Will Microsoft trump CELL by using Proximity Communication in XB720 CPU?

[Geo]

Then MS was really dumb. It is obvious that IBM could never deliver 3 OOOE cores running at 3.2 Ghz. They couldn't even get 1 OOOE core to go 2.5 Ghz on the same process. Even Intel and AMD couldn't get dual-core 3.2Ghz CPUs at the 90nm node. Unless they expected a much worse off chip, like 2 cores going at 2Ghz, what MS wanted was a pipedream.

Have you read Takahashi's book with extensive interviews with insiders all along the development process for XB360?

 

[Gubbi]

There's quite likely a serious floor to how "conservative" you can make an OoOE engine. In all OoO processors, you need some sort of instruction queue and a reorder buffer, both of which will need a minimum number of transistors and power usage. Plus much of the advantage of OoO is that you can make the CPU much wider than an in-order core, something that would be lost if you make the core that narrow. It's is quite reasonable that IBM decided that OoO is all but practically impossible for a triple core CPU going at 3.2Ghz.

OoO gives you a higher data dependency latency tolerance. That means that the average distance between dependent ops can be increased. Yes, this can enable a wide issue design, but it could just as well enable a deeply pipelined (and narrow) design.

Anyway, does it matter whether or not IBM could hit 3.2GHz with a narrow OoO, surely what is important is that they could a similar, or better, higher performance level with the same power and die budget ?

Look at PA Semi's new OoO PPC to see what *can* be done, when the right people get to work on it.

Cheers

 

[patsu]

Gubbi, would you be able to elaborate on PA Semi's OoO PPC design points for the uninitiated (like me !) ? I'm interested to learn more.

 

[Deusp]

Have you read Takahashi's book with extensive interviews with insiders all along the development process for XB360?

No. Is there something explained there that should change my opinion?

 

[Geo]

No. Is there something explained there that should change my opinion?

Well, not my style to tell you what it should do. "Might"? Dunno, you tell me.

But Spillinger said that the changes to the instruction set that Microsoft wanted – which meant changes in the cores that IBM had already created – meant that the entire design had to go through a complete verification from the ground up. Some of the instructions were not compliant with the PowerPC architecture and were owned by Microsoft itself. But IBM had the rights to the pieces of the chip that it was designing for Microsoft, and it had the right to use those again if it wanted.

“It’s truly a custom microprocessor,” Spillinger said.

On a day to day basis, Spillinger and his engineers stayed in touch with Jeff Andrews, the Microsoft CPU architect who worked in Mountain View. They argued about many things, but on a technical level. Given the legacy of battles between IBM and Microsoft, everyone knew that they had to be extra careful, and diplomatic, on this project. Microsoft stayed in touch with IBM every step of the way.

“We made the trade-offs together,” Spillinger said. “It started with communication between two teams, and then it expanded so that they talked to any of our engineers.”

A couple of the trade-offs were big ones. During 2003, IBM realized it had to scale back. Instead of hitting 3.5 gigahertz, IBM decided that it could only target 3.2 gigahertz speeds. (Sony had the same problem; it said its Cell chips would run at 4 gigahertz, but had to settle for 3.2 gigahertz). Otherwise, the yields on its chips might be too low, driving the costs up for both IBM and Microsoft.

Another setback was that IBM had also decided that it couldn’t do out-of-order execution. This was a modern technique that enabled microprocessors to dispatch a number of tasks in parallel. A sequence of instructions was broken into parallel paths without regard to order so they could be executed quickly, and then put back into the proper sequence upon completion.

Instead, IBM had to make the cores execute with the simpler, but more primitive, in-order execution. Out-of-order consumed more space on the chip, potentially driving up the costs and raising the risks. When Microsoft’s Jeff Andrews went to Jon Thomason and told him the news, it was like a bombshell. One by one, many of the Mountain View group’s biggest technological dreams were falling by the wayside.

“You always shoot for the best you can do, and then reality kicks in,” said Nick Baker. “You go through iterations and sometimes you get nasty surprises.”

“The schedule was a constant worry,” said Bob Feldstein, the ATI engineering vice president on the project. “2005 doesn’t move.

Interesting book, btw, and available as an e-book. It'd be cool to get that kind of soup-to-nuts insider story of PS3 development as well.

 

[Megadrive1988]

I borrowed his first Xbox book from the library and I bought the e-book version of Xbox 360 Uncloacked. both are exellent reads. I highly recommend both.

