SCEI & Toshiba unveil 65nm process with eDRAM

[BenSkywalker]

Faf-

Well GF1 needs a CPU to perform a great many things, so your comparing of chip alone to a system(you compared it to the whole PS2) is pointless.

Not when we are looking at rasterization between a CPU and a GPU. You can say that that comparison is pointless, but that is what we are discussing.

While on the subject, what magic copy of Giants did you run "well" on GF1 that even looked good if I may ask? The one I played runs like complete crap on anything below GF3 & 1+ghz cpu.

Run it under Win9X. Giants framerate drops ~50% and up under any NT based OS(the difference is roughly 100% under certain settings). No magic involved, rather common knowledge the game engine has never run well under 2K or XP

Even then, under Win2K@1024x768 with all details cranked 32bit(on the retail build patched version) I'm pushing just under 30FPS(just under 50FPS under Win98) with a GF2- both CPU limited on a GHZ Athlon. True if you drop the CPU down to ~700MHZ the framerate falls to around 35FPS under Win9x and 21FPS under Win2K. Given how the downgraded port runs on the PS2 I'd say it's closer to Win2K level performance when Win9X(choppy as hell on the PS2).

 

[Tagrineth]

And the PS2 can't run Giants without compromises, nor could it run JKII or Mafia(both without compromises of course) as a few examples. Obviously when coding you are going to take certain things into consideration on a fixed platform. The big difference is I can pull up examples that were not built from the ground up for the GF1 that won't run on the PS2.

To be fair, almost no PC ports are ever very optimised, with VERY few exceptions.

I bet if someone took up the cross and severely optimised JK2 or Mafia, they'd run brilliantly on PS2. Of course there would be the occasional trade-off, but barely any in the end.

 

[marconelly!]

And the PS2 can't run Giants without compromises, nor could it run JKII or Mafia(both without compromises of course) as a few examples.

You can't really consider quick & dirty PC ports made by people completely inexperienced with PS2 hardware and proclaim hardware incapable of running it. Does Mafia look much better than Getaway or Primal (both more or less free roaming games with no load times and lots of stuff going on the screen)?

How is Mafia and Giants working with GF1 in the first place? I know I'm not getting much more than 30ish FPS average in Giants on 1GHz + GF2 machine, all settings max.

My point is - I've never seen a PC game (running at more than 30FPS) on GF1 that is remotely as impressive as the best looking titles on PS2. Not when the GF1 launched, not now (that is several year later)

 

[BenSkywalker]

Tag-

I bet if someone took up the cross and severely optimised JK2 or Mafia, they'd run brilliantly on PS2. Of course there would be the occasional trade-off, but barely any in the end.

No, they wouldn't. You would have to seriously reduce the texture quality to get the games to the PS2 on either title(Mafia I don't see working at all). And remember, without compromise means trilinear and anisotropic filtering on

Marco-

. Does Mafia look much better than Getaway or Primal (both more or less free roaming games with no load times and lots of stuff going on the screen)?

Getaway and Primal compared to Mafia, different leagues(by a long shot). Mafia won't run on the XBox without downgraded graphics(not close to enough RAM for the textures used).

How is Mafia and Giants working with GF1 in the first place? I know I'm not getting much more than 30ish FPS average in Giants on 1GHz + GF2 machine, all settings max.

Run it under Win9X. Pair the GF1 with a 3GHZ P4 and you would likely be pushing close to tripple digits in Giants under Win9X running console resolutions.

My point is - I've never seen a PC game (running at more than 30FPS) on GF1 that is remotely as impressive as the best looking titles on PS2. Not when the GF1 launched, not now (that is several year later)

And as I stated, the PS2 has the enormous real world advantage of being a fixed platform.

 

[zurich]

On the flip side, isn't Square speccing a ~2ghz P4 w/a GF3 for the PC FF XI?

Same with GTA3, the game was unplayable on my 1Ghz Tualatin P3 & GF2. I know the PS2 version was sub 30fps, but the PC was sub 10fps. Doom 3 Alpha ran better.

 

[Fafalada]

Even then, under Win2K@1024x768 with all details cranked 32bit(on the retail build patched version) I'm pushing just under 30FPS(just under 50FPS under Win98) with a GF2- both CPU limited on a GHZ Athlon.

Actually that's about what I got on 2k, and similar setup. The problem is "just under 30fps" only holds as long as there's almost nothing going on the screen. Bring a few enemies, a bit of shooting, and fps started howering between 10-20, often lower. In the later levels really heavy with action I've seen sub 10fps quite often. But this was still on gf2...
(A dual CPU with a Quadro ran worse, so I won't even go there).

