[hexus.net]AMD claims it will power another gaming device

[AnandThenMan]

lol you make it sound like nvidia designed maxwell the day it got released. I am assuming that nvidia had maxwell in the works in a similar time frame coinciding with amd's work on their semi custom apu.

So what you're saying is Tegra should have been in the current consoles. Because last time I checked Nvidia only makes discrete GPUs outside of Tegra.

 

[monstercameron]

So what you're saying is Tegra should have been in the current consoles. Because last time I checked Nvidia only makes discrete GPUs outside of Tegra.

A little lost on this comment as I didn't bring up Tegra but it is also possible that a scaled up Tegra could suffice. Assuming arm a15, that should be close to jaguar in terms of ipc and we know it clocks upto 2ghz in a tablet form factor. We also know that the arm bus can support quite a few of these cores. Assuming a57 then that will easily out perform jaguar ipc wise, don't know how well it behaves in high clocks though. Tack those cores into a maxwell gpu and you got a pretty decent chip.

 

[AnandThenMan]

A little lost on this comment as I didn't bring up Tegra...

You brought up Nvidia, so I'm thinking of a possible way Nvidia could supply silicon to Sony and MS. That would be Tegra, which in theory could work if substantial development went into such a device. But no such development took place, plus Nvidia flat out said they were not interested in any console contracts because there was no money it in.

That could very well have been true for Nvidia, but for AMD the contracts are providing a nice source of revenue.

 

[monstercameron]

You brought up Nvidia, so I'm thinking of a possible way Nvidia could supply silicon to Sony and MS. That would be Tegra, which in theory could work if substantial development went into such a device. But no such development took place, plus Nvidia flat out said they were not interested in any console contracts because there was no money it in.



That could very well have been true for Nvidia, but for AMD the contracts are providing a nice source of revenue.

True, another point to make it that, it would have made much more sense for nvidia to attempt this than Intel. In any case the only other third party could've been imgtec. Just imagine their rogue architecture with a bunch of competent mips proaptiv cores. /wishful thinking. Gosh darned it, I want rt path/Ray tracing!

 

[positivedoppler]

Apparently I was wrong. It is the new Nintendo Console
http://wccftech.com/nintendo-confirms-console-powered-amd/

 

[monstercameron]

Apparently I was wrong. It is the new Nintendo Console
http://wccftech.com/nintendo-confirms-console-powered-amd/

Hopefully they bring back the gameboy brand.

 

[TurtleCrusher]

Apparently I was wrong. It is the new Nintendo Console
http://wccftech.com/nintendo-confirms-console-powered-amd/

Pretty exciting. The Wii U might be their relatively first home console flop. Hopefully they can release something 1.5x the power of the PS4 and they'll take the crown back. I would be able to justify with the wife on spending $400 on a home console while having a gaming PC, only because it'd have game franchises she's familiar with. If it had more oopmh than the current consoles then it'd be a no brainer.

The wife loves Donkey Kong and Mario Kart. I love sports and strategy games. If Nintendo can get hardware that PS4/XB1 games can be readily ported to the new console with power to spare, it'll be a winner.

 

[mrmt]

To match the AMD APU there would have to be a PowerPC/Nvidia hybrid with a shared pool or memory among other things. Custom chip designed from the ground up. Not feasible in fact the suggestion is fairly ridiculous. It took AMD several generations to get the APU to the state it's in now.

Why is the shared pool a mandatory item? And why Nvidia or Intel would have to *match* AMD APU levels of performance, why not go beyond it, which you can with a PC-like solution, with the trade off of having higher costs? You are trying to transform a business restriction (the consoles must reach a certain X level of performance within a Y budget) in a technical restriction (Nobody but AMD could reach a given level of performance, whatever this level of performance is).

