Predict: The Next Generation Console Tech

[TheD]

Only it won't look weak next to the existing consoles, which is what most console gamers have as a reference point. A 6670 XB3 will look a generational advance on the consoles, gamers will buy it as such, and it doesn't need to be the flagship box 5 years later. The current consoles look unimaginably pathetic next to what modern tech makes possible these days, but they still sell.
There's nothing illogical about it. It's a perfectly valid strategy. It might not be the one used, but don't go calling those with a different observation to you as illogical. There are multiple logical arguments that can be made for different strategies.

A 3x power increase would be much harder to see than 8-10x, It would have no wow factor to most gamers.
The 6670 is a low midend GPU as it stands right now in early 2012, if a console is launched in late 2013 or 2014 with it, it would be laughed at!
It would be lowend by then and using outdated tech with a poor tessellator.

That would not go down well with the core gamers. The core gamers are the people most likely to buy new hardware and make hype.

Casual gamers are not people that you can "bet the farm on".
How many casual gamers do you think would be wowed by a new console?
How are you going to sell a new console to people who got a console for fitness games or wii sports?
Do you really think that you can get people that are not really into games to spend any money at all on a new console when the one they got suits them fine?

 

[McHuj]

The 6670 is a low midend GPU as it stands right now in early 2012, if a console is launched in late 2013 or 2014 with it, it would be laughed at!

My suspicion on the 6670 rumor was that it's all related to the shader count. (Assuming the rumor to be true for this post) and with AMD 6670 having 480 shaders and the 6750 having 720 shaders, one could make a claim that a GPU with 512 GCN (or customized GCN equivalent shaders) would be close to the 6670, especially if you didn't know the final clock. At the time of the rumor, there was no 77xx out on the market, so there was no GPU with a 1 Tflop perfomance with such a low shader count.

 

[Legendary]

I don't believe for one minute the NextBox will have a GPU the equivalent of a 6670.

 

[Rodéric]

I don't believe for one minute the NextBox will have a GPU the equivalent of a 6670.

I assume you'd like it better.
I wouldn't expect much from next gen...

 

[Platon]

Would a 90watt Pitcairn Pro-based GPU be a realistic possibility for either the next Xbox or the PS4? Obviously both MS and Sony are going to be more conservative next-gen, but a mid-range 2012 GPU in a (more than likely) late 2013 console doesn't seem like that much of a stretch, but I'm sure there are factors I'm not considering...

To me it feels like the Pitcairn series make absolute sense as the next gen GPUs. They offer roughly 10x the transistor count of the 360GPU, at approximately the same die size and depending on the clock consume around 100 watt, it just fits. You also get the nice "10x powa" marketing slogan and so on. Maybe it is wishful thinking but I would like to believe that it is quite realistic, it is not like hoping for something like a 7990 GPU...

 

[AlNom]

Whatever the performance will be, I'm more interested in developers targeting next gen architectures. Personally, I've always done a mix of PC/Console gaming (I've got way more games this gen on PC in fact), so I don't care that much about console aside from what exclusive games will be able to pull off. *shrug* FWIW, I think the art side is much more important going forwards anyway... In that regard, exclusives will have a leg up on multiplatforms there - and who knows if devs will start using the WiiU as a base. >_>

 

[Platon]

...lots of sensible stuff...

I can see the 6670 rumor being trueish if we talk about a double GPU system, then it would kind of make sense. Maybe they could license the SoC part to TV manufacturers or what not, for the new generation of "smart" TVs, to be Live compatible, watch Tv shows and what not from xbox live and then for the xbox they would but one more GPU. I just can't see something equivalent to one 6670 being the GPU that will power the next 7 years of console gaming, even being conservative that would for sure be quite underwhelming...

 

[Gubbi]

Why on earth would anybody stick a 6670 derived part in a new console ?

