NGGP: NextGen Garbage Pile (aka: No one reads the topics or stays on topic) *spawn*

[Ruskie]

It's not only more FLOPS, it also has more CUs, more TMUs and more ROPs

It has two more CUs that are clocked lower and bring theoretical shading performance in almost exact same number (+~50FLOPS). It has the same number of ROPs too.

So, effectively Orbis GPU is 7850 and Durango is 7770, like I said. Interestingly, thats exactly the same thing Sweetvar26 mentioned too. That guy got everything completely right.

This is also off topic and I think I'm baiting a ban hammer on me.

 

[Hecatoncheires]

So, effectively Orbis GPU is 7850 and Durango is 7770, like I said.

No, according to the rumors Orbis has more CUs, more FLOPS, more bandwidth and more TMUs than a HD7850. Durango has more CUs than a HD7770 and it also has embedded RAM.

 

[Pugger]

It has two more CUs that are clocked lower and bring theoretical shading performance in almost exact same number (+~50FLOPS). It has the same number of ROPs too.

So, effectively Orbis GPU is 7850 and Durango is 7770, like I said. Interestingly, thats exactly the same thing Sweetvar26 mentioned too. That guy got everything completely right.

This is also off topic and I think I'm baiting a ban hammer on me.

I don't think your baiting a ban hammer. From the specs released and IF they are correct, theres no doubt that Durango is a slightly modified 7770 and Orbis is a slightly modified 7850. Thats the reason we have seen many quotes (from unnamed sources mind) that state the specs are very similar. They are. Yes Orbis will have more legs but there's no massive jump from one to the other. In fact this is as close as developers are going to get to a one system setup, they will be able to shift code between platforms easily.

 

[AstoundingHolmes]

Yep. For all we know, Durango is a TBDR device with API's cutting up workloads into small pieces.

You've mentioned this a few times before, could you expand on it a little? I'm not really familiar with the ins and outs of TBDRs, outside of grasping the very basic concept how they work. What would be the tradeoffs/gains with going the TBDR route? My limited understanding is that geometry and possibly lighting would take a hit due to them crossing tiles, but the processing performance/complexity of each tile can be increased due to being localized in cache, right?

 

[Ruskie]

No, according to the rumors Orbis has more CUs, more FLOPS, more bandwidth and more TMUs than a HD7850. Durango has more CUs than a HD7770 and it also has embedded RAM.

And how exactly is that different then what I'm saying? It has two more CUs that are clocked LOWER, bringing theoretical shading performance (TFLOPS) in almost equal terms (give or take few dozen FLOPS, like in Durangos case).

The whole system has more bandwidth, but that is shared between CPU and GPU (176GB/sec CPU/GPU vs 153GB/sec for GPU alone). Thats why I said, Orbis is effectively 7850 and Durango is 7770 (both have more CUs, lower clock and slightly more TMUs then their PC "brethren") except Durango can use 102GB/sec from ESRAM as well as pool of 68GB/sec from main RAM pool which is significantly more then HD7770 can use.

 

[dobwal]

You've mentioned this a few times before, could you expand on it a little? I'm not really familiar with the ins and outs of TBDRs, outside of grasping the very basic concept how they work. What would be the tradeoffs/gains with going the TBDR route? My limited understanding is that geometry and possibly lighting would take a hit due to them crossing tiles, but the processing performance/complexity of each tile can be increased due to being localized in cache, right?

I am not an expert but TBDR is more energy and bandwidth efficient because it depends on more local memory to avoid read-modify-write calls to main memory. TBDR dominates mobile devices because of its energy efficiency.

MS's Talisman worked similarly but it was a software based solution and the advent of gpus made it unnecessary.

 

[Averagejoe]

both have a gpu worse than 7900 mobile, 12 or 14 CU's is trivial, I don't know how people can be disappointed by one and happy about the other, in my opinion they are both in the same boat, both disappoint , both are in the best case on the mid-class league from a PC stand point, difference minimal, both overperformed by a mobile chip, what are we talking about? from two or three years from launch will be there smartphones with 4 GB ram CPU16+ cores @ 2+ GHz and probably a better gpu too
it's a joke?

12 vs 14 is trivial but you are forgetting something.

Orbis has 4 CU for compute jobs,if the game heavily use this on Orbis what will you use to emulate those on Durando.?

CPU + some of the 12 CU units.?

But that means you are using resources from those 12 CU and the CPU,while Orbis still has most of its CPU + 14 CU available.

Is the only reason i think there will not be parity if this specs are true,even less on exclusives.

 

[Ruskie]

12 vs 14 is trivial but you are forgetting something.

Orbis has 4 CU for compute jobs,if the game heavily use this on Orbis what will you use to emulate those on Durando.?

CPU + some of the 12 CU units.?

But that means you are using resources from those 12 CU and the CPU,while Orbis still has most of its CPU + 14 CU available.

Is the only reason i think there will not be parity if this specs are true,even less on exclusives.

There is still nothing we know of Durango's CPU. I'm not sure how much they could customize it, but if there are additional vector units on it FLOP count could shrink (albeit nothing like ~400 GFLOPs, its 1.6Ghz CPU after all).