The console APUs are pretty much mediocre chips from a performance POV. 8 weak jaguar cores, the kind of power someone would use for a NAS, and a mainstream 2012 GPU. It's not breathtaking by any standards and are a smaller performance leap when compared to the last generation. Both consoles are cheaper to build and sell than their previous generation counterparts were at similar points of their shelf lives. If AMD weren't in such a bad financial situation it would be a footnote on their financial statements, and on top of all that there are plenty of interesting things happening on the server front and mobile front, far more interesting than these mediocre chips. but who cares? Both Sony and Microsoft picked these chips and are building profitable business around these machines, and that's what matters.

 

[mrmt]

Apparently I was wrong. It is the new Nintendo Console
http://wccftech.com/nintendo-confirms-console-powered-amd/

The volumes AMD reported are far lower than the volumes that Sony and MSFT got. If this mysterious customer is really Nintendo, either are designing something in an experimental form factor or they are projecting sales to be very low.

 

[positivedoppler]

The volumes AMD reported are far lower than the volumes that Sony and MSFT got. If this mysterious customer is really Nintendo, either are designing something in an experimental form factor or they are projecting sales to be very low.

That would be pretty darn good news for AMD. The 1 billion profit they anticipate is probably for 1.5 million volume per year. If it is Nintendo, 1.5 million is way low ball. Just being the first console that can hit 1080P or maybe even 4k will give Nintendo a HUGE advantage.

 

[mrmt]

That would be pretty darn good news for AMD. The 1 billion profit they anticipate is probably for 1.5 million volume per year. If it is Nintendo, 1.5 million is way low ball. Just being the first console that can hit 1080P or maybe even 4k will give Nintendo a HUGE advantage.

It's not 1 billion profits, it's 1 billion in revenues, for two semi-custom products, and on a 3 years life cycle, that's a paltry 80 million per quarter, for *two* designs. That's way, way low for a mass market console.

 

[turtile]

That would be pretty darn good news for AMD. The 1 billion profit they anticipate is probably for 1.5 million volume per year. If it is Nintendo, 1.5 million is way low ball. Just being the first console that can hit 1080P or maybe even 4k will give Nintendo a HUGE advantage.

There are also rumors going around that Nintendo wants to make a low cost console for the developing market. It could be that the chips are just really small and therefore, don't make much money.

 

[positivedoppler]

It's not 1 billion profits, it's 1 billion in revenues, for two semi-custom products, and on a 3 years life cycle, that's a paltry 80 million per quarter, for *two* designs. That's way, way low for a mass market console.

Here's my math, and I think it's a good estimate

1 billion over 3 years is 333 million/year.
Assumption is both contract are about equal so that's 167million/year. They are selling Xbox one chips to MS for about $115 and $100 for PS4. So given 167 million/year that works out to be maybe selling 1.5 million units/year. That is too low for a console. But if the report is right and it is the next Nintendo console, then the assumption of 1 billion revenue over 3 years is based on 1.million units being sold is way too low.

 

[positivedoppler]

There are also rumors going around that Nintendo wants to make a low cost console for the developing market. It could be that the chips are just really small and therefore, don't make much money.

You know, that's a good theory. I was assuming it was going to be on 14nm node but maybe it will still be el cheapo 28nm meant for a 3rd world market.

 

[mrmt]

Assumption is both contract are about equal so that's 167million/year. They are selling Xbox one chips to MS for about $115 and $100 for PS4. So given 167 million/year that works out to be maybe selling 1.5 million units/year. That is too low for a console. But if the report is right and it is the next Nintendo console, then the assumption of 1 billion revenue over 3 years is based on 1.million units being sold is way too low.

If the customer is Nintendo, either AMD reported the take-or-pay commitment of the contract or the new console is a cheap mobile console. I think that the second hypothesis is far more probable than the first one.

 

[turtile]

You know, that's a good theory. I was assuming it was going to be on 14nm node but maybe it will still be el cheapo 28nm meant for a 3rd world market.

It all depends on when they release it. If it's a Nolan or Amur base, it will be 20nm for 2015. Who knows if GF will have 14nm ready on time! I don't think AMD will be using both TSMC and GF for the same architecture if there is no need. I don't know if it would be difficult, but they'd have port 14nm/16nm to do so.