The new ATI 7770 is about twice as fast and has a very modest die size of 123 mm^2. Double that and clock it slightly lower (ie. speculated 7850-ish performance)

Cheers

 

[AlNom]

Indeed. It's too early to lock down hardware right now.

 

[steampoweredgod]

Sorry for bringing this back up, but was looking for information, and if I'm reading things right it does seem that the 3960x does not fare that well in bandwidth, even with a high level of overclocking and even in a synthetic benchmark(which I hear are often optimized for intel architectures).

the 3960X has double the bandwidth to main memory that Cell does so if bandwidth is the basis for your reasoning that it can achieve a higher percentage of it's theoretical throughput then.. well...

[i7]An integrated memory controller with four channels of DDR3 1600 MHz offers memory bandwidth up to 51.2 GB/s.

Populating all four channels with DDR3-1600 memory, Sandy Bridge E delivered 37GB/s of bandwidth in Sandra's memory bandwidth test. Given the 51GB/s theoretical max of this configuration and a fairly standard 20% overhead, 37GB/s is just about what we want to see here.

The FlexIO interface is organized into 12 lanes, each lane being a unidirectional 8-bit wide point-to-point path. Five 8-bit wide point-to-point paths are inbound lanes to Cell, while the remaining seven are outbound. This provides a theoretical peak bandwidth of 62.4 GB/s (36.4 GB/s outbound, 26 GB/s inbound) at 2.6 GHz.

Reading or Input

[i7] the theoretical transfer rate is 25.6 GB/s (3.2 GHz × 2 bits/Hz × 32 bits/link ÷ 8 (bits per Byte)) per direction

[cell] two IO controllers are documented as supporting a peak combined input speed of 25.6 GB/s

Doesn't look like you can feed the i7 at 2x let alone 50% higher bandwidth than the cell.

i7 3960x L2 cache performance read 70GB/s
CELL Internal EIB bandwidth 300+GB/s

It would be interesting to see how the spe local store bandwidth fares.(the combined spe local store bandwidth utilized during computations would presumably be even higher still)

Again, even in synthetic benchmarks we see.

Sustained benchmark results of overclocked i7 170Gflops, cell sustained benchmark 200~Gflops.

On my 3960X at 4.7Ghz I'm getting about 160-163 GFLOPs with HT enabled. The key is temps. These boards and CPU will throttle quickly without good cooling. My 5Ghz GFLOPs are 165-167.-mdzcpa, evga

Even at 5Ghz in synthetic benchmark it doesn't approach cell sustained performance.

IF the ageia[said to be similar to cell in design] performance was real 200x higher than traditional cpu at physics. I'm most certain the latest i7 isn't 200times faster than the cpus available a few years back. It is likely that physics performance is higher on cell, unless someone can clarify and suggest otherwise.

 

[hoho]

i7 3960x L2 cache performance read 70GB/s
CELL Internal EIB bandwidth 300+GB/s

Isn't the I7 read from a single L2 cache and no from all of them in parallel?

 

[steampoweredgod]

Isn't the I7 read from a single L2 cache and no from all of them in parallel?

Don't know it's a performance benchmark I'd assume it's testing things full throttle as hard as possible, but I'm not familiar with this benchmark so it may or may not be..

edit:
The following is some information from the benchmark product page

Benchmark pages of AIDA64 Extreme Edition provide several methods to measure system performance. These benchmarks are synthetic, so their results show only the theoretical (maximum) performance of the system.

CPU and FPU benchmarks of AIDA64 Extreme Edition are built on the multi-threaded AIDA64 Benchmark Engine that supports up to 32 simultaneous processing threads. It also supports multi-processor, multi-core and HyperThreading enabled systems.

Memory Tests

Memory bandwidth benchmarks (Memory Read, Memory Write, Memory Copy) measure the maximum achiveable memory data transfer bandwidth. The code behind these benchmark methods are written in Assembly and they are extremely optimized for every popular AMD and Intel processor core variants by utilizing the appropriate x86, MMX, 3DNow!, SSE, SSE2 or SSE4.1 instruction set extension.

