GTX512/RSX analysis

[Carl B]

Well, I kind of did not want to be the one to make a thread like this, but then again it is something I want to discuss. We've been going over it to some extent in the 'PS3 spec upgrade' thread but I'd like to isolate focus onto the RSX alone, and what some recent developments with the G70 might mean for it.

I wanted to post some insights last night to get discussion rolling but thought I might as well take the time to work it into a full article, which I have since done. Anyway to keep the discussion to the things in which I have the most interest, here are the relevent RSX 'extrapolation' parts, rather than re-posting:

...First and foremost, for anyone that has been wondering whether or not it would be possible for Sony to reach it's stated goal of 550 MHz with RSX, this release should put those fears firmly to rest. Not only has NVidia been able to do it with a 110nm chip (RSX will be 90nm), they have done it without any of the advanced low-k and silicon-on-insulator (SOI) technologies Sony will be using. In fact since the first leaks of the GTX512's performance began to emerge a week ago, the tenor of the discussion has changed noticeably from wondering whether Sony would meet the 550 MHz clock in the first place to wondering how much they might exceed it by.

Unfortunately, it's not as simple as that. Remember that although PC cards can have relatively large cooling solutions attached and benefit from operation within the spacious confines (comparatively) of a PC, the heat and power restrictions placed on RSX will be much more severe. There is no doubt at this point that RSX could be clocked well beyond 550 MHz, but that doesn't mean it will be. Low-k and SOI will reduce power requirements (and thus heat) by a good bit, as will the move to a 90nm process, but remember applied voltage will play a very large role in determining just how 'hot' the chip will run. With a stated goal of 550 MHz and an architectural base with seeming room for speed, Sony will certainly have some options open to them. At the end of the day though, expect them to clock the RSX at the highest clock achievable at the same voltage at which they can comfortably reach 550 MHz. Maybe this means 550, maybe it means 560, maybe it means 600 MHz - who knows? Of course it can work in reverse, too. If Sony finds that going from 530 or 540 MHz to 550 is a full step in voltage, they may just opt to clock it a couple of MHz lower and save on power and heat. Indeed due to yield issues with the original EmotionEngine, Kutaragi launched the PS2 with a larger power supply and CPU voltage than he had originally desired, something which let's us know that such concerns feature prominently in SCE's thinking.

Increased clockspeeds aren't the only thing that we can be hopeful of receiving in terms of PS3 upgrades though. With ATI's new x1800 cards and the new GTX512, we are seeing a new class of GDDR3 memory entering the fray. 1.1ns RAM modules from Samsung and others have the potential to reliably reach speeds of 1800 MHz (DDR) and are being featured prominently in all of these new cards. Such memory was not available at the time of the XBox 360 and Sony PS3 E3 announcements, when the best memory available was indeed the GDDR3 memory they announced with. With it's later launch, Sony now has the opportunity to upgrade the class of GDDR3 it is using in the PS3 should it choose to do so. At what we can only assume would be a nominal increased cost over the memory presently slated to be used, RSX would gain access to greater memory bandwidth, something that has been seen by some as a potential bottleneck. This is not to say that they will of course, but the option is there where it wasn't before.

To end, some of the most exciting possibilities concerning the RSX don't come from analysis of the G70 itself at all, but from talk of certain changes the 90nm revision of G70 is rumored to feature. In addition to theorizing that the 90nm revision of the chip will sport a full eight pixel quads (32 pixel pipes) and 10 vertex shaders, Josh Walrath at Penstar Systems indicates that NVidia may be reworking the anti-aliasing unit in the 90nm revision in order to allow for multi-sample anti-aliasing and HDR lighting to be implemented concurrently. There is also talk that texture filtering may be raised back up to pre-NV40 series levels. If true, we might expect any or all of these architectural improvements to find their way into the RSX - something that all those anticipating the PS3 should find fairly exciting.

At the moment all of the above is all speculation. But with G70 our only tangible hardware lead into what RSX may eventually turn out to be, at the very least it is informed speculation.

Earlier this year Sony announced that final dev kits will begin shipping to PS3 developers this December, and that those dev kits will include functional RSX chips. It's hard right now in this information void we find ourselves to say whether those dev kits are getting set to ship or not - indeed we don't even have full confirmation that RSX has even taped out - but perhaps once those development kits do start shipping, we will gain a little more insight into just what exactly the RSX is...