 

[Madpacket]

I know this won't be a popular opinion but I think these low power Jaguar cores is the best thing to happen to the game industry in a long while.

By having many weak cores it absolutely forces engine developers to built proper multi threaded games which in the end will benefit the industry as a whole.

For much too long we have game engines that rely on 1 or 2 fast cores to get the job done so these low speed jaguars are great in my opinion.

 

[NostaSeronx]

-Off-topic-

It all depends on when they release it. If it's a Nolan or Amur base, it will be 20nm for 2015.

Amur is a gaming smartphone/gaming tablet SoC. Nolan is the what is replacing Mullins/Carrizo-L is what is replacing Beema.

The AMD Amur SoC development team was mostly acquired from the ST-E NovaThor development team who did L8540/L8580.

Amur = 28nm SLP, could be ported to 28nm FDSOI. (GF28A/Advanced)
Nolan = 20nm LPM, could be ported to 14nm FDSOI. (GF20AN/Advanced Next)
Carrizo-L = 28nm SLP, could be ported to 28nm FDSOI. (GF28A/Advanced)

● Designed and implemented Camera HAL v2.0 & v3.0 from scratch, for Amur APU.
● Developed Android Camera Hardware Abstraction Layer (HAL) for STE 85XX SOC.

For the nodes AMD will use look for; http://i.imgur.com/v0Sxx5E.png
(Higher is better//DDC process ~ FDSOI process in height.)

 

[NTMBK]

Apparently I was wrong. It is the new Nintendo Console
http://wccftech.com/nintendo-confirms-console-powered-amd/

From that quote it sounds like they are still in the early planning stages. This probably isn't what AMD was talking about.

 

[MisterLilBig]

Its quite meaningless isnt it. The Cell in theory can provide 230Gflops. Yet its much slower. My Haswell is also 200Gflops or so. Yet its not twice as fast as a 3570K.

For flops it should be, no? I did point out it was a very simplistic data point of showing change, improvement and so on.

Why would you need 2 cores for OS/background? I am sure you could settle for 1. If it was even needed at all.

For all the processes that are running in the background, faster switching, services, etc...some of these are on dedicated processor units also.

But I find amusing the notion that AMD was the only company able to provide a solution for the console OEMs. IBM and Nvidia already had expertise in the area, and there was nothing preventing Intel from doing so, and these companies could devote far more resources to the venture than AMD. AMD's main advantage is that nobody with the resources to develop the console chips were willing to go as low as them in terms of price (mid-teens margins) and was in so much need of cash as them (taking upfront cash from the OEMs, further eroding the margins). AMD simply offered the best "bang for the buck" for the console OEMs, and given that the consoles were fundamentally cost-constrained this time, this is *the* selling point, regardless of all the technical limitations anyone can see on the solution.

AMD was the only company that meet all of the criteria, and the best choice to be picked for PC gamers in the end. The only reason they didn't went ARM was because it had no 64-bit support. Going ARM also has a great business case, greater than x86, for the market, that's for sure. But, at the moment, x86 was a better choice.

Do you think you can make a game like on the level of the planned SC complexity to run on 6 weak cores? you just cant, a minimum of 2 are gona be used for physics and another 2 for API rendering, that leaves you with only 2 for game logic, AI and audio, not good, time to use the chainsaw. And im not even sure if 2 are gona be enoght for rendering.

Is SC, StarCraft...? If so, my answer is,Yes. There used to be a time where you would separate a system and put it on a whole core, that should not be done anymore, now parallel as much as you can and go from there.

Just the fact that a tablet cpu could get probably about 50 to 60% of the cpu power in a "gaming device" is stunning. And the ones with the Z3770 and Z3775 will get even closer.

So? Nobody will ever push those things to their max capacity as developers will a console.

I've seen repeated many times that the current gen of consoles brings nothing new or innovative to the table. Then there is the citation of other consoles that did indeed contain some breakthroughs in graphics. The above is true, but it is more to do with timing than anything else.