 

[liolio]

Why on earth would anybody stick a 6670 derived part in a new console ?

The new ATI 7770 is about twice as fast and has a very modest die size of 123 mm^2. Double that and clock it slightly lower (ie. speculated 7850-ish performance)

Cheers

Well because I can discard rumors about specs but I can't discard the rumor about the thing either a GPU or a SoC being tape out.

It will take a lot to convince me about why AMD would hand over GCN IP quiet a long ago when their own chip just launched either for a standalone GPU or in a SOC.
They are not even using it GCN fore their upcoming APU, a critical segment in their strategy.

Assuming that MS or another one for that matter will have access to whatever AMD has because last time, AMD offered MS something custom is when empirical approach fails.

What intensive AMD has to give MS something as important for them as GCN when even "Evergreen" based still offers best perfs per watts and mm2 that whatever Nvidia has? It makes sense for AMD to increase the amount of money they ask if you want newers IP.

As it is Rumors have Nintendo to go with RV7xx architecture why would they do that if not for costs?

As for MS rumors hints at something based on Northern Island chip.

Everybody act as if AMD was in as bad a situation as they was prior 2005 and more importantly as no matter what they are not in a good position to negotiate. Sorry but that just wrong. They are in a position of strength as they have the best products.

Either way 6850 is more likely than 7850 if you see what I mean. Or it's better to qualify the option by number of SIMD arrays and the architecture.

Thing is now in this topic and in the whole console forum too many things are taken for granted, like this point or others (like 360 crowd is technophiles which 1/2 played a graphic powerhouse GeoW2 on sdtv => spotty teenagers...) etc. etc.

Honestly I like your posts it's too bad there is no longer a rep system but I had to say that what is going on is wrong here in the console forum (it's happening more and more). We have to do better it's only the beginning of the silly season if we let us take every assumptions (and repeatedly) as fact without thinking we will end pretty low. As example look at what I've said nobody will come with a good argument about it, it will be purposefully misquoting like "AMD needs the money" and purposefully ignoring my complete statement that is AMD will get the money anyway as they have the best products. Then few pointless posts and some pages ago or in another thread we are here again.


I will give you my answer about why something akin to 6670 could be considered and it simple: it's because it's dirty cheap and mass production is not a problem.
Just as I explain they may want a main chip not big either like Xenon, Xenos, Valhalla were and decided again that is the maximum limit for the size of the chip (for yields, volume early on).
From there the stuff is underwhelming and they add a dirty cheap GPU in dual graphic mode.
I don't believe in something akin to the HD6670 can be the only GPU. it would be a very unbalanced transistor budget (cpu/gpu balance) hence to if it's true to me it has to be a SoC +GPU aka dual graphic set-up.
And for the 6xxxx instead of newer architecture that easy too, AMD asked too much /refuse to hand over something they haven't deploy them selves in all their products.


EDIT

For what it is worth the name of the project could be Durango according to Kotaku

 

[bkilian]

Ahh, that explains why Wii is still selling gangbusters and xb360/ps3 are both in the bargain bin at $99 and still selling less yoy.

That also explains why Apple invested significantly less on the graphics end for ipad2 than ipad1 and then brought the same graphics light chip to their phones.

... or wait ... maybe I'm confused ...

Really? you're finding Xboxes at $99? I'd love to know where, that's cheaper than I can even get at the company store.

 

[TheChefO]

Really? you're finding Xboxes at $99? I'd love to know where, that's cheaper than I can even get at the company store.

I was being sarcastic.

 

[liolio]

I can see the 6670 rumor being trueish if we talk about a double GPU system, then it would kind of make sense. Maybe they could license the SoC part to TV manufacturers or what not, for the new generation of "smart" TVs, to be Live compatible, watch Tv shows and what not from xbox live and then for the xbox they would but one more GPU. I just can't see something equivalent to one 6670 being the GPU that will power the next 7 years of console gaming, even being conservative that would for sure be quite underwhelming...