Full article

Anyway what are people's thoughts on how RSX might be shaping up? It's obviously the case that GTX512's launch has not provided us any hard information on RSX, but at the same time I feel it has somewhat shifted the expectations for the chip. Has anyone else heard about a possible AA and texture reworking for the G7x 90nm revision? And if such a change were in the works, would there be any reason to doubt that RSX would benefit from these tweaks?

PS - Though I wouldn't normally use those symbol options, decided to stick a 'Playstation' symbol onto the post since there's a lot of XBox symbols floating around nowadays.

 

[jvd]

...First and foremost, for anyone that has been wondering whether or not it would be possible for Sony to reach it's stated goal of 550 MHz with RSX, this release should put those fears firmly to rest. Not only has NVidia been able to do it with a 110nm chip (RSX will be 90nm), they have done it without any of the advanced low-k and silicon-on-insulator (SOI) technologies Sony will be using. In fact since the first leaks of the GTX512's performance began to emerge a week ago, the tenor of the discussion has changed noticeably from wondering whether Sony would meet the 550 MHz clock in the first place to wondering how much they might exceed it by.

the gtx 512 has no bearing on the rsx .


The gtx has its own cooler and i believe its a 2 slot cooler . It also consumes alot of power . Alot of power and dishes out a lot of heat .

These things aren't looked for in a console part.


The console part while on 90nm wont save it from being a power hog and a heat monster . You can see by the x1800xt that they can still use alot of power and produce alot of heat at high clock speeds .

The ps3 cooling capacity will be much reduced over what the pc cooling factor offers .


Even if sony can produce a chip faster than 550mhz (which i have no doubt they can) power and cooling requirements will keep it at the 550mhz speed.

Just like we heard ati was getting higher speeds on the xenos . They still went with 500mhz as it fit the cooling parameters of the system

 

[Carl B]

Yeah no doubt JVD, I mean you read what was written, right? Because that's addressed. But the point is before, some people were thinking there was just no way that even on 90nm RSX could reach 550 MHz. Well, that thought has since been dispelled. The question now becomes simply one of heat and power, as stated. If you read beyond the first paragraph, you'll see that.

 

[expletive]

Why is this article any different than any of the other speculative stuff we've already seen a hundred times? Is there new information available?

 

[Carl B]

Why is this article any different than any of the other speculative stuff we've already seen a hundred times? Is there new information available?

I would consider the AA and texture information new, but maybe that's just me. Anyway if you don't have anything to add Expletive, by all means don't.

 

[Shifty Geezer]

It be nice to hear where these rumours of an updated G70 on 90nm come from and whether they should be trusted more than other rumours.

 

[Titanio]

Why is this article any different than any of the other speculative stuff we've already seen a hundred times?

It's xbdestroya's own commentary from another thread.

As tempting as it is to consider, I think architectural differences at 90nm revisions are unlikely. The main benefit of 90nm will be being able to put a 550Mhx RSX into PS3, making that feasible, IMO

edit - didn't see the Penstar report/rumours, but grain of salt. We also have to constantly remember nvidia's next high-end needs, and those or rsx, may not coincide entirely!

 

[Carl B]

@Shifty: Since it was just a straight copy/paste, the link didn't carry. But I've since added it to the post. I highly encourage anyone who has not yet read the entirety of Walrath's 'origins of GTX512' article to do so, but that is where the 90nm revision news comes from.

 

[jvd]

Yeah no doubt JVD, I mean you read what was written, right? Because that's addressed. But the point is before, some people were thinking there was just no way that even on 90nm RSX could reach 550 MHz. Well, that thought has since been dispelled. The question now becomes simply one of heat and power, as stated. If you read beyond the first paragraph, you'll see that.

it doesn't really talk about how serious these issues are . Heat and power doesn't go away with just a micron drop , as I said the x1800xt is running very hot at 90nm . Don't forget that even if its just am odified g70 there are still changes like fp32 and a new memory controller and yields on 7 month old product may not reflect what a modified chip on a new micron process with much smaller cooling capacity and power draw will produce.

It also doesn't factor in that the diffrence between these two chips are most likely less than 6 months (march 2006 launch) and the gtx 512 meg chips are said to be cherry picked gtxs not a new spin and are produced in very low quanitys (compared to what would be needed for a ps3 launch)




As for the ram. Once again this is ram on a 600$ video card . It has a whole diffrent pricing structure and on the gtx and x1800xt ram is most likely the most expensive part of the board (Even above that of the chip which is expensive) I don't see sony going with that .