Based on specs GPU in Durango will be very well "fed", and by looking at 16 ROPs I guess they aim at 1080p 30fps. Sony aims at 1080p 60fps, but will be interesting what happens if devs opt for MS target since that would make 32 ROPs less of an advantage.

 

[Averagejoe]

Yes Cell was good but at a cost and is one reason why it had a weak gpu. And if all Sony could get was a G71 from Nvidia whats what makes you think they would of got Xenos from AMD.

Are you for real dude.?

The PS3 cost $800+ dollars to manufacture far far more than the 360 cost to manufacture,if sony would have chase power with that kind of budget it would have out power the xbox 360 in a considerable way..

Let me put it this way erase Blu-ray,and add a Xenos alone side Cell with 512MB of DDR3 and 256 XDR for system ram.

Some of you may say it was impossible,but sony bom on the PS3 far exceed that of the 360 basically from every point but GPU.

Is the reason why sony loss so much how people can ignore this is crazy.

 

[dragonelite]

There is still nothing we know of Durango's CPU. I'm not sure how much they could customize it, but if there are additional vector units on it FLOP count could shrink (albeit nothing like ~400 GFLOPs, its 1.6Ghz CPU after all).

Based on specs GPU in Durango will be very well "fed", and by looking at 16 ROPs I guess they aim at 1080p 30fps. Sony aims at 1080p 60fps, but will be interesting what happens if devs opt for MS target since that would make 32 ROPs less of an advantage.

Does that work like that?
Doesn't the cpu/gpu also need to recalculate stuff at 60 fps?

Had hoped to get better hardware then my laptop but then again its adapter is almost as big as a nuclear reactor and it cost 2 times as much as a console..

 

[Ruskie]

Does that work like that?
Doesn't the cpu/gpu also need to recalculate stuff at 60 fps?

Had hoped to get better hardware then my laptop but then again its adapter is almost as big as a nuclear reactor and it cost 2 times as much as a console..

Well, after 1st wave of games for the console there will be frame rate reduction, but I don't see a way in which Orbis won't be able to run all launch titles at 1080p/60fps. There is no bandwidth/ROP bottleneck that would limit it for now. Later, I of course expect them to drop 60fps and cram as much as they can in 30fps budget.

 

[patsu]

I said it several times already: DX11 level of graphics is full of effects that require GPU computing: You need it for lighting, for ambient occlusion, for depth of field, bokeh, motion blur, lens flare, fluids, textiles, particles, smoke, etc. A desktop PC GPU like the HD7850 has to detach some computational resources for these graphics effects as well. Having 4 Orbis CUs ready for GPGPU is no downgrade in terms of graphical capabilities unless you want DX9 level of graphics without these fancy effects. Take a look at Uncharted: Naughty Dog used the Cell SPEs to support the RSX with graphical computations (lighting, ambient occlusion, etc). If it's there then it'll be used.

Yes, besides graphics work, the SPUs also run audio, video, security, uncompress, AI, physics and other odd jobs. For next gen, the "compute units" seem more focused and specialized for concurrent AI, physics and visual work. The rest are fixed functions. In Cell the devs have to keep the processors in sync using their own "glue". I expect the new, tighter pipeline to simplify synchronization and reduce overhead. I think most AI will still be handled by the CPU.

If the devs can code to the metal, I expect stunning results once they understand the h/w.

I suspect Sony may be prepped for shorter or even heterogenous console lifecycle this time round though. I wonder if their dual GPUs patent can be used for better software compatibility in these refreshes. The one that does JIT compiling when commands are issued into the queue. The newer box won't have 2 GPUs, perhaps the new/different GPU will take over the role/"queue" of the old one to run the dynamically recompiled commands.

 

[dobwal]

Are you for real dude.?

The PS3 cost $800+ dollars to manufacture far far more than the 360 cost to manufacture,if sony would have chase power with that kind of budget it would have out power the xbox 360 in a considerable way..

Let me put it this way erase Blu-ray,and add a Xenos alone side Cell with 512MB of DDR3 and 256 XDR for system ram.

Some of you may say it was impossible,but sony bom on the PS3 far exceed that of the 360 basically from every point but GPU.

Is the reason why sony loss so much how people can ignore this is crazy.

What's the point of putting gpu like processors on a CPU then have them work to bolster a gpu? Sony would of been better off sacrificing some spes and getting a more robust gpu.

One of the problem with the ps3 was its esoteric design.

 

[Averagejoe]

There is still nothing we know of Durango's CPU. I'm not sure how much they could customize it, but if there are additional vector units on it FLOP count could shrink (albeit nothing like ~400 GFLOPs, its 1.6Ghz CPU after all).

Based on specs GPU in Durango will be very well "fed", and by looking at 16 ROPs I guess they aim at 1080p 30fps. Sony aims at 1080p 60fps, but will be interesting what happens if devs opt for MS target since that would make 32 ROPs less of an advantage.

There is several things we don't know about either machine all is rumors and leaks.