The Xbox had 21.6 gigaflops of computing power, or the equivalent of 58 Cray YMP supercomputers circa 1988.

so instead of the 80 GFLOPs often touted for the original Xbox (GPU),
we got a more realistic total FP figure in Opening The Xbox.

 

[pjbliverpool]

Then you're looking in the wrong place! We've seen plenty of scientific applications where the x86 or similar CPUs are eclipsed by Cell.

You misunderstood me, I was saying that Cell performs well in some scientific applications, its in actual games that it has yet to demonstrate its superiority in my view.

Folding@home, anyone? For games, there's obviously not much out yet. But in tech demos, Crysis doesn't strike me as the most impressive demo of physics so far. The Rubber Ducky E3 demo was more impressive.

To you perhaps, but can you explain why an x86 couldn't handle the duck demo? I have run some pretty impressive physics demo's on my old XP. I really don't see why a Core2 couldn't handle the duck demo.

The DMM from LucasArts too (we don't know what platform that was, and may well be PC from what's shown).

The fact that you point it out as an example of first class physics but can't tell if it was running on an x86 or not speaks volumes IMO.

We also have developers of physics engines telling us they run better on Cell.

Do we? I missed that but if we have then I would like to see who.. and why. Im not saying its not true, in fact I would expect it to be true but I don't recall anyone confirming it. I would love to hear one way or the other though.

And we have the Maya cloth demo. There's no reason to doubt Cell's abilities. You can doubt they'll make it into games, but you count discount the potential due to the existing demonstrations that have shown marked improvements possible on Cell.

Well, its the same answer as previous to that. Sure Cell is better in some ways, e.g. F@H, but I still have yet to se an example of PS3 gameplay, existing or upcoming that simply isn't possible on the PC or 360. By all means, point them out.

 

[Deusp]

Well, not my style to tell you what it should do. "Might"? Dunno, you tell me.

Sounds like you were gonna tell me.

Interesting book, btw, and available as an e-book. It'd be cool to get that kind of soup-to-nuts insider story of PS3 development as well.

It definitely would be. This book however, didn't give a date for when IBM told MS that OoO wasn't going to happen, so I assume it must have been fairly early on.

 

[Geo]

Sounds like you were gonna tell me.

Did it? I assumed that if "might" were a possibility you'd (or any reasonable person) would want to know about it to integrate new information into your decision making.

 

[Deusp]

Did it? I assumed that if "might" were a possibility you'd (or any reasonable person) would want to know about it to integrate new information into your decision making.

That's not necessarily the case. For example, sending a man to the moon is difficult and requires a big rocket. If someone disputes the need for a big rocket, I'll simply tell him otherwise. A book detailing the Apollo missions is not really needed for the discussion.

 

[Titanio]

but I still have yet to se an example of PS3 gameplay, existing or upcoming that simply isn't possible on the PC or 360. By all means, point them out.

Ohoho, come now, that's such a fuzzy point and you know it. The nature of gameplay is something that's still hotly debated even, something completely subjective, so that's a point you know you'll never ever get an answer on.

You'd do better to restrict your interest to algorithms that are applied in games and how they compare, of which some have been publically discussed and written up.

 

[Shifty Geezer]

To you perhaps, but can you explain why an x86 couldn't handle the duck demo? I have run some pretty impressive physics demo's on my old XP. I really don't see why a Core2 couldn't handle the duck demo.

Well then, we can compare impressive demos.

http://www.research.scea.com/research/pdfs/gdc2006_kokkevis_vangelis_physics.pdf (page 48)

Hundreds of rigid bodies on a fluid surface, piled high, running on one SPE at 60 fps. Even if you consider they were using bounding spheres rather than mesh collisions, that'd be equivalent to something like 500 (assuming 250 ducks. The presentation says 600 rigid bodies) spheres in a pile on water solved in 15 ms. I've never seen nor heard anything like that on PC. Maybe it exists and I just missed it (and if that's the case, what's the use of Aegia's PhysX?). Maybe you can find such physics on PC, and if you do find such a thing, maybe you can show that it could handle 4x as many ducks, seeing as the ducks physics was running on one SPE only.

The fact that you point it out as an example of first class physics but can't tell if it was running on an x86 or not speaks volumes IMO.