Even with the SoC GPU power would be nothing to wow about but it makes more sense.
Any decent CPU upgrade would have MS spending transistors in ~50/50 between the CPU or GPU or worse. They are not dumb so I believe indeed that dual graphic is the only reasonable option.

Well It's unclear if they want to last 7 years without a replacement next time. In any case we have to take in account that the pace/rate of performance increases in the PC is decreasing as they hit the power wall and optical reticule size limit.

I could be pleased with a "traditional" set-up like a quad core power7 derivative and a GPU akin to hd6850 with a 128bit bus or 192 bits to main memory.
The question is enthusiast here would be more happy? May be, I'm not confident here.
will it make a sensible difference to the intended market? I'm not confident either.

Between my set-up and a hd6850 the are only 3 SIMD array of difference, either +33% or -25% in computational power. AMD card have texturing power in spare anyway. There is the number of ROPs to consider but it's somehow linked to the memory bandwidth/bus width and the HD 6850 akin thing could scaled down in this regard (people here were adamant that anything but 128 bits bus is impossible too).

Anyway, I guess that were the EDRAM magic bullet enters the show and how it is adverse to modern form of (deferred) rendering that require a lot of RAM. That means tiling, another bus, another (big) chip and possibly games not rendered at 1080 xAA. Things is people here are kind of adamant that games have to be rendered at 1080p either it's BS not worth upgrade.

I'm not sure you see were I'm heading but at some point it's kind of bothering not matter what manufacturers do ultimately.

 

[pjbliverpool]

Sorry for bringing this back up, but was looking for information, and if I'm reading things right it does seem that the 3960x does not fare that well in bandwidth, even with a high level of overclocking and even in a synthetic benchmark(which I hear are often optimized for intel architectures).

Populating all four channels with DDR3-1600 memory, Sandy Bridge E delivered 37GB/s of bandwidth in Sandra's memory bandwidth test. Given the 51GB/s theoretical max of this configuration and a fairly standard 20% overhead, 37GB/s is just about what we want to see here.

So the SB doesn't achieve it's peak theoretical bandwidth, what is that supposed to prove? Neither would Cell. And by the very nature of development for both chips, the code designed to extract performance from Cell is always going to be far more optimised than code designed to extract performance from a number of different x86 architectures. If you have an apples to apples test I'd be happy to see it.

Incidentally, the memory controller will achieve higher bandwidth than that with faster memory:

http://www.legitreviews.com/article/1773/5/

The FlexIO interface is organized into 12 lanes, each lane being a unidirectional 8-bit wide point-to-point path. Five 8-bit wide point-to-point paths are inbound lanes to Cell, while the remaining seven are outbound. This provides a theoretical peak bandwidth of 62.4 GB/s (36.4 GB/s outbound, 26 GB/s inbound) at 2.6 GHz.

In the PS3 much of that FlexIO bandwidth is given over to communication with the GPU. If you want to compare total off chip bandwidth then I'm sure you'll want to consider the 80GB/sec worth of PCI-E lanes the 3860x has available to it. As well as the 25.6GB/sec QPI link. And of course the 4GB/sec DMI 2.0 link just for good measure.

Reading or Input

[i7] the theoretical transfer rate is 25.6 GB/s (3.2 GHz × 2 bits/Hz × 32 bits/link ÷ 8 (bits per Byte)) per direction

[cell] two IO controllers are documented as supporting a peak combined input speed of 25.6 GB/s

Doesn't look like you can feed the i7 at 2x let alone 50% higher bandwidth than the cell.

If you're going to post quotes then you need to post their source. For example that first link, does it refer to Sandbridge or Sandybridge E? As far as I'm aware the memory channels on an i7 of any type are bi-directional which means that 51.2GB/s can be all read, all write or any combination inbetween.

