Cell on NVIDIA Graphics Cards?

[pelly]

Hey guys,

I saw this in the Yahoo Financial forums for NVIDIA.......I'd bet it is 99.9% garbage, but an interesting thought regardless.

http://finance.messages.yahoo.com/b...ard=15784668&tid=nvda&sid=15784668&mid=381696

"Rumor is that Nvidia and Sony will make an integrated video card with a Cell processor and Nvidia graphics chip combined on one card. Stay tuned. "


-A few SPE's and you'd have a strong physics processor on the card...any ideas?

 

[Arun]

Bullshit.
Either that guy is high on drugs and/or stupidity, or he's misinterpreting the speculated 65nm shrink of RSX+CELL as being a standalone product.

Uttar

 

[pelly]

I'm with you on that Uttar....lol.....

Regardless, this is one of the more amusing rumors I've heard in a long time.....

 

[AlNom]

I'd like to think he was just misinterpreting instead of just being stupid.

 

[karlotta]

well they have to use the "low" yield cell chips somewhere....think it might run the NVDA ,Netgateway and Firewall on the lil cell that could...

 

[rwolf]

Don't the SPE processors handle integer math only. I don't think that would be useful for floating point math calculations. I understand that Sony had to run to Nvidia because cell couldn't handle graphics processing. Might be useful to pull functions of the driver and have them run on the card itself though.

 

[KimB]

Don't the SPE processors handle integer math only. I don't think that would be useful for floating point math calculations. I understand that Sony had to run to Nvidia because cell couldn't handle graphics processing. Might be useful to pull functions of the driver and have them run on the card itself though.

No. The SPE's are primarily floating-point.
Linky

If I remember correctly, the PhysX actually is just a Cell processor.

 

[ShootMyMonkey]

If I remember correctly, the PhysX actually is just a Cell processor.

It's not a *Cell* specifically... just that AGEIA openly states an apparently large set of similarities between the two. The PhysX as is is a whole lot smaller, as there are a lot of things it doesn't have (e.g. cache)... Unless of course you were speaking metaphorically, in which case, I'll go in a corner and be silent.

Cell-like Global Structure of PPU

By the abstract of the PPU architecture, you'll immediately notice the commonality with Cell. Both of them are parallel processors with huge floating point processing units, have no cache hierarchy, and manage inter-memory data transfer by software programs.

If you replace PPU Control Engine (PCE), the RISC core in PPU, with PPE(Power Processor Element), the PowerPC core in Cell, and Vector Processing Engine (VPE), PPU's data processing engine, with SPE(Synergistic Processor Element), Cell's data processor, they almost correspond with each other.

In both the architectures one RISC core does global control and many vector data processors does data processing in parallel. As for the affinity in the architectures Hedge said:

"If you look at the very high level they are very alike. Both of them are huge parallel engines, have floating point processing units, and control each internal memory. But the difference is also big. For example, Cell does internal data transfer by a ring bus (so it has limited bandwidth). On the other hand, our architecture has far higher (internal data) bandwidth.

But it's also true that Cell is relatively suitable architecture for physics processing. In PS3, the GPU is "GeForce 7900+" architecture, but it has Cell. So in PS3 it can do physics in a PPU-like architecture (Cell), not in GPU."

Looking at the PPU architecture like this, you can imagine AGEIA has a relatively good affinity in PS3 library development.

 

[KimB]

It's not a *Cell* specifically... just that AGEIA openly states an apparently large set of similarities between the two. The PhysX as is is a whole lot smaller, as there are a lot of things it doesn't have (e.g. cache)... Unless of course you were speaking metaphorically, in which case, I'll go in a corner and be silent.

Well, actually I was speaking literally, but I wasn't sure on the point. I was sure it was at least similar to Cell.