Anyway, the other problem is, that your assumption Giants is a rasterizer benchmark is probably quite far off the mark, at least when it comes to the PS2 port.
PC -> console ports are pretty much always CPU limited, even when target is XBox, which makes it practically by default with other two platforms.
This is all the more likely when you consider that majority of all PS2 games to date have been limited by r59k memory accesses, not by any of the specialized components (rasterizer/vu's etc.).
In other words, chances are rather good that Giants port on PS2 is mostly leaving the rasterizer you're trying to benchmark - idle.

Pair the GF1 with a 3GHZ P4 and you would likely be pushing close to tripple digits in Giants under Win9X running console resolutions.

Oh cool. I'll just compare it to 1ghz EE then.

 

[BenSkywalker]

Zurich-

Same with GTA3, the game was unplayable on my 1Ghz Tualatin P3 & GF2. I know the PS2 version was sub 30fps, but the PC was sub 10fps. Doom 3 Alpha ran better.

I'm getting ~50FPS running GTA3 with a GHZ Athlon and GF2(actually measured, not a guess) under Win2K, and that's at roughly quadruple the resolution of the PS2 version(have to try and crank it up to see if I can get it close to as slow as the PS2 version, haven't tried applying AA at high res yet).

I can't find anything on the specs for FFXI, do you have any links?

Faf-

Actually that's about what I got on 2k, and similar setup. The problem is "just under 30fps" only holds as long as there's almost nothing going on the screen. Bring a few enemies, a bit of shooting, and fps started howering between 10-20, often lower. In the later levels really heavy with action I've seen sub 10fps quite often. But this was still on gf2...

Giants has a benchmark built in to the latest patch. I am using benchmark averages, not standing still without enemy numbers

Anyway, the other problem is, that your assumption Giants is a rasterizer benchmark is probably quite far off the mark, at least when it comes to the PS2 port.

Dot3 removed, texture quality significantly reduced and bilinear filtering(it should be using trilinear and anisotropic)- that is what I mean by downgraded graphics(Dot3 I can see the CPU relevance, and that is a big downgrade in Giants, but not proper texture filtering techniques which based on this discussion should be handled in software if the GS can't cover it ).

PC -> console ports are pretty much always CPU limited

Yes, which backs up exactly what I've been saying. Relying on software for increasing amounts of rasterization you will be CPU limited even moreso then currently(although I know that your comment relates directly to PC ports).

In other words, chances are rather good that Giants port on PS2 is mostly leaving the rasterizer you're trying to benchmark - idle.

This discussion isn't about the GS. Any feature the GS is lacking has to be covered by the EE to back Vince's general line of thought. I'm saying that this is not viable, not now and almost certainly not with the PS3. Sony has to have a strong rasterizer to be able to compete.

 

[zurich]

Ben,

I read about it on a post on Tech-Report.com. Square released a benchmark, and only systems sporting 2000+ points are deemed fast enough for the game.

ftp://212.43.238.130/nvchips/pub/benchs/FFXiBench.zip

My rig got 1500 What is interesting, is that enabling 2xFSAA and even 4xFSAA did virtually nothing to the score, meaning that it's highly CPU limited...

 

[marconelly!]

Getaway and Primal compared to Mafia, different leagues(by a long shot).

How is Getaway in different league? Conceptually, it's almost exactly the same game, as Mafia as far as I can see.

Btw, I was running Giants under Win 98, that's what I had back then.

And as I stated, the PS2 has the enormous real world advantage of being a fixed platform.

Well, this has always been saving grace of PC users in dicussions like this. Perhaps we'll never see how much would GF1 be able to accomplish in reality, but even it's theoretical limits are not all that impressive.

 

[Vince]

This discussion isn't about the GS. Any feature the GS is lacking has to be covered by the EE to back Vince's general line of thought. I'm saying that this is not viable, not now and almost certainly not with the PS3. Sony has to have a strong rasterizer to be able to compete.

Ben, how many times now did I state I was comparing a 'general processor' like the EE - VU's specifically - to the TCL front-end on the nVxx's?

My point based on programmability > fixed function in preformance if it has one of those advantages was not based around the notion of a totally software defined scheme. P10 is an excellent example of the type of 'general computing' I speak of.

Common buddy, We even agreed that there will be some form of conventional hardware raster functionality. This full software talk was just to talk about the feasibility.

 

[Fafalada]

Giants has a benchmark built in to the latest patch. I am using benchmark averages, not standing still without enemy numbers

Imo average fps numbers don't really tell much about the game running at playable framerate or not, especially if benchmark is ran with v-sync disabled.

Yes, which backs up exactly what I've been saying. Relying on software for increasing amounts of rasterization you will be CPU limited even moreso then currently(although I know that your comment relates directly to PC ports).

I know what you're trying to say, but you're taking contexts all over the place. "software" does not equal "general purpose cpu" for one. :\
If I am to use DOT3 on GS, I will emulate it through GS - in software - yes, but not on the CPU.
If I need a complex pixel shader on NV3x to simulate surface of say, human skin, I will again, write Software - shader to extend the rasterizer function. It won't use CPU to rasterize either.