The second link simply repeats what was mentioned earlier with regards to peak FlexIO transfer capacity without any regard to what other off chip communication might be required. The simply fact remains that the PS3 implementation of Cell has 25.6GB/s of main memory bandwidth (IIRC it's fixed at 12.8GB/s in either direction but I may be wrong) and that's half the main memory capacity of an i7.

i7 3960x L2 cache performance read 70GB/s
CELL Internal EIB bandwidth 300+GB/s

Why are you comparing internal bus bandwidth in Cell to L2 cache??! Surely you should be comparing this to the ring bus bandwidth of Sandybridge (which I'm not aware of).

It would be interesting to see how the spe local store bandwidth fares.(the combined spe local store bandwidth utilized during computations would presumably be even higher still)

While we're comparing random numbers why not look at L1 cache bandwidth while we're at it which has a measured bandwidth of ~600GB/s on older i7's:

http://techreport.com/articles.x/21813/6

Again, even in synthetic benchmarks we see.

Sustained benchmark results of overclocked i7 170Gflops, cell sustained benchmark 200~Gflops.

On my 3960X at 4.7Ghz I'm getting about 160-163 GFLOPs with HT enabled. The key is temps. These boards and CPU will throttle quickly without good cooling. My 5Ghz GFLOPs are 165-167.-mdzcpa, evga

Even at 5Ghz in synthetic benchmark it doesn't approach cell sustained performance.

And like the last time you posted this I'll ask again, was that measuring single or dual precision performance?

And how efficient is a generic x86 benchmark going to be compared with hand optimised code built specifically for Cell? Does that benchmark even leverage AVX? Unless you have a benchmark that's actually comparable between the two architectures this is a complete waste of time.

IF the ageia[said to be similar to cell in design] performance was real 200x higher than traditional cpu at physics. I'm most certain the latest i7 isn't 200times faster than the cpus available a few years back. It is likely that physics performance is higher on cell, unless someone can clarify and suggest otherwise.

So your using 5 year old marketing claims from a dead company and a fairly wild assumption that the Ageia PPU would have had the same physics performance as Cell because they have very vaguely similar architectures to prove your point? Shall we just end this now?

 

[liolio]

Isn't the I7 read from a single L2 cache and no from all of them in parallel?

I'm not sure in which way his comparison is relevant. Actually one has nothing to do with the other.
BY the way SnB of 394 GB/s of aggregate bandwidth to L3 stat it and it offers coherency, etc. (the traffic has to be higher).
http://www.anandtech.com/show/3922/intels-sandy-bridge-architecture-exposed/4

 

[TheChefO]

Even with the SoC GPU power would be nothing to wow about but it makes more sense.
Any decent CPU upgrade would have MS spending transistors in ~50/50 between the CPU or GPU or worse. They are not dumb so I believe indeed that dual graphic is the only reasonable option.

Well It's unclear if they want to last 7 years without a replacement next time. In any case we have to take in account that the pace/rate of performance increases in the PC is decreasing as they hit the power wall and optical reticule size limit.

I could be pleased with a "traditional" set-up like a quad core power7 derivative and a GPU akin to hd6850 with a 128bit bus or 192 bits to main memory.
The question is enthusiast here would be more happy? May be, I'm not confident here.
will it make a sensible difference to the intended market? I'm not confident either.

Between my set-up and a hd6850 the are only 3 SIMD array of difference, either +33% or -25% in computational power. AMD card have texturing power in spare anyway. There is the number of ROPs to consider but it's somehow linked to the memory bandwidth/bus width and the HD 6850 akin thing could scaled down in this regard (people here were adamant that anything but 128 bits bus is impossible too).

Anyway, I guess that were the EDRAM magic bullet enters the show and how it is adverse to modern form of (deferred) rendering that require a lot of RAM. That means tiling, another bus, another (big) chip and possibly games not rendered at 1080 xAA. Things is people here are kind of adamant that games have to be rendered at 1080p either it's BS not worth upgrade.

