Xenos - RSX - What was left out?

[scooby_dooby]

The needed to compare because sony completely blew the FLOP count out of proportion.

They completely overplayed it's importance in terms of actual effect on total system power(according to Anandtech the numbers are next to meaninglesss)

You believe MS, despite knowing full well their system was much more powerful at GP processing should have not said anything? Why would they do that?

On the subject of FLOp count, I'll give you the 218GLOPS for CELL, but where did sony get 1.8tflops for RSX? Please correct me if I'm wrong, but if MS is pegging the Xenos at ~90GLOPS how the heck is the RSX 1.8?

Or is that a completely bogus number? And If you think it is, I'm sure you'll agree that was a low blow, and that MS was probably sent reeling. To honestly not expect them to come out with guns blazing on the subject of the PS3's lack of GP power is totally unrealistic.

 

[Titanio]

The needed to compare because sony completely blew the FLOP count out of proportion.

They completely overplayed it's importance in terms of actual effect on total system power(according to Anandtech the numbers are next to meaninglesss)

Did they? I don't remember Kutaragi pointing out ratios about CPU usage in games and the like. And it would still be questionable if they were blowing it out of proportion anyway. It's implicit in the importance they've placed on it in their design, but at E3 they simply said "look, this is our floating point performance, this is our competitors". And their numbers were transparently deriveable and all was well.

You believe MS, despite knowing full well their system was much more powerful at GP processing should have not said anything? Why would they do that?

How much more powerful..? See, that's the problem. And that's the problem with making a comparison like that and trying to wrap it up in a number.

On the subject of FLOp count, I'll give you the 218GLOPS for CELL, but where did sony get 1.8tflops for RSX? Please correct me if I'm wrong, but if MS is pegging the Xenos at ~90GLOPS how the heck is the RSX 1.8?

This is a different debate. Oh, and MS is pegging Xenos at 900Gflops.

To honestly not expect them to come out with guns blazing on the subject of the PS3's lack of GP power is totally unrealistic.

Qualify "a lack of GP power". What's a lack of it?

 

[blakjedi]

On the subject of FLOp count, I'll give you the 218GLOPS for CELL, but where did sony get 1.8tflops for RSX? Please correct me if I'm wrong, but if MS is pegging the Xenos at ~90GLOPS how the heck is the RSX 1.8?

Beyond the MS number being 900 gflops this is exactly the issue when comparing power of rsx to xenos

 

[scooby_dooby]

At Sony's E3 they said something to the effect of "FLOPS will be the defining measure of performance in next generation systems"

Bottom line is, they made it seem like having 2x the flops meant the system was twice as powerful.

Remember it's not like microsoft made an official comparison at all regarding actual GP performance. Allard only pointed out games ran mainly GP code and that PS3 had less GP power, he gave no numbers. Major Nelson's blog was unofficial, and as far as I'm concerned it's fighting fire with fire considering the FLOP crap sony pulled.

Granted the x360 has nowhere near 3 times the GP performance, even less of an advantage when you consider the SPE's contribution. At the same time, FLOPS is NOT the defining measurement of system performance and that's a total load.

So both companies are spreading FUD, but I think Sony made it clear at E3 through their FLOP crap, and CGI demo's that they want to play dirty. It's been game on ever since then.

 

[scooby_dooby]

On the subject of FLOp count, I'll give you the 218GLOPS for CELL, but where did sony get 1.8tflops for RSX? Please correct me if I'm wrong, but if MS is pegging the Xenos at ~90GLOPS how the heck is the RSX 1.8?

Beyond the MS number being 900 gflops this is exactly the issue when comparing power of rsx to xenos

sweet back on topic!

the 90 was a typo. so you mean RSX actually does do 1.8 tflops? does that mean anything? can we verify that? how do they come up with these numbers?

it seems to me like flops is a rating that can be manipulated very easily, given the fact MS released it's specs first it seems fishy to see a FLOP rating of double the Xenos, when Xenos is supposed to be a very powerful GPU.

shouldn't the total flop rating of both systems be within 10% of eachother? given the GPU's account for ~90% of this number?

 

[Titanio]

Remember it's not like microsoft made an official comparison at all regarding actual GP performance. Allard only pointed out games ran mainly GP code and that PS3 had less GP power, he gave no numbers.

He did actually. He harped on about 3x the "integer" or "GP" performance, the 80/20 ratio etc. His remarks were undoubtedly linked to the nelson blog, and it was as "official" a comparison as you're going to get.

Granted the x360 has nowhere near 3 times the GP performance, even less of an advantage when you consider the SPE's contribution. At the same time, FLOPS is NOT the defining measurement of system performance and that's a total load.

But it is a common metric used for quantifying chip power. MS's, on the other hand..

As for the importance of floating point power..well looking at Xenon I don't think they would actually disagree with Sony as much their PR might like you to believe

So both companies are spreading FUD, but I think Sony made it clear at E3 through their FLOP crap, and CGI demo's that they want to play dirty. It's been game on ever since then.