The other problem is you're trying to establish lacking features of a port(which won't be looking for ways to create new stuff) as a benchmark of what is possible on target hw. While stuff like anisotropic filtering in software is rather feasible at low performance overhead on the PS2, it takes TIME to implement.
Morevoer, whether the visual result of something like that is worth the implementation cost/headache alone is already arguable, before we even mention performance considerations.
That said, you're likely to see stuff like Dot3 eventually in PS2 games, but I have doubts anisotropic would be high on anyone's list to implement(Imo, at least).

Anyway, as you've said yourself "The next GS has to have a decent amount of feature support", is pretty much true. However, Imo it is irellevant if those features will be hardwired or in software - as long as they are available to developers to use from the get go.

 

[randycat99]

If I am to use DOT3 on GS, I will emulate it through GS - in software - yes, but not on the CPU.
If I need a complex pixel shader on NV3x to simulate surface of say, human skin, I will again, write Software - shader to extend the rasterizer function. It won't use CPU to rasterize either.

Is this part really true for the GS, or were you just speaking figuratively? It seems to be pounded into our heads (PS2 fans, collectively) that there exists absolutely no programmable function support in the GS, but are you saying that there is? This is certainly good news and bodes well for the future in PS2 development, but I just want to make sure I heard you right.

 

[V3]

It's not speculation. A TNT2 is roughly fifty times faster then a GHZ P3 rendering real time graphics(trilinear filtering etc, etc, not software compromised code). Given the P6 core's IPC edge it's going to take the P4 some time to catch up.

TNT1 ? The new P4 has a bandwidth advantage over TNT1. TNT2 was pretty slow at doing trilinear, I am not even sure TNT1 able to do it.

 

[BenSkywalker]

Marco-

How is Getaway in different league? Conceptually, it's almost exactly the same game, as Mafia as far as I can see.

Visually. The Getaway may be the game that actually gets me to purchase a PS2, but its visuals aren't even close to Mafia's.

Btw, I was running Giants under Win 98, that's what I had back then.

And it was that slow? It was much faster then that for me(via running a bench).

Well, this has always been saving grace of PC users in dicussions like this. Perhaps we'll never see how much would GF1 be able to accomplish in reality, but even it's theoretical limits are not all that impressive.

No, the theoretical limits of the GF1 aren't all that impressive. Neither are the PS2's.

Vince-

Common buddy, We even agreed that there will be some form of conventional hardware raster functionality. This full software talk was just to talk about the feasibility.

I thought that was the discussion we were having right now? I haven't argued that TnL functions can be handled by a CPU with proper considerations.

Faf-

I know what you're trying to say, but you're taking contexts all over the place. "software" does not equal "general purpose cpu" for one. :\
If I am to use DOT3 on GS, I will emulate it through GS - in software - yes, but not on the CPU.
If I need a complex pixel shader on NV3x to simulate surface of say, human skin, I will again, write Software - shader to extend the rasterizer function. It won't use CPU to rasterize either.

When you say not on the CPU, do you mean use the VUs? I'm assuming you do.

That said, you're likely to see stuff like Dot3 eventually in PS2 games, but I have doubts anisotropic would be high on anyone's list to implement(Imo, at least).

So anisotropic may not be a viable desire(given the overall weak texturing seen in PS2 games it wouldn't pay the dividends it does on the PC anyway), how about trilinear? Why isn't that pervasive in PS2 games?

Imo average fps numbers don't really tell much about the game running at playable framerate or not, especially if benchmark is ran with v-sync disabled.

I'm very confused by this statement, do you enable VSync when you play games on your PC.....?

 

[marconelly!]

Visually. The Getaway may be the game that actually gets me to purchase a PS2, but its visuals aren't even close to Mafia's.

Not even *close*? Well, compare these:

Mafia Inside the building:

Getaway inside the building (I'm using supersampled shots because I can't find Mafia in appropriate, comparable resolution of 640x):

Mafia outside:

Getaway outside:

I guess your definition of 'not even close' is very different than mine, then.

Also, consider that Primal has much better graphics (textures, effects) than what I've seen of Getaway. It's a linear game, but has a seamles world loading like the Getaway, GTA3, etc.

 

[Fafalada]

randy, it's math capabilities aren't the greatest, but duh yeah. Every rasterizer has programmable capacity to some extent.

I'm very confused by this statement, do you enable VSync when you play games on your PC.....?

V-Sync off can artifically inflate average FPS in any game that can run faster then V-Sync limit, even though those cases may be only when you stare at a wall. Not to mention in console title you don't have the option to do so, and most games never disable it, even when framerate drops, so comparing the two is not really possible.
Personally I can't be bothered to fiddle with settings much nowadays. When I want to play a game, there's nothing more annoying then having to push sliders around to get playable fps (just one of the reasons why my preferred genre on PC are strategy games).