Its all nice to think sony can upgrade the ps3 and i'm sure they could. But why ? They already have the spec advantage . What else do they need ? How much would a 600mhz rsx help ? How much wuold it cost ? How would it affect avalible units ?

How much would 800mhz gdr help ? Cost ? How avalible would it be ?

I'd be surprised if there are more than 100k 7800gtx 512 meg cards made before the ps3 comes out vs the 4-5 million ps3s made in the first year of its release (if not more)

 

[Guden Oden]

How can an analysis be made of a product whose specifics haven't been announced? Just seems to be a particulary large waste of time guessing about unknown hardware...

 

[Carl B]

it doesn't really talk about how serious these issues are . Heat and power doesn't go away with just a micron drop , as I said the x1800xt is running very hot at 90nm . Don't forget that even if its just am odified g70 there are still changes like fp32 and a new memory controller and yields on 7 month old product may not reflect what a modified chip on a new micron process with much smaller cooling capacity and power draw will produce.

JVD frankly I think I address the power issue as best as anyone has on the subject thus far; certainly I've been the only one to really bring the talk of voltage and indeed the limits on what Sony might clock the chip at up in our other discussions on the topic. A process shrink 110nm to 90nm, everthing else equal *should* indeed reduce heat and power - not to mention that low-k and SOI will reduce it a *lot*. You can't point to x1800 and say that 'you see, 90nm does nothing,' when x1800 is a transistor and clockspeed monster.

As for the ram. Once again this is ram on a 600$ video card . It has a whole diffrent pricing structure and on the gtx and x1800xt ram is most likely the most expensive part of the board (Even above that of the chip which is expensive) I don't see sony going with that .

The RAM before was for a $600 card also - what's your point? This new RAM probably occupies the price slot the former RAM did, with the former having dropped in price. It's stated as something that could happen, not something that will.

Its all nice to think sony can upgrade the ps3 and i'm sure they could. But why ? They already have the spec advantage . What else do they need ? How much would a 600mhz rsx help ? How much wuold it cost ? How would it affect available units ?

Well in theory it would help devs be able to achieve more, which could spread to stronger word of mouth later on in the gen. Certainly XBox benefited later into this gen from a perception of improved graphical prowess over it's competitors; why would Sony choose not to pursue a similar path if the pain would only be in the short-term?

 

[Carl B]

How can an analysis be made of a product whose specifics haven't been announced? Just seems to be a particulary large waste of time guessing about unknown hardware...

Guden if that were the case, why would any of us have been posting on these consoles over the last several months/years?

Ah well, I think the 'analysis' makes sense, even if ya don't.

* RSX: no doubt now able to reach 550

* Will it?: Depends on power and heat considerations

* GDDR3 class available to Sony that wasn't before - not to say they'll use it

* Rumors of AA and texture tweaking, pipeline increases on it's sister chip

Not saying you have to 'buy into' it or not Guden; indeed there's nothing I'm trying to sell. Rather it's simply the presentation of some facts and qualified speculation in a 'food for thought' manner. Certainly no one is twisting your arm on any sort of RSX religion here.

 

[expletive]

I would consider the AA and texture information new, but maybe that's just me. Anyway if you don't have anything to add Expletive, by all means don't.

It seemed like this was another 'please speculate based on someone else's speculation' thread, and while i'm all for those, i just wanted to make sure i wasnt missing anything more concrete. No offense intended, i enjoy reading these threads as much as you enjoy starting them.

 

[jvd]

JVD frankly I think I address the power issue as best as anyone has on the subject thus far; certainly I've been the only one to really bring the talk of voltage and indeed the limits on what Sony might clock the chip at up in our other discussions on the topic. A process shrink 110nm to 90nm, everthing else equal *should* indeed reduce heat and power - not to mention that low-k and SOI will reduce it a *lot*. You can't point to x1800 and say that 'you see, 90nm does nothing,' when x1800 is a transistor and clockspeed monster.

All things are not equal , the g70 and rsx are diffrent , there are more diffrences than you'd think , memory controllers being a huge one , then ofcourse fp32 hdr logic .

Woh knows what else has been removed or added the x1800 is an example of a chip running very hot on 90nm

The RAM before was for a $600 card also - what's your point? This new RAM probably occupies the price slot the former RAM did, with the former having dropped in price. It's stated as something that could happen, not something that will.

