Can the RSX framebuffer be split between GDDR3 and XDR RAM?

[Crossbar]

This article seems to indicate that:
http://psinext.e-mpire.com/index.ph...99&PHPSESSID=784800d1964413bf17a3fe11f1c80bc2

PSINext: As previously discussed, beyond it's high quality one of the primary reasons for the use of NAO32 is that it saves bandwidth in a bandwidth-hungry environment. In the future do you feel RSX will be at a disadvantage to Xenos when it comes to framebuffer effects due to the 128-bit bus and lack of eDRAM?

Marco: Not at all; in fact for many framebuffer effects I believe RSX will have an edge over Xenos. Don't want to go into details, but let me just point out that RSX is connected to two seperate busses, not just one.

 

[sonyps35]

Excellent article here

PSINext: So NAO32 is a means for you to preserve memory while at the same time retaining image quality, is that correct?

Marco: Correct. The main idea behind NAO32 is that we want to trade shading power to regain memory space and bandwidth (very precious resources on a console). So instead of encoding our HDR colors into a FP16 or FP32 frame buffer, we devised a scheme to use RSX pixel shading units to convert an RGB color in a CIE Luv color that only requires a common RGBA8 frame buffer (4 bytes per pixel, half the space of a FP16 pixel) to be fully stored.

The quality of this format is really outstanding. Even if it uses half the space/bandwidth of common HDR rendering solutions, it really makes no compromises at all in image quality.

There's no magic here: HDR rendering costs are shifted from memory to shaders, and so our shaders are a bit longer now (between 3 and 5 cycles). We believe it's a very good trade-off. Furthermore, it enables HDR rendering and multisample anti-aliasing on GPUs that do not natively support AA with floating point render targets such as FP16 and FP32.

We also developed a faster 3 bytes per pixel version called NAO24 (predictable, isn't it?) that supports a narrower dynamic range with less accuracy. And although the quality was quite decent in most cases, we decided against making any compromises, and so in the end we did not use it.

My final answer is totally positive: NAO32 can be considered a real HDR format.

I'm wondering if this interviewer isn't Titanio..perhaps arranged from B3D PM's or the like in fact..

Edit: It's Xbdestroya apparantly..

 

[DeanA]

I think he's simply saying that because RSX apparently has access to two different memory pools connected via separate buses, as discussed here, it may become possible to balance your rendering operations in order to get an effective amount of bandwidth greater than that available by just working only in VRAM.

So, for a framebuffer effect example, traditionally you may have all your source textures and destination buffer in VRAM. Reads from your source buffers and writes to your destination buffers all take place over the 128-bit bus to VRAM, so you may potentially end up being bandwidth limited. As an alterative approach, if you moved one or more of your source textures over into XDR, you might end up with a more balanced system that uses both XDR bandwidth and VRAM bandwidth, and as such is possibly able to transfer more.

Cheers,
Dean

 

[Crossbar]

I think he's simply saying that because RSX apparently has access to two different memory pools connected via separate buses, as discussed here, it may become possible to balance your rendering operations in order to get an effective amount of bandwidth greater than that available by just working only in VRAM.

So, for a framebuffer effect example, traditionally you may have all your source textures and destination buffer in VRAM. Reads from your source buffers and writes to your destination buffers all take place over the 128-bit bus to VRAM, so you may potentially end up being bandwidth limited. As an alterative approach, if you moved one or more of your source textures over into XDR, you might end up with a more balanced system that uses both XDR bandwidth and VRAM bandwidth, and as such is possibly able to transfer more.

Cheers,
Dean

Thanks for that clarification.:smile: I actually read that thread you pointed out but it's a mixture of vague statements from people in the know and speculations of less informed people, so it's hard to make head or tail from it.

You are suggesting there is more work being done over the FlexIO interface to the XDR RAM in parallel with post effects being applied to the frame buffer in the GDDR3 RAM. Is the bandwidth requirement of the texture transfers really significant or can you point out some more examples of possible usage?

 

[DeanA]

Is the bandwidth requirement of the texture transfers really significant or can you point out some more examples of possible usage?