When you say not on the CPU, do you mean use the VUs? I'm assuming you do.

Depending on the case. DOT3 is useless without T&L assist, regardless if you have the feature in hardware or not. The VU work involved with the GS emulation of DOT3 is pretty much nonexistant though - it boils down to doing what a VS already does for DOT3 normally, and then kicking slightly modified display list several times (it's a multipass process). The actual 'work' is being done by GS, which will also determine the speed limit.

So anisotropic may not be a viable desire(given the overall weak texturing seen in PS2 games it wouldn't pay the dividends it does on the PC anyway), how about trilinear? Why isn't that pervasive in PS2 games?

Well you need to use mipmapping first, if you want trilinear, something lots of PS2 titles avoided at first. Second reason is GS the texel cache scheme which can break performance with trilinear enabled if you're not carefull. On that thought, anistropic might actually work better in some cases, although I'd need to test that stuff out.

 

[randycat99]

Daaayamn, I'd have to say the much vaunted textures of PC Mafia are highly overrated. I don't think I'll be missing much if that is what uber PC textures are like...

 

[BenSkywalker]

Marco-

Despite the details being turned down in the GameSpot Mafia shots(very obvious in the inaccurate shadows in some screens and completely missing ones in others not to mention the significantly reduced texture quality in many of their shots), they still have some better shots then what you posted-

That is about what I would expect to see the XBox version look like. Downgraded from the PC version of course, but still not in the same realm as The Getaway.

Randy-

Daaayamn, I'd have to say the much vaunted textures of PC Mafia are highly overrated. I don't think I'll be missing much if that is what uber PC textures are like...

It's not, they have details turned down. Some of the shots that GameSpot has up all settings are set to their lowest possible(compressed, low detail, 16bit, low geometry settings, shadows off).

Faf-

V-Sync off can artifically inflate average FPS in any game that can run faster then V-Sync limit, even though those cases may be only when you stare at a wall.

If you have VSync on then you are pretty much always landing framerates quite a bit lower then you should, particularly running nV hardware(in game with heavy action too).

Not to mention in console title you don't have the option to do so, and most games never disable it, even when framerate drops, so comparing the two is not really possible.

If you are running a GF board you can force it off for all games, it pretty much always has a decent impact on noticeable framerate(not just in benches).

Personally I can't be bothered to fiddle with settings much nowadays. When I want to play a game, there's nothing more annoying then having to push sliders around to get playable fps (just one of the reasons why my preferred genre on PC are strategy games).

The curse and blessing of PC games You really should disable VSync though, you do it through the control panel twice(once for each API) or use RivaTuner if your driver rev doesn't have the option. Do it once and forget it.

Depending on the case. DOT3 is useless without T&L assist, regardless if you have the feature in hardware or not. The VU work involved with the GS emulation of DOT3 is pretty much nonexistant though - it boils down to doing what a VS already does for DOT3 normally, and then kicking slightly modified display list several times (it's a multipass process). The actual 'work' is being done by GS, which will also determine the speed limit.

So if you are clever coding and have the VUs handle the dot product calcs then the GS is capable of handling Dot3, OK.

Well you need to use mipmapping first, if you want trilinear, something lots of PS2 titles avoided at first. Second reason is GS the texel cache scheme which can break performance with trilinear enabled if you're not carefull. On that thought, anistropic might actually work better in some cases, although I'd need to test that stuff out.

I'm not quite getting the part I bolded. Are you saying that using ani in conjunction with trilinear would work better then tri alone, or are you saying that ani w/bi would work better then tri alone?

 

[randycat99]

Those textures do look better, but they still betray the premise of "uber PC textures". Still, they are blurry, and the polycounts aren't that hot, either. If it is just the choice between blurry textures and less blurry textures, then it's hardly a "world apart". If that is what you get with another $1300 in hardware, I got to wonder...

 

[BenSkywalker]

Those textures do look better, but they still betray the premise of "uber PC textures". Still, they are blurry, and the polycounts aren't that hot, either. If it is just the choice between blurry textures and less blurry textures, then it's hardly a "world apart". If that is what you get with another $1300 in hardware, I got to wonder...

The details are turned down, I could have sworn I mentioned that(there are numerous options for Mafia, by default the game will detect what you are running and choose an 'optimal' setup). Besides lowering texture details, it also reduces poly count, draw distance and shadows. Check out Quake3 with everything turned down to everything cranked to get a general idea of what kind of impact adjusting details has on a game.

As far as spending an additional $1300.. my rig would cost about a quarter of that to build(although I almost never do a complete rebuild, upgrade the parts I need when I need to), where do you shop?