The gddr 700mhz ram or 1400 effective is older ram , for introduction of a product almost a year after the introduction of the ram. The price would have scaled much better than low volume gddr 800 or higher which has just come out for a system that has to go into production in as little as 4 or so months for a launch in spring 2006

It could happen and at the same time pigs can fly.

Well in theory it would help devs be able to achieve more, which could spread to stronger word of mouth later on in the gen. Certainly XBox benefited later into this gen from a perception of improved graphical prowess over it's competitors; why would Sony choose not to pursue a similar path if the pain would only be in the short-term?

Xbox had more to it than just faster ram or clock speeds. It had a completely diffrent generation of 3d hardware. The gs in the ps2 was a fillrate beast but had the feature set of a voodoo1 or 2 . The nv20a in the xbox was based on dx 8 tech . There was a huge diffrence in feature sets that really gave it the graphical edge.

Another 50mhz over the clock speed or even a 100 is not going to make huge performance increases . Esp if only clock speed or ram speed go up. You would need to make a few changes for you to really see any benfit .

Just increasing clock speeds will give u a smaller gain than the actually clock speed raise. Same with ram.

Now increasing both is going to icnrease both costs and also with a faster gpu and faster ram your heat is going to go up .

It also doesn't translate into better games and sony already has the spec hype wrapped up .

So what do they gain besides a more expensive unit .

 

[Carl B]

It seemed like this was another 'please speculate based on someone else's speculation' thread, and while i'm all for those, i just wanted to make sure i wasnt missing anything more concrete. No offense intended, i enjoy reading these threads as much as you enjoy starting them.

LOL, well I wanted to start it yesterday, but like I said I ended up writing an article on it instead. By all means I'm not asking for punches to be pulled - and worry not no offense was taken - I'm certainly happy to debate my points of view all day long (as many know) so certainly I wouldn't have posted this in the first place if I didn't think it'd result in at least some contention of views.

In a way it *is* a 'please speculate on someone else's speculation' thread, except that in this case that other person is me, so it's ok.

 

[jvd]

* GDDR3 class available to Sony that wasn't before - not to say they'll use it

sony would have had gddr product maps . HTey would have known the projected prices and ram speeds well into 2007 .

They didn't just say oh wow gddr 700 ? wow sounds fast lets just stick that in .

They did reserch into the scaling of price , quanitys avalible and of course all this in comparison to other speeds that would be out in thier time frame .

 

[Shifty Geezer]

Its all nice to think sony can upgrade the ps3 and i'm sure they could. But why ? They already have the spec advantage . What else do they need ? How much would a 600mhz rsx help ? How much wuold it cost ? How would it affect avalible units ?

I felt the potential for a clock upgrade was the least concern in XBD's post. The majority of text posted was about how clocks and yields were progressing, just as evidence that 550 MHz is possible (was it ever in doubt?). But XBD's (having pointed out the inhibiting factors of a faster component quite amply IMO) actual question was

...but at the same time I feel it has somewhat shifted the expectations for the chip. Has anyone else heard about a possible AA and texture reworking for the G7x 90nm revision? And if such a change were in the works, would there be any reason to doubt that RSX would benefit from these tweaks?

Specifically, news on a variant 90nm G7x part with extra features and it's relation to RSX.

@xboxdestroya : I'll check the article when I have time.

 

[Carl B]

All things are not equal , the g70 and rsx are diffrent , there are more diffrences than you'd think , memory controllers being a huge one , then ofcourse fp32 hdr logic .

I mean, haven't you until now been heavily in the 'just a modified G70' camp? I seriously doubt the heat and power profile will be insanely different than the NVidia low-k 90nm revision of the G70 architecture. I would expect a fair bit lower actually due to SOI. And yes I expect relative power savings on this process over the 110nm process. I think that makes sense. x1800 has nothing to do with anything; certainly G70 is the *most* similar architecture to compare it to.

The gddr 700mhz ram or 1400 effective is older ram , for introduction of a product almost a year after the introduction of the ram. The price would have scaled much better than low volume gddr 800 or higher which has just come out for a system that has to go into production in as little as 4 or so months for a launch in spring 2006

It could happen and at the same time pigs can fly.

And I could have a smoothie for lunch - any and all of these things might happen. Personally, I give the RAM speed increase a higher chance than the pigs. Price scaling though is exactly the reason Sony might want to consider it though, as the pricing of such modules would fall off rather quickly to the point of essentially being the same as the RAM they were going to use before a year or so into it.