I wasn't pointing to DMM as a Cell case-study, but a better choice of comparing CPUs than Crysis. Crysis isn't the be-all-and-end-all of physics. If an x86 can handle DMM as well as Cell, that'd be the proof of equality, rather than a glob of bounding-box rigid-body particles.

Do we? I missed that but if we have then I would like to see who.. and why. Im not saying its not true, in fact I would expect it to be true but I don't recall anyone confirming it. I would love to hear one way or the other though.

Well I've more important things to do then go research all this! If you've missed all this discussion, you should have been paying attention!

Well, its the same answer as previous to that. Sure Cell is better in some ways, e.g. F@H, but I still have yet to se an example of PS3 gameplay, existing or upcoming that simply isn't possible on the PC or 360. By all means, point them out.

How can I point out games that show this? The games out there are what we have now, which we know aren't the limit of Cell's abilities. There's no argument for this. You are only willing to believe real-game situations (though point to CryEngine 2 demos...) and aren't willing to look at the other evidence. As an analogy, consider Woking FC and Accrington Stanely in the Nationwide Conference. You've seen Accrington play, and they've never lost a match. I say to you Woking will beat them. You say you don't believe it. I say Woking got investment from a Russian Billionaire. I give a list of the 22 world-class players on the team. I point to reports and articles about how well these players have played in their other clubs. You say 'yeah, but until I see them playing at Woking against Accrington, I won't believe they can beat them.'

The fact Cell has shown incredible performances in other fields is proof it can bring that to games. How can you flatly refuse to recognise that, as though the moment you get into a game situation, all those benefits are lost? Why is it Alias Wavefront's cloth simulator runs 8x faster on Cell than x86, but the moment we're in a game we're to believe Cell is no better than x86? At the end of the day, the Doubting Thomas will never be convinced until he sees for himself, so there's no point in this discussion. We'll just have to wait until games get to grips with Cell, and more importantly that comparative games come out on other platforms too. eg. We've no way of knowing if x86 can handle LBP or not until someone tries to recreate that game, and if no-one does, there's no comparison. And also those games will have to be doing on the CPU what Cell does. If they do exist but use the GPU's abilities, it's not going to be a fair comparison either.

 

[Neb]

Reminds me of the EE (Emotion Engine) for PS2 versus PC/xbox CPU. The EE was faster on practically all the benchmarks presented, especially graphical. Yet the xbox CPU delivered the best in almost all departments for games (GPU assisted with rendering).

You are only willing to believe real-game situations (though point to CryEngine 2 demos...)

Obviously you haven't seen Crysis real-time gameplay videos which has nearly/same effects/IQ as the CryEngine2 techdemos.

Crysis real-time, physics/effects, visual IQ.
http://www.incrysis.net/images/crysis_heliblast1.gif
http://www.incrysis.net/images/crysis_heliblast2.gif
http://www.incrysis.net/images/crysis_landslide.gif

We've no way of knowing if x86 can handle LBP or not until someone tries to recreate that game.

I believe a PC CPU wouldn't be able to run it fluid as they aren't designed to do graphics alone. But if you pair it with a GPU then I dont see a problem rendering LBP. And not to mention the jaggy edged shadows, no AA (blur filter), questionable physics/graphis for rope (bends like sticks)... no, lightning is what makes LBP shine.

 

[Shifty Geezer]

Obviously you haven't seen Crysis real-time gameplay videos which has nearly/same effects/IQ as the CryEngine2 techdemos.

Crysis real-time, physics/effects, visual IQ.

1) Dont link directly to animated .gifs, as per the forum rules.
2) Of the realtime demo's I've seen, framerate gets very choppy, without huge amounts of physics going on.
3) The whole Crysis thing doesn't help in any sort of argument on Cell's abilities and contributions to games. It's held as an example of the best physics to date. I argue it's not, because things like DMM are more impressive. Furthermore it's only physics in a game environment so isn't comparable to physics outside of games, which is mostly what we've seen on Cell. Lookijg to games to give comparable performances isn't going to say what vector-strong CPU designs like Cell are or are not able to contribute.

I believe a PC CPU wouldn't be able to run it fluid as they aren't designed to do graphics alone. But if you pair it with a GPU then I dont see a problem rendering LBP.