I'm not sure you see were I'm heading but at some point it's kind of bothering not matter what manufacturers do ultimately.

So your essentially suggesting a SOC + GPU with the idea that the SOC will be universally utilized in multiple other devices outside of gaming ...

I get where you're coming from concept wise, but that would seem to be a bit too high powered for other embedded devices.

Modularity in design is a good concept (Sony attempted the same with Cell) but the target use cases have to make sense.


I will say one thing that would be a coup for MS on a profitable from day one console:

Windows8

If MS isn't worried about losing money on the console at the start, they can also release Windows8 on it, or just put it in the box at the start without worrying about how many games people buy.

HD Kinect for a mouse interface and voice recognition for a keyboard (or optional on-screen variety with HD Kinect).

Implications of such a system would be huge for parents looking to buy the family a cheap computer and would help adoption and integration of Windows8/metro.

It would suck for the loss of Hardcore gamers and they'll likely get shafted by developers going forward, but in the longrun, it might be the best strategy.

In that light, your proposal makes a lot of sense.

Core $299 (windows8 + KinectHD sold separately)
Ultimate $499 (with KinectHD, HDD and Windows8)

It would sell.

They'd lose the core gamers, a chunk of developers, a chunk of XBL money, and the attache rate (licensing fees) that core gamers brought to the table ... but they might make up for it in casuals.

And honestly, they're not really losing the core if the core heads to PC gaming aka Windows land. But this would ultimately lead to unsatisfied gamers as devs target such a low spec again, gamers might just lose interest and not feel the need to support the industry financially (aka more piracy). This would lead to developers/publishers saying screw it, and going belly up, leading to the death of the gaming industry as we know it ...

But we'd always have Angry Birds.

On second thought, that idea sucks.

Seriously though a low spec console does have merit for MS if OS integration is in the plans, but then where does that leave Sony?

Do they follow suit with a low spec ps4? Or do they embrace the opportunity and snatch up the core gamer for themselves?

If the opportunity presents itself, I hope Sony has the financial ability and vision to pull it off.

 

[liolio]

So your essentially suggesting a SOC + GPU with the idea that the SOC will be universally utilized in multiple other devices outside of gaming ...

Hum no that's Platon' POV. I'm not sure that the SoC would be a good target for the kind of uses you/he describe(s).
If it were to happens I can't see it used in a Xbox, more something like a hypothetical MS-TV. I would place the odds close to zero through.

The main motive I see for MS to use such underwhelming system is costs, keep the cost a low as possible and achieve high production volume at launch. So SoC would includes more suckys CPUs and a few SP I'm not sure that by it-self the thing would have any interest. It's part of a bigger picture.

The idea is trading any convenience for max throughput in the most costs constrained environment:
IO cpu cores, 2 different GPUs still using the same architecture, possibly few ram, etc. Not about having some use for the SoC by it-self.

MS is committed now to the ARM architecture so I don't think that the option is relevant.

I believe each manufacturers deserves its own thread as we are closing in now. For Sony. I think that they have to go down the road of KK vision, provide an open computer, so for me they have to adapt Android as anything in house is set to fail and so an unnecessary expense.
It's been my pov for a while, Sony should adopt Android asap and lead. That's for ps4, psvita, xperia plays, tabs. By the way Sony Erickson is officially done it's Sony mobile communications (sucky name...).

But Sony case is tough case. They are in a way worse shape than what most here dare to acknowledge. I believe that Gakkai CEO were speaking of them when she said, one manufacturer could pass on next gen. That's how bad their financials are. Sony can't afford loss leading model now, PSV could prove tough to sustain already in their condition.
It's a shame that people get fire when the responsible for such a series of bad decisions are still around. I can't believe Sony get themselves overall in such a mess. I'm convinced that with good decisions (and not even great ones) they would have pushed MS out and let them with ultimately +6billions in losses...