Xbox had more to it than just faster ram or clock speeds. It had a completely diffrent generation of 3d hardware. The gs in the ps2 was a fillrate beast but had the feature set of a voodoo1 or 2 . The nv20a in the xbox was based on dx 8 tech . There was a huge diffrence in feature sets that really gave it the graphical edge.

Another 50mhz over the clock speed or even a 100 is not going to make huge performance increases . Esp if only clock speed or ram speed go up. You would need to make a few changes for you to really see any benfit .

You lose me here. But Xenos and G70 as it is already are essentially from different gens. THe RSX from this gens 'halfway point' and the Xenos from some sort of Twighlight Zone, sent from the future.

Just increasing clock speeds will give u a smaller gain than the actually clock speed raise. Same with ram.

What the hell? Does this sentence not contradict itself completely?

Now increasing both is going to icnrease both costs and also with a faster gpu and faster ram your heat is going to go up .

It also doesn't translate into better games and sony already has the spec hype wrapped up .

So what do they gain besides a more expensive unit .

Like I've said elsewhere, at least on the GPU side faster doesn't mean more expensive, as long as the processor is for the most part comfortable across the majority of chips reaching a certain speed at a certain voltage. If basically all the chips that can reach 550MHz on say, 1 volt, could reach 570 as well, why would Sony *not* go for 570?

 

[Carl B]

@xboxdestroya : I'll check the article when I have time.

Indeed Shifty check out. I link to the last page of that Penstar article, but the whole thing is quite interesting, though the rest deals more on the struggle of NVidia to position it's 7800 line of cards aganst what they thought would be a much sooner launched R520.

 

[jvd]

I mean, haven't you until now been heavily in the 'just a modified G70' camp? I seriously doubt the heat and power profile will be insanely different than the NVidia low-k 90nm revision of the G70 architecture. I would expect a fair bit lower actually due to SOI. And yes I expect relative power savings on this process over the 110nm process. I think that makes sense. x1800 has nothing to do with anything; certainly G70 is the *most* similar architecture to compare it to.

I allways felt it was a modified g70 but that doesn't mean its just a g70 chip. As i said there are some new featuers and of course new memory controllers along with other specific changes that sony wnated. I don't have acess to what they have.

110nm may give process savings if it was jsut a non modified g70. HOwever it isn't .

The x1800 is to show that power and heat requirements can go up . The g70 isn't whats being talked about the rsx is and since we don't fully know what it is we can't just say oh 90nm will make it run cooler.

Aside from that even if we say well its close enough , how much cooler and how much power will actually be saved ?

Will it be enough to matter .

And I could have a smoothie for lunch - any and all of these things might happen. Personally, I give the RAM speed increase a higher chance than the pigs. Price scaling though is exactly the reason Sony might want to consider it though, as the pricing of such modules would fall off rather quickly to the point of essentially being the same as the RAM they were going to use before a year or so into it.

You just give no plausible reason for sony to increase the price of thier console for limited gains . Its a ****** dream that makes very very little sense to sony.

Why do you believe they are going to invest more money and move to rarer ram speeds ? What does it gain them that they don't already have? They already have the spec advantage . So what else can more powerfull hardware give them ?

You lose me here. But Xenos and G70 as it is already are essentially from different gens. THe RSX from this gens 'halfway point' and the Xenos from some sort of Twighlight Zone, sent from the future.

Both the rsx and xenos are from the same dx 9 generation they just take diffrent routes to reach the end goal which is dx 9 platform for the console .

What the hell? Does this sentence not contradict itself completely?




Like I've said elsewhere, at least on the GPU side faster doesn't mean more expensive, as long as the processor is for the most part comfortable across the majority of chips reaching a certain speed at a certain voltage. If basically all the chips that can reach 550MHz on say, 1 volt, could reach 570 as well, why would Sony *not* go for 570?

You have to first prove that

1) the chips don't run hotter that = higher costs 2) the chips don't take more voltage = more cost 3 ) the chips have the same yield rates at higher speeds

Also just because they run at the same voltage doesn't mean they will produce the same heat.

My athlon 64 300+ runs at 1.5 volts at 1.8 ghz and at 2.4ghz however the cooling requirements are much greater at 2.4 ghz and so is the power draw even though it runs at 1.5 volts at both speed.

You also don't explain what the gain is gonig from 550mhz to 570mhz and the reason why is its not a big gain at all. Not in performance and not in marketing specs .