Predict: The Next Generation Console Tech

[Prophecy2k]

Arthur Geis' info. is old. It's all stuff we've heard regurgitated and twisted over and over many times now.

Currently I don't think anyone is really in the know particularly about what is going on with both Next-gen consoles. Even publisher execs won't know the exact specs, only general performance targets and likely some key pieces of info. like RAM amount and target memory bandwidth.

I think like Nintendo, Sony and MS have learned to keep their cards closer to their chests this time. I don't suspect that we'll know exactly what is the PS4 and Durango untill after they've released.... Not unless either of the two parties are feeling particularly candid about sharing, or alternatively they truly believe they have a BEAST of a next-gen console, want to boast heavily about it and use that as a selling point, whilst believing it to be too late in the game for the competitor to switch things up and make ammendments.

I believe both console will have the same or similar memory amounts. Particularly for Orbis, i'm confident that this was changed not too recently (well as confident as anyone an be in second hand whisperings).

 

[ultragpu]

The news about AMD 8000 series releasing in the 2nd quarter of 2013 is rather tantalising, I wonder if any party has moved onwards to that SI chip yet. I hope CES brings us the bacon in 4 weeks.

 

[bkilian]

Arthur Geis' info. is old. It's all stuff we've heard regurgitated and twisted over and over many times now.

Currently I don't think anyone is really in the know particularly about what is going on with both Next-gen consoles. Even publisher execs won't know the exact specs, only general performance targets and likely some key pieces of info. like RAM amount and target memory bandwidth.

Execs and actual devs know a lot more than that. They know pretty much everything about the hardware except maybe the IP providers, but at this point they know the full hardware system architecture. I'm stunned at how little info is out there considering how many people are fully read in.

 

[almighty]

4 Core Haswell would be twice that - 409.6 GLFOPS via AVX2.

Current Sandy and Ivy bridge processors peak at 204.8 with 4 cores at 3.2Ghz.

A 5Ghz 2500k only generates ~125Gflops when running Intel Burn Test so I'm failing to see where this 204Gflops figure comes from.

 

[Hornet]

A 5Ghz 2500k only generates ~125Gflops when running Intel Burn Test so I'm failing to see where this 204Gflops figure comes from.

We are talking about single precision arithmetic, not double precision.

 

[Proelite]

Execs and actual devs know a lot more than that. They know pretty much everything about the hardware except maybe the IP providers, but at this point they know the full hardware system architecture. I'm stunned at how little info is out there considering how many people are fully read in.

How likely is it for ram, number of cores, and other specs outside of clock rates to change at this point?

 

[Gipsel]

How likely is it for ram, number of cores, and other specs outside of clock rates to change at this point?

My uneducated guess would be that the number of cores and their capabilities are fixed at this point in time. They may have changed the amount of RAM recently. That's the only thing somewhat flexible (besides the clocks) at this stage I suppose. So the devs who got briefed should have quite a good knowledge of what is to come.

 

[almighty]

We are talking about single precision arithmetic, not double precision.

If it takes a ~4.8Ghz 2500k to hit 100 DP Gflops do you think a stock clocked 2500k can achieve over 200 SP Gflops?

Because I don't.

 

[Hornet]

If it takes a ~4.8Ghz 2500k to hit 100 DP Gflops do you think a stock clocked 2500k can achieve over 200 SP Gflops?

Because I don't.

The peak throughput is 204.8 GFLOPS at 3.2 GHz. The actual throughput in Linpack is usually between 80% and 90% of the peak for most architectures. It is generally slightly higher in matrix multiplies (sgemm). By the way, I don't see the point of using an overclocked CPU as the basis for comparison. A stock Xeon E5-2680 achieves 157.7 GFLOPS in double precision linpack compared to a peak of 172.8 GFLOPS (91% of peak).

 

[Ruskie]

How can you go (in Sony's case) from 2GB of GDDR5 to 4-8GB unless you completely change system architecture in favor of slower memory since 8 GDDR5 would be lots and lots of chips and lots and lots of watts.

 

[nunomoreira]

How can you go (in Sony's case) from 2GB of GDDR5 to 4-8GB unless you completely change system architecture in favor of slower memory since 8 GDDR5 would be lots and lots of chips and lots and lots of watts.

It could be same number of chips, just denser, not really an increase of heat there.
and 8gb is a bit farfetched, its not like the console is running windows on the background..

 

[McHuj]

How can you go (in Sony's case) from 2GB of GDDR5 to 4-8GB unless you completely change system architecture in favor of slower memory since 8 GDDR5 would be lots and lots of chips and lots and lots of watts.

