-developing a fully custom computing architecture with other corportations and they learned very important insight & experience they used in the future. It is not really true to say the CELL was a disaster as it was heavily supported and used in many devices. The CELL the was world's first iteration of a non-homogenious computing chip and is the direct forerunner to modern day APUs as seen in the PS4 & xbone. Without the CELL APUs would not exist as many thought that a non-homogenious chip could not be commercially viable. It is also false to say Ken was fired as that never happened. Many like to trump up this lie to say SCE failed but it is only that a lie. It is important to understand japanese culture as with important projects its not uncommon for working men to not take days off or see their family for extended periods of time, sometimes even many years. Ken merely took some extended time off to see his family and deal with family issues. During that time he was & is still employed as a consultant for SCE and in fact was even one of the engineers for the PS4.

AMD does most of the work here really, they make GPUs, if Sony/Nintendo were so good at it they would make their own, but they don't and are well advised to not. 

Actually no AMD does not "make GPUs". You are wrong AMD has not made any GPUs or CPUs for a long time. They are a fab-less chip design & consulting firm now. Why would SCE make their own GPU design that would take decades of R&D not to mention patents & licensing issues. SCE went to AMD and licensed a reference design for their APUs and then substantially modified it with new tech. (This is actually really common in the computing industry. ARM licenses their ARM cpu reference design to the dozens of ARM cpu companies that then heavily customize & manufacture the CPUs for their own applications. ARM doesn't have real intimate knowledge and cannot legally disclose what makes Qualcomm's ARM cpus better to another company, Apple for instance)

The R&D cost of the chip is not really so much the issue. The R&D cost of SOFTWARE development is more the issue nowadays, sure you can have a 3 TFLOP machine. 4 TFLOP. Fuck. Make it 5 TFLOP. 

This shows you have no clear understanding of the computing industry and don't really know. R&D costs will typically outweigh all the manufacturing & marketing costs for a chips useful lifetime. Say you have 1000 highly qualified engineers paying each engineer a modest salary of 150K, thats 150 million a year just for their base salaries. So say a chip takes just 6 years to R&D you are already at nearly $1 billion. Add in the cost of benifits (healthcare, vacation, pension, bonuses, etc) and all the il-liquid material costs (real estate, equipment, prototyping, etc) and any srewups or failures along the way and you will quickly see R&D go up to the billions. Its not uncommon to see a single chip cost $3 billion or more to develop. And can't "make" a chip be XX terraflops as terraflops is a human measure of how powerful a chip is. You can't just throw in more processors as that will increase the chip size which will exponentially decrease the yield rate (%). And there are other things to worry about too such as TDP, watts, bandwidth, etc. (Large firms like Intel have over 10,000 engineers)

The real question is who the hell is going to develop for that and at point does pushing such a chip to its limits basically mean that a developer has to gamble with its future existence? 

So it's this. It's a certain hardware manufacterer telling developers no you can't develop games on more powerfull hardware "we know better". Since when will a dev go bankrupt by developing on more powerful hardware? SCE certainly hasn't and with revenues and profits higher than ever.

The PS4 and XB1 GPUs from my understanding aren't even that different, MS was so butthurt from the RROD on the 360 that they played it probably too safe. It's not like Sony had some magic powder in their chip that only they could create. 

Well your understanding is wrong. Yes both MS and SCE went to AMD and licensed reference designs but thats is where things start to diverge. MS left much of the reference design in place chose only a modest bump in core specs in favor of the whole ESRAM setup. SCE chose to look at where the computing industry will be going in the next few years and add those technologies in such as HSA and hUMA support among others. And not going the esram route allowed SCE to really ramp up the core specs. Fun fact: the SCE APU has smaller chip surface area and had higher yield rates resulting in lower costs than the MS APU while also simultaneously more powerful and having higher bus speeds than the MS APU. I never claimed SCE had some sort of magic power but instead spent many years and billions on R&D. This is also something that Nintendo will also have to do but even more as they lack SCE's engineering experinece with the CELL.