Anyway, this is one big reason why I don't see hardware physics ever catching on. At the very least, Aegia has a long, hard uphill battle to get the cards out there. And by the time they have a chance to have any marketshare, I fully expect that we'll have CPU's that are rather similar to Cell, negating any need for it (to make good use of a Cell-like architecture, you need highly-parallizable code, and there isn't much code in games other than 3D graphics and physics that is very highly-parallelizable, except perhaps the AI, but in any case a Cell-like architecture would go largely unused in games without physics processing).

 

[Nick]

...I fully expect that we'll have CPU's that are rather similar to Cell, negating any need for it...

Actually that might already happen in two months when Core 2 Duo is released.

PhysX presumably runs at 500 MHz, and has 8 vector units each capable of 4+1 FP, resulting in 20 GFLOPS (AGEIA specifies 20 giga-instructions). Core 2 Duo ticking at 3.0 GHz has 2 complete SSE units each capable of 4 FP, per core, resulting in 48 GFLOPS. Even if you only consider one core available for physics processing that's 24 GFLOPS.

 

[KimB]

I don't think you can really consider SSE to be the same thing as additional separate processing elements.

 

[Nick]

I don't think you can really consider SSE to be the same thing as additional separate processing elements.

Sorry, what do you mean?

 

[KimB]

Sorry, what do you mean?

It's still ILP instead of TLP. Unless Intel changes the processor to allow as many threads per core as they have SSE units + scalar units, that massive parallelism isn't going to be used nearly as well as if they had multiple separate processors in real-world applications.

 

[Nick]

It's still ILP instead of TLP. Unless Intel changes the processor to allow as many threads per core as they have SSE units + scalar units, that massive parallelism isn't going to be used nearly as well as if they had multiple separate processors in real-world applications.

They have four decoders, three issue ports and two SSE units per core. Also, everything is fully pipelined and out-of-order (including loads). So it should be capable of keeping the SSE units busy most of the time.

Hyper-Threading would help indeed, and definitely 'inverse' Hyper-Threading, but for the time being I believe Core 2 Duo will be excellent for something like physics processing. Increasing TLP is the way forward but it's no panacea.

 

[caboosemoose]

Is it possible that Sony has some plans down the road for integrating Cell and RSX onto a single chip in the name of economy? I don't really see it as being relevant for PCs, but unless I'm mistaken, that is the approach taken later in the life cycle of many consoles - ie integrating more and more features onto a single chip.

 

[_xxx_]

that is the approach taken later in the life cycle of many consoles - ie integrating more and more features onto a single chip.

That goes for every single kind of electronic device

 

[KimB]

Is it possible that Sony has some plans down the road for integrating Cell and RSX onto a single chip in the name of economy? I don't really see it as being relevant for PCs, but unless I'm mistaken, that is the approach taken later in the life cycle of many consoles - ie integrating more and more features onto a single chip.

I doubt it, due to the memory interface. Currently each has its own memory.

 

[caboosemoose]

That goes for every single kind of electronic device

Well, I take your point, but would argue that it doesn't always apply. Intel took the P4 from 180nm to 65nm without integrating any significant extra functionality onto the die.

 

[Arwin]

Let's alter the question slightly. If you were Nvidia and you were looking to have a Physics processor on board, where would you look? There are a couple of things here:

- having the physics processor close to the gpu seems to me pretty important. The Ageia chip has trouble partly because it can't very efficiently interact with the graphics chip.

- RSX and Cell already work together on the PS3 so a fair bit of the hardware interfacing has already been done at the very least in a first version, which always helps a lot.

- Cell has potential to become very cheap as it is very much a mass product

- Cell will continue to be further developed with low power consumption in mind (which I expect to be a big issue)

- Cell looks like it has a bright future with IBM as well and is well supported by IBM

- Nvidia have good relations with Sony

 

[zed]

gpu's are becoming more like cpus every year (and visa-versa a bit)
in the future (perhaps even as soon as the next console generation) cpu's + gpu's will be combined into one uber-chip, thus im pretty sure both nvidia + sony (also others) are researching this avenue