The whole question is about CPU's and where you get the best performance for consoles. This is NOT a PC versus PS3 thread. It's not asking what a PC can handle. It's considering real-world performance in games by the CPU. The OP is actually about what's good for XB3000, and from that we've moved onto whether multiple OoO monolithic cores are suitable, with people suggesting a big dual-core x86 is as capable as something like Cell. LBP wasn't presented as a test-case (PC's may well be able to produce a similar game) but only an example of where referencing a game as evidence of performance superiority fails, because to compare platforms by games, you need the same games with the same level of implementation.

 

[Geo]

That's not necessarily the case. For example, sending a man to the moon is difficult and requires a big rocket. If someone disputes the need for a big rocket, I'll simply tell him otherwise. A book detailing the Apollo missions is not really needed for the discussion.

Well, that's certainly the way it was done. If someone tells you there's a way to send a man to the moon without a big Apollo-style rocket, then you could, of course, shout him down without finding out what he has in mind. But then you might miss out on learning about a cool possibility of sending a man to the moon without a big Apollo-style rocket. But if you don't care, you don't care. Ta.

 

[Shifty Geezer]

Well, that's certainly the way it was done. If someone tells you there's a way to send a man to the moon without a big Apollo-style rocket, then you could, of course, shout him down without finding out what he has in mind. But then you might miss out on learning about a cool possibility of sending a man to the moon without a big Apollo-style rocket. But if you don't care, you don't care. Ta.

I've got a teleporter working here. I've already beamed a pot-plant and a dog to the moon, and a man from the top of Canary Wharf to Birmingham. I'd like to tell people about how it can beam a man to the moon, but some people just don't want to listen

 

[Geo]

I've got a teleporter working here. I've already beamed a pot-plant and a dog to the moon, and a man from the top of Canary Wharf to Birmingham. I'd like to tell people about how it can beam a man to the moon, but some people just don't want to listen

Ha. You really think a huge Apollo-style rocket is the only way to do it? Tsk, tsk. A lot of Apollos decisions were based around the top-down requirement that it be done "by the end of the decade". And then new ideas came later as well. But this is a digression, of course. The point being that one of the reasons you read these kind of books is to realize that what a lot of know-it-alls some years after the fact will treat as "historically inevitable" decisions, weren't actually. (Edit: Not that anyone cares, probably, and I didn't chose the example, but the concept I had in mind as the key enabler to a non-huge Apollo rocket method for getting people to the moon was first proposed in 1895. Tho it became more doable in 1957. And there are companies today still pursuing it. You'd still be on a rocket for some portion of the trip, but the rocket/fuel itself would be dainty compared to a Saturn V)

 

[pjbliverpool]

Well then, we can compare impressive demos.

http://www.research.scea.com/research/pdfs/gdc2006_kokkevis_vangelis_physics.pdf (page 48)

Hundreds of rigid bodies on a fluid surface, piled high, running on one SPE at 60 fps. Even if you consider they were using bounding spheres rather than mesh collisions, that'd be equivalent to something like 500 (assuming 250 ducks. The presentation says 600 rigid bodies) spheres in a pile on water solved in 15 ms. I've never seen nor heard anything like that on PC.

No doubt thats impressive and perhaps its not possible on a dual or even quad core x86. But the fact that we have never seen its equivilent doesn't really tell us much. Sony are deliberatly trying to market Cell as a processor that will enhance gaming so they have a very strong drive to produce demoes like that. Who has a similar drive in the PC space? Intel and AMD are the closest but they are targetting their processors at a much wider market and thus have less motivation to produce something like this. That said, Intel did produce that Ice demo thingy which I personally didn't find that impressive but on a technical level, how does it compare?

BTW, there were some interesting points in those slides to note. Slide 29 shows a demo that I was running on my XP years ago, obviously I don't know the context of its involvement in that presentation but its worth pointing out. Slide 44 shows a comparison of PPE to SPE performance with 4 SPE's only being about 3 times faster in the best case. Again, we don't know context and I would be suprised if far bigger gains can't be seen in other situations but given how the PPE compares to a modern x86 core, thats not particularly impressive. Also slide 47 states that there were early PC prototypes of some pieces of the of the duck demo. Who knows what performance was like or whether they were using one, or multiple cores but it does suggest thats its not completely outside the realms of possibility for an x86.

