Digital Foundry: next-gen PlayStation and Xbox to use AMD's 8-core CPU and Radeon HD

[Fx1]

On the OP linked article....

i have the original PS3 and its worked fine. Desktop Class GPU and all.

A 28nm modern desktop GPU would work easily but the consoles are known for cutting corners in costs.

1.6ghz CPU thats a laugh AMD 1.6ghz too. like a 1ghz intel CPU.

Console graphics suck and they have always sucked. Can barely handle 720p today and they might just about handle 1080p in next gen.

But in 2 years time it will be dead in the water again as newer games cant be handled on a poxy 1.6ghz CPU and mobile GPU

 

[magomago]

WOW and SC2 in a console.... really?



my dad is using a celeron 420 right now, single core and 1.6ghz...and it runs half life 1 just fine...

i know, i know...old game...but this 8-core ship is like 9-10 time faster than my dad's computer, and i really think that with a decent gpu it can run HL2 (just igp at the moment)

I don't think you understood the point at all. I'm talking about CPU intensive games...I'm not talking about SC2 or WOW specifically; those were used as an analogue as a reference of CPU bound games on the PC. I'm actually not referencing any more not because there are none, but because I don't keep up in the same way, and my PC has been able to play everything on high details for me that I don't bother trying to figure out what/where my bottleneck is.

Your reference to "a decent gpu to run hl2" itself is making a point: the game came out over EIGHT years ago.

 

[Ferzerp]

10? Minimum here is two (enough for an emoticon).

 

[ViRGE]

This is going to be the only warning I'm going to give about anti-social behavior. Accuse another poster of being a fanboy, thread crap, or otherwise make a menace of yourself and I'll delete your posts and throw you out of this thread.

-ViRGE

 

[Ferzerp]

Anyway, slightly on topic with some of the comments here: How exactly did PC gaming benefit from sharing CPU architecture with the Xbox? That's right, it did not.

 

[itsmydamnation]

9-10 times faster?

More like 6x with near perfect scaling. Just compare the dualcore Brazos with the singlecore Celeron 420. But getting near perfect scaling and gaming. Thats gonna be more than tricky.

Thing to remember, Xenons is a very well understood beast and lots of time is spent optimizing code for it. How many people spent time optimizing code for bobcat. Expect game developers to get much higher then talked about 1.1 IPC from "general workloads" on Jaguar. Given that Xenons gets you 0.2 IPC per thread on optimised code, 9-10 times is certainly possible.

Anyway, slightly on topic with some of the comments here: How exactly did PC gaming benefit from sharing CPU architecture with the Xbox? That's right, it did not.

thats a very poor straw man, how many xbox's vs PS2 where sold. How many non 1st party games was the xbox the primary development platform. Microsoft had an interest in limiting PC version of Xbox games, see Halo for example.

 

[Fx1]

You're right, they could do it. It's been done before.

I don't think the issue here is heat* though, it's pricing. The PS3 sold at a loss for many years before it became profitable, so I'd wager that it's Sony (and Microsoft's) intent to sell these consoles at a profit from the start. An 8-core Jaguar design is going to be small, cheap, and powerful for the price/size.

*Beefier power supplies and heatsinks required for high TDP silicon also add to cost.

Yes and its a shame.

But other costs are cheaper now. HDD, Ram and Blu rays. 28nm GPU and CPU vs a 90nm GPU/CPU

Its just a profit grab or maybe they expect this generation only to live 3-4 years rather than the awful length this generation has made me endure

 

[parvadomus]

Amazing, I hope this is true, it will make developers to multithread their games as much as possible

 

[ShintaiDK]

Thing to remember, Xenons is a very well understood beast and lots of time is spent optimizing code for it. How many people spent time optimizing code for bobcat. Expect game developers to get much higher then talked about 1.1 IPC from "general workloads" on Jaguar. Given that Xenons gets you 0.2 IPC per thread on optimised code, 9-10 times is certainly possible.



thats a very poor straw man, how many xbox's vs PS2 where sold. How many non 1st party games was the xbox the primary development platform. Microsoft had an interest in limiting PC version of Xbox games, see Halo for example.

So you say code for x86 will be more optimized than code already is today for x86?