2GB of GDDR5 I believe was probably based on 8 2Gb GDDR5 chips. The rumored BW of 192 GB/sec probably indicated a bus width of 256. Previously in this thread it was posted that 4Gb GDDR5 chips will be available in 2013. So I think they could go from 2GB to 4GB of GDDR5 on the same width bus and only 8 4Gb chips, but I imagine the cost will be really high, so I'm not sure it's economically feasible.

Having a system consisting of 2GB GDDR5 and 2-4 GB of DDR3/4, would certainly be a significant change from a GDDR5 only system. You need to have two separate buses, 2 memory controllers, etc.

I'm actually not sure this type of setup would work with an APU as well. The APU would probably need to support a huge bus (256-bit for the GDDR5 plus another [at the very least] a 64-bit one for the DDR3/4). Die shrinks and cost reduction would be very hard, IMO. I guess they could reduce the bus for the GDDR5, but that would seem to be a worse solution.

 

[MrFox]

How can you go (in Sony's case) from 2GB of GDDR5 to 4-8GB unless you completely change system architecture in favor of slower memory since 8 GDDR5 would be lots and lots of chips and lots and lots of watts.

I wonder if they planned to use wide memory from the start instead of GDDR5, current chips are is 4Gbit per die, and I think Elpidia was supposed to be sampling 4 stack high 16gbit already this year. So maybe they planned the design for 4 chips wide, they can do 2GB without TSV (low cost, lower risk), or the liberty to change for up to 8GB with TSV. No change to anything since the interface and the chip size are exactly the same.

Either that, or it's a major change and it explains why AMD is 6 month behind schedule for Sony, but on schedule for Microsoft (cause microsoft planned it correctly from the start)?

 

[MrFox]

its not like the console is running windows on the background..

Why not?
I can see microsoft wanting to have it run Windows8 RT, it's not such a crazy idea now that they have a walled-garden version with app store and all.

 

[liolio]

If it takes a ~4.8Ghz 2500k to hit 100 DP Gflops do you think a stock clocked 2500k can achieve over 200 SP Gflops?

Because I don't.

THe calculation is nothing fancy
4 cores that work on 8 SP FP elements and can pull one FMA (2 FLOPS per cycle). At 3.2 GHz that is:
4x8x2x3.2=204.8 GFLOPS
Haswell can do 2 FMA per cycle.

 

[lefizz]

I think next box running some version of Windows 8 seems incredibly likely. Having a common interface between console, PC and mobile and a similar if not the same app store seems pretty desirable.
Hopefully (from their point of view) they would get some crossover customers to there Mobile platform which they desperately need.

Also why reinvent the wheel several times it simply not needed and costly for no gain.

 

[jonabbey]

What makes me kind of sad is that whereas it never shipped when I read the 'few" here that had the chance to play with Larrabee I've the, may be wrong, sense (reading to much into), that they really enjoyed working on it.

Larrabee actually has shipped. Not as a GPU, but as a compute cluster on a chip, called the Xeon Phi.

See Charlie at
http://semiaccurate.com/2012/11/13/what-will-intel-xeon-phi-do-to-the-gpgpu-market/

 

[tongue_of_colicab]

I think next box running some version of Windows 8 seems incredibly likely. Having a common interface between console, PC and mobile and a similar if not the same app store seems pretty desirable.
Hopefully (from their point of view) they would get some crossover customers to there Mobile platform which they desperately need.

Also why reinvent the wheel several times it simply not needed and costly for no gain.

Didn't the xbox and xbox360 use a modified version of windows 2000?

Obviously they won't be making a totally new OS just for their console and it will have it's base in some windows OS.

 

[ERP]

Didn't the xbox and xbox360 use a modified version of windows 2000?

Obviously they won't be making a totally new OS just for their console and it will have it's base in some windows OS.

Only in the loosest possible way, the kernel was based on the existing NT kernel at the time, the user level API's are for the most part similar.
It's not the same as "running a version of windows", there is for example nothing that looks like the windows DDM, and the game app has effectively exclusive access to the GPU.
If they are going to "run a version of Windows" on it there is a good chance they lose a lot of performance doing it.
FWIW I also think that it would distract from their focus on gaming, once Windows is involved making decisions inside MS becomes a LOT harder.

Having said all of that the patent that was circulated months ago, to me looks like a hardware solution to the political problem of having to run windows on a console, so who knows.

 

[bkilian]

How likely is it for ram, number of cores, and other specs outside of clock rates to change at this point?

Highly unlikely. I could see some clock rates changing, but even that would be unlikely. Any change at this point would likely result in a release slip of months at the minimum. Also folks (like Charlie) have been rumbling that the new console was originally one architecture and then changed to another one. That's simply not true, there has only been one architecture considered for the device from the start of the project. A few things did change, but not big changes. It's how I know Charlie is full of crap.