Predict: The Next Generation Console Tech

Status
Not open for further replies.
I'm comparing Llano to the wii and to xbox 360. Compared to those its very capable and very cheap and would allow nintendo to go the wii route again.

Of course if you wanted to build a power house consle that blows Llano out of the park you could. But it be much more expensive.

I was just saying how necessary bandwidth will be for Llano in a machine centered on gaming. Why go through the expense of having such a "capable" chip in console specifically for gaming when it would be severely bottlenecked by bandwidth (assuming 128 bit + DDR3)? Around ~25 GB/s of bandwidth would match fine with a part with 4 x86 cores + 200 SPs, but with 400 SPs? It would be a waste of silicon, so GDDR5 + 128 bit pretty much would be mandatory for a Llano based system.

Llano + 1 GB GDDR5 I would bet is pretty affordable and shouldn't consume too much power, probably in the 40W area when using the video of Llano vs SB as a starting point for guessing. Of course I am assuming 1 GB of GDDR5 vs 4 GB of DDR3, no monitor/speakers, simplified motherboard and I/O.
 
That is false. No x86 tax on a CPU will make it cheaper. A Bobcat CPU is pathetic even compared to the current 360 and PS3 cpus.

not always, I remember John Carmack on the x360 CPU : half the single thread performance of a pentium 4 3.2GHz. hence the slow quake 4 port on the console's launch.
so maybe bobcats would match a xenon in integer performance :p, not conclusive on everything. Sure it gets its ass handed over for those paralle, simd-friendly tasks that manage to use the gigaflops.

you might increase clock a bit (but pipeline is maybe tuned for low frequency) and have four bobcats cores.

perhaps we're comparing crap CPUs with crap CPUs.
they're crap for different reasons, bobcat has weak execution resources but use them efficiently, xenon has a lot execution resources and is wasteful.
 
Are there any SpecInt numbers for XBOX360?

PS. 423 SpecInt 2000 for the PPE, anyone with an E350 which can run it? :)
 
Last edited by a moderator:
Per-clock is one part of the overall equation. Neither Atom or Bobcat can get near the clock speed of Xenon, so even a modest per-clock advantage does not mean there is an overall improvement.


There are also situations where OoOE does not yield significant performance gains, such as when there are memory latencies too large to hide, or the code is scheduled well enough that minimal reordering is necessary.
In areas of code with low ILP like some long chains of pointer-chasing, clock speed and the memory subsystem become dominant, and other parts of the architecture do not increase performance.
Without a redesign, Bobcat is not capable of scaling its clock close enough to Xenon to make up for the large clock disparity, at least not at acceptable voltages.


That might require some additional engineering. AMD does not have an uncore like Sandy Bridge, nor does it seem to have one planned for a while. It tends to use a crossbar that takes up a fair amount of space and is less scalable to higher client counts. The modularized Bulldozer keeps the crossbar client count lower by making the pairs of cores share an interface.


The average IPC for a number of workloads for Athlon was around 1 per cycle, at least on some desktop media applications when Anandtech looked into it a while ago.
There are low periods and burst periods. If a core is narrow, and if it is fighting for shared resources, those burst periods take longer. The probability, particularly under load conditions, goes higher because the bursts take longer to get through.


There are limits to what can be gained by adding more threads. The more thread-level parallelism is exploited, the more serial components dominate. Console single-threaded performance is not yet so high that it can be ignored, much less made worse.


If you are stating that if Bobcat were redesigned so it was more like Bulldozer and had more resources to burn that it would benefit more from being put into a module, then I have not said anything to the contrary.


AVX has 8-wide SIMD, not 256.
Fermi and Cayman have 16-wide units, but due to their batch sizes, their minimum granularity is 32 and 64, respectively.


The cores running the MLAA algorithm are most likely larger than the silicon found in the ROPs.
SRAA in particular uses higher-resolution buffers, which would take longer to build if the ROPs were scaled back.
3dilettante thanks for your insights ;) (in regard to AVX units I meant 256 Bits wide).