My point is 1.6Ghz Jaguar would be turtle slow.

9-10x is certainly not possible. Remember the compare is a Celeron 420.

As I say, try run your desktop at 1.6Ghz and see how it goes. And then your desktop might even still wastly outperform Jaguar.

 

[Greenlepricon]

Yes and its a shame.

But other costs are cheaper now. HDD, Ram and Blu rays. 28nm GPU and CPU vs a 90nm GPU/CPU

Its just a profit grab or maybe they expect this generation only to live 3-4 years rather than the awful length this generation has made me endure

Thank goodness blu rays are so much cheaper now. The cost is definitely one incentive for the companies to make "weaker" systems. I don't want a $300 Haswell and a $400 7970 in my console. That's what my computer is for.

Anyway while this was a really long generation for consoles, I don't think it was that bad. I'm excited to see what new developments come out of next generation though, especially if a beefy gpu is used and a 8 core cpu offers multithreading. This is definitely giving me something to daydream about.

 

[Ventanni]

Yes and its a shame.

But other costs are cheaper now. HDD, Ram and Blu rays. 28nm GPU and CPU vs a 90nm GPU/CPU

Its just a profit grab or maybe they expect this generation only to live 3-4 years rather than the awful length this generation has made me endure

I wouldn't necessarily call it a shame. You want console makers to make a profit. I mean, this hardware may be cheap, but it's definitely capable.

 

[itsmydamnation]

Yes and its a shame.

But other costs are cheaper now. HDD, Ram and Blu rays. 28nm GPU and CPU vs a 90nm GPU/CPU

Its just a profit grab or maybe they expect this generation only to live 3-4 years rather than the awful length this generation has made me endure

actually i think your quite wrong
1. Ram is going to be way more expensive because DDR3/4 are just to slow so ether GDRR5 on a wide bus (256bit) some form of embedded ram or stacked 2.5/3d/imposer solution will be needed.

2. HDD from a relative perspective will be about the same cost, 100gb bluray games vs 8 gb xbox games for example.

3. what does 90nm vs 28nm have to do with anything? if anything cost per mm or cost per watt ( likely limitation) is higher @ 28nm then 90nm.

4. this generation hasn't really been mach longer then any of the previous and the last generation, particular the PS2 was selling very well even after both PS3/360 had been out for a significant amount of time so there was a big crossing over period.

 

[ShintaiDK]

Amazing, I hope this is true, it will make developers to multithread their games as much as possible

The question is at what scaling ratio.

 

[itsmydamnation]

So you say code for x86 will be more optimized than code already is today for x86?

My point is 1.6Ghz Jaguar would be turtle slow.

9-10x is certainly not possible. Remember the compare is a Celeron 420.

Yes, your optimizing for a processor, its particulars, how the oooE works, L/S works, how it decodes, what kind of throughput you get with different operations etc .

If jaguar is turlet slow then Xenon was sloth/koala slow when it was released ( both sleep 20 hours a day)

 

[ChronoReverse]

So you say code for x86 will be more optimized than code already is today for x86?

My point is 1.6Ghz Jaguar would be turtle slow.

This isn't hard to comprehend you know.

For example, we know hand optimized assembler for x86 is superbly fast while even C code isn't even close.

This is why ZSNES was able to run the majority of SNES games at full speed on really old x86 processors while all the ones written in C or C++ require, *gasp*, 10x faster CPUs to achieve the same thing.

So x86 code can be more optimized than x86 code.

 

[VirtualLarry]

I wouldn't be so sure. We're talking about a console after all. There's something to be said for having known, deterministic performance, as opposed to the uncertainly turbo would introduce.

I was wondering the same thing. I recall reading about one game for the 3DO, that wouldn't work on the newer "slim" 3DO console, because the original code in the game was tuned for the exact performance of the original disc drive, and the newer version of the console had a different disc drive. I think the game was one of those video-streaming choose-your-path games. Don't recall if it was Dragon's Lair, or a different one. That was a popular genre on the 3DO.

 

[Ventanni]

So you say code for x86 will be more optimized than code already is today for x86?