I am going by this specs if MS has something else who knows,but is impossible that people who know hardware here look at this 2 set ups getting GPU from the same company one with higher specs than the other and expect them to perform the same.

Is not just 1080p 30 fPS vs 60 FPS,i think more physics calculations,better lighting and several other things could be at hand here.

But i am just going by specs i know that comparing hardware is not always apples to apples,but GPU talking this seems to be the case here.

7850 a little beef up vs 7770.

Hell even that Orbis is say to have a 7850,is not exactly so it has lower speed but more CU.

 

[Ruskie]

There is several things we don't know about either machine all is rumors and leaks.


I am going by this specs if MS has something else who knows,but is impossible that people who know hardware here look at this 2 set ups getting GPU from the same company one with higher specs than the other and expect them to perform the same.

Is not just 1080p 30 fPS vs 60 FPS,i think more physics calculations,better lighting and several other things could be at hand here.

But i am just going by specs i know that comparing hardware is not always apples to apples,but GPU talking this seems to be the case here.

7850 a little beef up vs 7770.

Hell even that Orbis is say to have a 7850,is not exactly so it has lower speed but more CU.

Yea, except Durango's GPU isn't exactly 7770. Its more powerful with significantly more bandwidth at disposal.

 

[Arwin]

12 vs 14 is trivial but you are forgetting something.

Orbis has 4 CU for compute jobs,if the game heavily use this on Orbis what will you use to emulate those on Durando.?

CPU + some of the 12 CU units.?

But that means you are using resources from those 12 CU and the CPU,while Orbis still has most of its CPU + 14 CU available.

Is the only reason i think there will not be parity if this specs are true,even less on exclusives.

I am in the camp that thinks we've simply still to hear anything about Durango's 'APU' component. But I could be wrong.

 

[Arwin]

2 more GB would not be any problems,that is Rangers opinion.

If sony want those 6GB that is what the PS4 will have,it will not break the back,considering the PS4 use CPU and GPU that are not that expensive,compare to Cell Jaguar is dirt cheap.

But i don't see what 6GB,hell 1GB more would do,most current top of the line GPU have 3 or 4 GB of ram the PS4 one is not even close to be high end GPU,so i think it will be ok.

Most GPUs are in computers that also have a lot of 'CPU' RAM, that is however still available for everything just not as fast.

 

[expletive]

I am in the camp that thinks we've simply still to hear anything about Durango's 'APU' component. But I could be wrong.

What would a 'beefier' CPU do to the bandwidth requirements of Durango? If Durango would have a more capable CPU to do the things the +4CU in orbis are purported to do, would that require additional bandwidth in the system or would it simply 'stay on the CPU' and not need to be moved around as much?

Given that 8GB seems like a primary requirement for Durango, it would make sense to have the compute power stay in the CPU to minimize bandwidth requirements since DDR3 was already challenged in that regard

 

[Prophecy2k]

What's the point of putting gpu like processors on a CPU then have them work to bolster a gpu? Sony would of been better off sacrificing some spes and getting a more robust gpu.

One of the problem with the ps3 was its esoteric design.

I would imagine Sony's intention was for Cell to enable stuff like better physics, AI, animations as well as stuff like Blu-Ray and fantastic audio (which it did to a large extent).

Cell was only relegated to doing alot of GPU stuff in many MP games because RSX struggled to keep up with the Xenos in MP game graphics workloads. On PS4 first party games, where games were designed around RSX and its weaknesses, CELL was able to do a lot more.

CELL wasn't merely designed to do just GPU helper stuff. It was just found that it was fast enough, and useful enough to do so in MP games this gen.

This time however, it looks as if Durango's GPU will be struggling to keep up with PS4's GPU alone (whether you're disingenuously counting 14CUs, or the full 18CUs). Unfortunately, Durango won't have a "Cell" to close that kind of performance gap. Even if it packed some modified Jaguar that doubled the flops rating over PS4's version, there's still some 410GFLOPs worth of CU rendering or compute (and that ignores the ROPs and TMU difference) on PS4 that would be impossible for Durango to match.

Pretty bad situation in my mind.

 

[Arwin]

What would a 'beefier' CPU do to the bandwidth requirements of Durango? If Durango would have a more capable CPU to do the things the +4CU in orbis are purported to do, would that require additional bandwidth in the system or would it simply 'stay on the CPU' and not need to be moved around as much?

Given that 8GB seems like a primary requirement for Durango, it would make sense to have the compute power stay in the CPU to minimize bandwidth requirements since DDR3 was already challenged in that regard

Well that's the tricky thing. We have no good (enough) overview of the dataflows in the system yet. Perhaps texture and vertex data streams from the DDR3 through some kind of Compute unit to the GPU and then into the VRAM in one pipeline that manages to minimise DDR3 access to great extent. Perhaps a secret OS component has access to DDR memory over a separate bus. We just haven't a clue (at least I don't).

Which is probably why for now we have to speculate based on the graphs we do have, and those suggest a certain number of components, and then we can compare these with what we know of Orbis. But everyone has to keep in mind that's all we can do - speculate based on very likely incomplete data, of which we're not even certain if it is 100% reliable or correct.