I managed to read a proper review of Bobcat (as expected from techreport :) ). They compare to plenty of modern CPUs and the overall result is that bobcat is indeed way more suckier than I thought. In regard to high hardware utilization, in some benchs Atom & Bobcat end consumming consistently more energy than beefier CPUs as execution time is too/so long.
So clearly I overestimated it by a fair amount, I'm calling for something that's such a leap from Bobcat that there would actually no reason to call it bobcat 2.
Let say I pass on bobcat as a basis but I may dare a last question.

Let forget X86, PowerPC, ARM, and their respective vendors. Let's take the whole thing from a higher level. Next generation systems are likely to chip with motion sensing pad, an IR camera on top of the system if manufacturers want to avoid prohibitive prices or abysmal losses they have to do more arbitration on the "core" hardware than they did this gen. So I'm a believer that to achieve this their best chance is a well designed fusion chip. So they will be limited in die size, power, thermal consumption. I assume they won't go for the highest single thread performance but they will want better parts than what they have now (which should not be that tough). I'm not sure either about really wide SIMD, say 256bits / 8wide, as by looking at what Intel did with SnB you need to make a significant investment to get those beasts (costly in silicon) fed and making the most out of their potential. So I'll put by bet on better 4 wide SIMD than wider ones.
So I would bet on pretty narrow OoO CPUs, 2 or 3 issues with potent 4 wide SIMD with dynamic speed clock and other good power containment features. The question is which form of multi-threading, SMT or CMT? CMT looks like a good idea which may not only be appreciated by comparing AMD bulldozer performances to their Intel counterparts.
As I understand it either you save power and die space through CMT but you can also reinvest those gains on a better front end and various architectural improvements.
What is you're opinion on the matter?
 
Last edited by a moderator:
Anybody get the feel, here at the 5 year mark where we would normally be seeing new consoles, that the software makers at GDC are making their push for next gen? Between Battlefield 2, the next gen UE demo, and even Cryengine 3. I definitely do. I imagine the hardware makers may be feeling a little bit of pressure. Or they may not care at all :LOL:

I do find it funny, that at least in Epic's case, and Cryteks a bit too, the pressure might come from software devs, when they're always the ones crying about multiplied higher costs with each gen :LOL: Bit hypocritical.

I also think the Wii2, if it ever comes out anytime soon, could actually be the catalyst that pushes MS and Sony into next gen. If it's more powerful than PS360, it could force some hands. Something like 1Gb of RAM would seem a doddle for Wii2 seeing thats in phones nowdays, and thats twice as much as PS360.
 
I agree Rangers but I suppose it all comes down to the dollar.
If people are still buying the consoles in droves and they are being sold at a profit then why should they move on.
Microsoft has it easy at the moment, hardly any first party offerings and they still outperform sony. I have asked my friends why they prefer xbox360 and none of them have kinect but they prefer the xbox for 2 reasons and that's, xbox live and better multi platforms.
They all think halo has dragged on as a franchise and to them it's not a system seller anymore.
None of them are loyal to any console and they will all jump on whatever console is the most powerful next gen and so will I.
It was strange that not one of them mentioned wanting any first party games from any of the console makers but they are curious to know when they arrive.
This is from gamers who don't even open a games magazine.
 
No one will launch a next gen console with more than a 12 month head start. Nintendo can probably get away with launching first with an HDWii3D and not worry too much about Sony and MS launching much more powerful consoles. Sony and MS will probably want to be careful about launching a console that will be easily overshadowed within 12 months.
 
Wasn't the xbox360 meant to have 256MB RAM but EPIC told them to up it to 512.
That cost them 1 billion$.
If that was true and they didn't launch with 512MB then the PS3 would have truly been the more powerful console.
I can't wait to see what both company's come up with but what will probably push me towards MS is the amount of XBLA games I have. I would like to think they will be compatible and hopefully have some sort of enhancements to 1080p and 60FPS.
What worries me is they no longer have the staff that thought up the achievements and friends list ideas. They appear to be alot less of the company that launched the xbox360.
 