Agreed. There is nothing new. These are evolutions of old things or old things finally in consoles. One could say that Microsoft was the one bringing the most innovation and it failed, because of people.

What if the CPU performance targets of the console chips were higher than AMD could provide with the cat cores, what technology would they have deployed for the consoles?

More cores. It is a valid argument. We have supercomputers made of many cores, not few super fast cores. "More cores" is a very acceptable proposition.

But change this set of constraints by moving them close to the bleeding edge (and consequently pay more for the console chips) and Intel and Nvidia might have provided a better value proposition.

Bleeding edge what? For what? As a single company solution, Intel has nothing good for a console, even the i7-4770R is totally out performed. And NVIDIA only has Tegra, so the Tegra TK1 64-Bit Denver @ 2.5GHz, which at less than 365 Gflops per SoC(!), you would need 5 of them to match the Sony/MS consoles.
Where is the value proposition? And if going with an ARM solution, why even go with Nvidia? Generic ARM IP would of been best for many, many reasons. And if going to a custom ARM solution, why go with Nvidia? Qualcomm has millions of its solutions out there already, that would of been a wiser choice. And then, there's Img on all those Apple devices.

A little lost on this comment as I didn't bring up Tegra but it is also possible that a scaled up Tegra could suffice.

It would had, but 64 bit support was a requirement.

Apparently I was wrong. It is the new Nintendo Console http://wccftech.com/nintendo-confirm...e-powered-amd/

There is no mention of it being from any company.

Why is the shared pool a mandatory item? And why Nvidia or Intel would have to *match* AMD APU levels of performance, why not go beyond it, which you can with a PC-like solution, with the trade off of having higher costs? You are trying to transform a business restriction (the consoles must reach a certain X level of performance within a Y budget) in a technical restriction (Nobody but AMD could reach a given level of performance, whatever this level of performance is).

It has been said many times that shared memory is the future, everyone is on that boat now, in some form or another. It was a requirement.
If I ask you to build me the best thing at X price and Z features and you turn out with AMD stuff, well that's it. But you show me Y, at a slightly higher price, then I will tell you..."Don't waste my time and money on something I did not ask for."

I have given examples of why the console hardware was the best choice, people have only complained.

Is there better hardware out there? Yes.
Could a better solution at a higher price be made? Yes.
Who picked AMD? Microsoft and Sony.
Why? They got what they wanted.

 

[AnandThenMan]

Why is the shared pool a mandatory item?

It's not. But I'll let you think about why it's a better solution.

And why Nvidia or Intel would have to *match* AMD APU levels of performance, why not go beyond it, which you can with a PC-like solution, with the trade off of having higher costs?

You will have to ask Sony, Microsoft and Nintendo why they are not pursuing the higher priced, more PC like approach. The STEAM Box will be more along those lines, if/when it gets released.

You are trying to transform a business restriction (the consoles must reach a certain X level of performance within a Y budget) in a technical restriction (Nobody but AMD could reach a given level of performance, whatever this level of performance is).

The restriction has nothing to do with me, it has to do with what Sony/MS/Nintendo want, what price point they want to hit, and what performance they can get at said price point. AMD was able to supply a solution no one else could. Very simple.

The console APUs are pretty much mediocre chips from a performance POV. 8 weak jaguar cores, the kind of power someone would use for a NAS, and a mainstream 2012 GPU. It's not breathtaking by any standards....

You want breathtaking performance at $400? Please point me in the direction of such a system I'll buy it!

 

[III-V]

More cores. It is a valid argument. We have supercomputers made of many cores, not few super fast cores. "More cores" is a very acceptable proposition.

No, it's not valid. Supercomputers compute problems that are easily parallelizable. Game code is largely serial, not parallel, in nature.

 

[AnandThenMan]

So then all processors should be a single, extremely powerful core, right?

 

[III-V]

So then all processors should be a single, extremely powerful core, right?

Yes, supercomputers should use the worst possible processors for their tasks. Brilliant suggestion.

 

[AnandThenMan]

So it's just a gaming processor that should be a very powerful single core. Got it.