I would like to make it clear at this point though that im not trying to claim Cell wouldn't be better than a dual core in physics. I think it probably is, my argument is against this whole "order of magnatude" or more difference between x86 (almost regardless of how many cores are involved) and Cell. I think over the course of PS3's life we will see little, if anything that couldn't be done the same, or very similarly on a powerful, multi-core x86 based CPU. Whether the number of cores is 2, 4, or 8 though, im not quite sure yet.

Maybe it exists and I just missed it (and if that's the case, what's the use of Aegia's PhysX?).

I would ask that question regardless of this discussion I really don't think there is a point in all honesty. If HL2's physics can run just fine on a 733Mhz Celeron then a 3Ghz 4 or 8 core Peryn/Nehelam would seem to make the PPU a bit redundant if you ask me. Add to that I don't think we have seen the PPU do anything particularly impressive. The most impressive thing IMO was Cell factor and you can actually get that working almost as fast on a CPU only (single core I believe).

Maybe you can find such physics on PC, and if you do find such a thing, maybe you can show that it could handle 4x as many ducks, seeing as the ducks physics was running on one SPE only.

Based on the seperation of work shown in the slides the PS3 Cell could only handle twice as many ducks since 5 SPE's are already being used for other things (including the one reserved for the OS). So there is only one spare. Besides, the interaction of the ducks with each other is the least impressive thing about that demo IMO, as I said there are "interacting blocks" demo's on the PC aswell with at least as many units. The impressive thing to me was the water.

How can I point out games that show this? The games out there are what we have now, which we know aren't the limit of Cell's abilities. There's no argument for this. You are only willing to believe real-game situations (though point to CryEngine 2 demos...) and aren't willing to look at the other evidence. As an analogy, consider Woking FC and Accrington Stanely in the Nationwide Conference. You've seen Accrington play, and they've never lost a match. I say to you Woking will beat them. You say you don't believe it. I say Woking got investment from a Russian Billionaire. I give a list of the 22 world-class players on the team. I point to reports and articles about how well these players have played in their other clubs. You say 'yeah, but until I see them playing at Woking against Accrington, I won't believe they can beat them.'

I point to the CryEngine 2 gameplay demo because its actually in a game so its relevance is clear. What im saying isn't clear is the relevance of superior F@H performance, BFS on large sets and imaging performance to how much better games can be on Cell. Perhaps they are relevant and the previous comments about the BFS by Patsu suggest that they might be, on other hand if its already fast enough for gaming purposes then its not relevant.

Obviously Cell has a long way to go and we may yet see the evidence that im speaking of but all im saying is that so far, to me at least there is nothing out there that conclusvely proves Cell is well beyond a powerful multicore x86 as a CPU in a games machine.

The fact Cell has shown incredible performances in other fields is proof it can bring that to games. How can you flatly refuse to recognise that, as though the moment you get into a game situation, all those benefits are lost? Why is it Alias Wavefront's cloth simulator runs 8x faster on Cell than x86, but the moment we're in a game we're to believe Cell is no better than x86?

I actually hadn't seen that demonstration but it does sound like the kind of evidence im looking for, do you have a link? Assuming we are talking about relatively equal levels of optimisation and multithreading then that certainly does sound like compelling evidence.

 

[TheChefO]

...If HL2's physics can run just fine on a 733Mhz Celeron...

Great post overall but I thought this bit deserved a 2nd mention for contrast.

 

[function]

Great post overall but I thought this bit deserved a 2nd mention for contrast.

It's worth pointing out that HL2 actually calculates it's physics on a relatively small number of objects at once. It's the way the HL2 uses physics that's so cool, rather than the amount of physics work it's doing. That's where we get in to the real debate IMO, which is about how Cell (or anything else) is going to change the face of gaming.

Put simply, it isn't, but neither is a PPU. It's going to be great for eye candy though. No doubt you can do "more" of the kind of stuff that you get in sandbox games which rely heavily on stuff like physics, but that doesn't change the nature of the gameplay. Scale, like eye candy, is cool though. It just means you're never going to see a type of gameplay, or a game experience, that you can't do just about as well on an A64 X2.

For this reason, the Cell naysayers will never be proven "wrong" in terms of "show me a game you can do on Cell that you can't do on a PC" etc.

Don't forget that the clock is ticking on the Cell (as seen in the PS3) too. While the folks here may be calling for patience in seeing Cell's significant potential trasferred into games, the platform is on a countdown to obsolesence for which the idea of 'patience' doesn't really mean a great deal.