Nah, what he's saying is that in a console environment, it's much easier to code at a low level given the fact that all consoles are the same. The beauty of a PC environment is that it allows for an enormous magnitude of products and devices to interface with the operating system, allowing for hardware developers to write their drivers for the API it interfaces with, not specifically for the hardware itself. This flexibility and convenience comes at a high performance price though, as those API calls must be interpreted by the drivers of the hardware, then send to the hardware itself.

In a console environment, a lot of that simply isn't necessary, and games can make calls directly to the hardware. This can be done on a PC as well, but would require software developers to code for all hardware combinations imaginable. It's just easier to code for the API.

 

[ChronoReverse]

Turbo is fairly deterministic though. Because console coders know about the loads their code needs and uses, I'd imagine they'd profile to see whether a load that triggers Turbo is more effective than one that doesn't.

 

[itsmydamnation]

This isn't hard to comprehend you know.

For example, we know hand optimized assembler for x86 is superbly fast while even C code isn't even close.

This is why ZSNES was able to run the majority of SNES games at full speed on really old x86 processors while all the ones written in C or C++ require, *gasp*, 10x faster CPUs to achieve the same thing.

So x86 code can be more optimized than x86 code.

he is also ignoring all the API overheads that aren't there in a console. Draw calls being the perfect example.

 

[Hubb1e]

People keep saying a 1.6 ghz Jaguar will suck. Dude, you've got 8 of them. Add them up and that's like a 12.8ghz Jaguar. Yeah I understand that it won't scale like that but you're forgetting that Brazos was only 2 cores with 25% less IPC and can still run some PC games. Since the console will be able to optimize around 8 threads you're still looking at some decent performance. I was expecting a 2 module Piledriver but Jaguar certainly makes some sense considering you've got Kinect and other peripherals that like threads.

The old PowerPC core in the 360 had terrible IPC. Worse IPC than a P4 at the time so you can't look at clock speed to compare them. It seems to me like an 8 core Jaguar will be plenty fast enough for gaming.

The GPU is a lot more than I was expecting. I thought it would be closer to 6670 performance, 7770 tops, and this GPU is a lower clocked 7850 which is an extremely respectable GPU even on the desktop today and also remember that 720p looks perfectly fine on a TV so you're shooting for a lower resolution than many desktops were you're sitting close to your screen. There's no way a 7970 would fit under the hood of a console. That thing pulls 250W on its own. GPUs have gotten MUCH more power hungry than they used to be in the 7800GT era.

I don't play consoles but this thing sounds like it will make for a very nice upgrade over the 360/PS3.

 

[ShintaiDK]

Turbo is fairly deterministic though. Because console coders know about the loads their code needs and uses, I'd imagine they'd profile to see whether a load that triggers Turbo is more effective than one that doesn't.

Turbo also depends on temperature. Something you cant predict at the customer end.

 

[itsmydamnation]

Turbo also depends on temperature. Something you cant predict at the customer end.

I dont think there would be a turbo for consoles, the goal is always to extract maximum performance, if something is idle then your doing it wrong, so to speak. From this point of view you should never hit turbo conditions.

 

[VirtualLarry]

Turbo is fairly deterministic though. Because console coders know about the loads their code needs and uses, I'd imagine they'd profile to see whether a load that triggers Turbo is more effective than one that doesn't.

I wouldn't say that it is that deterministic, since it likely depends on the ambient temps of the console.

Would users in AK get a different game experience than gamers in HI, for example? And would those differences be acceptable?

 

[ChronoReverse]

It's dependent on load actually (and correspondingly Power). TDP is called Thermal Design Power but the important part is really Power. The profiling I'm talking includes the loads on the system so it can be accounted for.

From Anandtech's article about the A10

Power is still estimated based on workload, which AMD claims has less than a 1% error rate, but the new model gets accurate temperatures from those estimations.

Nonetheless, I'd hope a provision would be made for the console coders to have more direct control over Turbo (i.e., a way to disable it at the very least).

 

[scannall]

Dont be stupid.

My PS3 has a 90nm CPU and 7800 GTX GPU inside also based on 90nm That was a desktop class GPU

They could easily put a 7870 or 7950 GPU inside and a mobile i7 with HT 3GB DDR5 and be done with it

Depends on what price point you are aiming at. Using high end Intel and high end GPU's in a console will relegate it to taking up shelf space in the store.