You're moving into different territory there, but MS wasn't averse to using CGI either.

WRT the GPUs, those numbers aren't like CPU flop figures..they're basically counting every last thing programmable or otherwise, and some cases "equivalents". I believe nao was keeping track of perhaps more accurate and comparable figures.

 

[Bobbler]

...
Major Nelson's blog was unofficial, and as far as I'm concerned it's fighting fire with fire considering the FLOP crap sony pulled.
...

Minor correction -- the article Major Nelson posted was from MS. MS sent it to various places to post (one of them happened to be Major Nelson). It was very much official and didn't originate from Major Nelson (it came from "engineers" at MS).

Regardless, you are correct that both companies are spreading FUD -- it is part of marketing. I don't really remember Sony claiming Flops were what mattered, but I'll take your word for it. Neither companies are saints in the marketing world.

 

[Neeyik]

Another thread that's borderline closure due to the level of off-topic discussion - folks it doesn't help people trying to get into a thread when so much of it has little to do with the actual thread title. Argue about FLOPS elsewhere please.

 

[MechanizedDeath]

The problem is that MS completely ignores the SPEs. If they attributed even a little GP performance to them, the comparison wouldn't have looked as impressive, but at least would have been closer to being realistic. Sony didn't short MS on the FLOPS front, but MS shorted Sony on the GP front by ignoring the SPEs. And the SPEs are the meat and potatoes of Cell.

It would be like Sony only counting one thread for Xenon in their comparison, and making the claim that 360 games will be single-threaded, so you can't really factor in the others. It's an unfair comparison. Not that the comparisons mean much, but really, if the SPEs can contribute to the GP performance of Cell (even if you have to move heaven and earth to do it), you can't just completely ignore them, can you?

This is off-topic, and all I have to say on this. PEACE.

 

[mckmas8808]

but MS shorted Sony on the GP front by ignoring the SPEs. And the SPEs are the meat and potatoes of Cell.

I wasn't aware that they weren't counting the SPE's.

 

[scooby_dooby]

Sony didn't short MS on their flop count, but the completely exaggerated their own FLOP numbers in regards to the 1.8tflops for the RSX.

Is it really any different?

 

[j^aws]

Sony didn't short MS on their flop count, but the completely exaggerated their own FLOP numbers in regards to the 1.8tflops for the RSX.

Is it really any different?

http://www.beyond3d.com/forum/viewtopic.php?t=23228

Neeyik was pretty clear. Also it's been discussed in this thread...

 

[j^aws]

That's the diplomatic answer!

I'd say on paper, the PS3 definitely has the raw power advantage but it seems it will be harder to leverage that than the X360...

Jaws, IYO, what role will tile based rendering play in this gen.

I see that Xenos uses some form of it to fit the frame buffer? (pardon my ignorance) into the Edram but is it going to be implemented similarly to how it was used in the Dreamcast?

Well they both use hardware tiling. PowerVR used small micro-tiles to fit into small on-chip buffers. Xenos uses very large tiles to fit into the 10MB eDRAM

IOW, will it be reducing the need for having all of the polygons rendered on screen by rendering only those visible to the eye and if so, does this reduce the number of shading operations for each scene as well?

They both use tiling but PowerVR is deferred rendering. Basically it discards all non-visible geometry before fragment shading so there's zero overdraw and only visible fragments are shaded so essentially it has no z-buffer. Immediate mode rendering GPUs including Xenos use various techniques to discard geometry early so that fragments aren't unnecessarily shaded for the final frame. But there will be overdraw unlike PowerVR. Xenos can use 'all' it's Shader ALUs to do a z-pass to make this process more efficient though it's not exclusive to Xenos. I could also see RSX doing something similar using the SPUs...

Does PS3 use a similar process or is this one of the "elegant" features of Xenos in order to make up for its lack of brute force compared to PS3?

G70 isn't a hardware tiler but there have been discussions of tiling using the RSX, i.e. the backbuffer to the SPE's local stores (and they are not caches but SRAM) and also splitting the colour and z buffers into different memory pools. The NV patent below also discusses, independent of GPUs being tile based, double-buffering and tiling by integrating logic to memory interfaces,

http://www.beyond3d.com/forum/viewtopic.php?p=563702#563702

So there are options for CELL-RSX and would definitely like to see more official info...but I do agree, it's an 'elegant' Xenos feature.

it sure sounds like most everyone here is concluding for now that both consoles will be pretty much equal in terms of power and output. they just go about it in very different ways

That's the diplomatic answer!

I'd say on paper, the PS3 definitely has the raw power advantage but it seems it will be harder to leverage that than the X360...

paper doesn't mean much, what matters most is final output to me.

Yep, and so it should and as long as your satisfied that's all that matters. Though this is a technical forum and you'll get plenty of those discussions here!

 

[Tap In]

good info there Jaws

thanks

I've really been wondering why tiling wasn't getting as much attention as I thought it deserved in light of all the bandwidth and shader power discussions.