So I would bet on pretty narrow OoO CPUs, 2 or 3 issues with potent 4 wide SIMD with dynamic speed clock and other performant power containment feasures.
The original hope for Xenon was that it was going to be OoO, but the time-to-market was so short that IBM told Microsoft they would not have the time to properly design and validate the part.
This time around, given the apparent lengthening of the design cycle, the odds are that they will have the time.
The escalating power costs for higher clocks will probably leave designers looking to increase IPC.

The question is which form of multi-threading, SMT or CMT? CMT looks like a good idea which may not only appreciate by comparing AMD bulldozer performances to their Intel counterparts.
As I understand either you save power and die space throught CMT but you can also reinvest those gains on a better front end and various architectural improvments.
What is you're opinion on the matter?

If the chip is OoO, SMT would be cheaper to implement than a Bulldozer-like module, although the performance would be more variable and generally a little lower.
Some facets of AMD's solution are more appropriate for servers, and may not be the best for a console chip.
AMD's design assumes the FPU will be used significantly less than the integer side, which is why it is shared. Consoles don't even run a large amount of integer programs, so this may not be a win for that space.
Bulldozer's front end (Icache, branch preditor, decoder) is also very heavily overdesigned for the console space, so once again I am unsure if it will be as significant a win.
It may be better to have decent SIMD units on a decent number of actual cores, in order to ensure that the FPU throughput is improved with the next generation. Sony may not want to cut down on the number of FPUs, for example, since it now heavily leans on its multitude of separate SPEs.

I'm not entirely sure, since we haven't seen how BD turns out, and since the next generation looks to be delayed a bit, the parameters may change again by the time the design efforts resume.
 
Wasn't the xbox360 meant to have 256MB RAM but EPIC told them to up it to 512.
That cost them 1 billion$.
If that was true and they didn't launch with 512MB then the PS3 would have truly been the more powerful console.
I can't wait to see what both company's come up with but what will probably push me towards MS is the amount of XBLA games I have. I would like to think they will be compatible and hopefully have some sort of enhancements to 1080p and 60FPS.
What worries me is they no longer have the staff that thought up the achievements and friends list ideas. They appear to be alot less of the company that launched the xbox360.

The PS3 was slated to be 256 MB XDR originally if I remember correctly as well. The 360 moving up to 512 and Sony having to resort to using an off the shelf GPU made them add the GDDR3.
 
If Microsoft goes to for example AMD for the CPU, am I right to assume that it kills the chances of backwards compatibility for the 360? Or would there still be hope of at least emulating the "important" titles, like the 360 did for the Xbox 1?
 
Does this signify any movement on the next gen front?

http://www.wired.com/gamelife/2011/03/epic-unreal-engine-demo-gdc/

The demo was created by a 12-person team in about two months, Epic Games President Michael Capps said during a Thursday press briefing at GDC. “Samaritan” was running on a custom-built system using off-the-shelf PC parts, Epic said, including three Nvidia GTX 580 graphics cards, which cost about $500 each.

Very few if any consumers today would be able to run the demo, but with the rapid advancement of technology, it’s not unlikely that we’ll be playing something that looks like this before too long.

“We have a pretty good idea of what’s coming next,” said Capps, who added that Epic is already in preproduction on a brand new title for the next generation of game machines
.
 
Emulating Cell would be nigh impossible on anything other than a similar architecture including instruction set.

Doesn't that depend on the outright performance and structure of the next generation architecture? For instance the Cell is 7 SPE + 1 PPE, surely this could be emulated on a PPC7 architecture given they share similar instruction sets because anything that can run on the PPE can be made to run on the SPEs and vice versa, so essentially a far away improved PPE could run everything?
 
I would think biggest problem would be emulating SPE's local memory. It's not exactly too easy to emulate that.
 
Status
Not open for further replies.
Back
Top