If I had to guess, I'd expect it to vary based on the number and type of textures being dealt with.. and possibly the sized of the addressed texture areas too (I'm thinking about texture cache usage here.. as cache reloads will end up contributing to the total bandwidth used). Heavy use of multitexturing would seem like something that may benefit from careful bandwidth management, no?

Dean

 

[PSman]

I'm disappointed TBQH, Heavenly Sword really only run at 30fps? . HS is a fast pace action game (correct me if I'm wrong) so 30fps is unacceptable IMO.


Why can't they make it 60fps!?

 

[Dave Baumann]

Marco: Not at all; in fact for many framebuffer effects I believe RSX will have an edge over Xenos. Don't want to go into details, but let me just point out that RSX is connected to two seperate busses, not just one.

I think that does needs some more details, because the only element discussed there, two seprate busses, isn't actually that different between RSX and Xenos.

 

[Carl B]

I'm disappointed TBQH, Heavenly Sword really only run at 30fps? . HS is a fast pace action game (correct me if I'm wrong) so 30fps is unacceptable IMO.


Why can't they make it 60fps!?

Of course they could if that was their primary design focus... but it's not. Rather, the focus is on the appearance and gameplay. I rather have a solid 30fps (and solid seems the goal) than a jumpy higher framerate anyway.

I have no doubt that if they targeted 60FPS instead of 30, they would have to trim back on some of what's made HS the #1 early PS3 game to watch at this point. Most will agree the game looks beautiful; it's all a matter of trade-offs. And I'm pretty sure that if the E3 demo was running at 60FPS, the trade-offs made to reach that goal would have kept this game from ever reaching the public awareness that it has.

Think of Gears of War, another graphics flag-bearer, that also will probably target 30FPS rather than 60.

The game plays smooth - to me that's what matters.

 

[PSman]

Well it just that both 360 and PS3 are VERY powerful next-gen cosole, so i expect them to be able to run game like Gears of Wars and HS at 60fps without any sacrifice.

 

[Acert93]

I'm disappointed TBQH, Heavenly Sword really only run at 30fps? . HS is a fast pace action game (correct me if I'm wrong) so 30fps is unacceptable IMO.

I like 60fps as much as the next guy (well, not quite as much as the "all games must be 60fps" but they are never happy!) but my main concern would be stability. A rock solid 30fps is better than a jumpy 60fps. Likewise no tearing (or at most minimal tearing) at 30fps is better than excessive tearing at 60fps.

If they have rock solid framerates at 30fps at all times (inside and outside) and no tearing I think they will be doing quite good.

Why can't they make it 60fps!?

I think they have already been pushed to Q1 2007. My guess it simply comes down to getting the game out on a timely manner. Kind of like the PGR3 debate from the 360 launch. Some people would have preferred months of delays in exchange for 60fps. But at some point getting a game out and working on your next title is important for the company, the platform, and even gamers. And as much as you and I may love 60fps, in the past it has not been a criteria of purchase by the mainstream consumer. Pretty graphics at 30fps seems to rule the roost so to speak.

 

[Shifty Geezer]

Well it just that both 360 and PS3 are VERY powerful next-gen cosole, so i expect them to be able to run game like Gears of Wars and HS at 60fps without any sacrifice.

Think 'irresistable force meeting an immovable object'. Yes their powerful, but the graphics are very demanding, and, certainly at this point in the system's lifecycle where developers are just starting out on the hardware, there aren't the resources to provide both the graphical advances and better framerates over last gen. Plus screenshots count for a great deal in first impressions and garnering interest. At 30fps you can afford to have twice as much graphical pazazz per frame (and screenshot) than at 60 fps, making your game appear twice as pretty as a similar game spread across 60 fps.

I like 60 fps and would love it if it was a de facto target, but stable 30 fps is fine and inevitable when stills still contribute much to the media attention you get for your game.

 

[nAo]

I think that does needs some more details, because the only element discussed there, two seprate busses, isn't actually that different between RSX and Xenos.

All details are already there since I was talking about frame buffer effects: in the vast majority of cases you'll end up being texture bw limited and last time I checked Xenos does not have 2 separate busses to fetch textures.

 

[PSman]

Thanks guys for clearing that up:smile:

 

[Dave Baumann]

All details are already there since I was talking about frame buffer effects: in the vast majority of cases you'll end up being texture bw limited and last time I checked Xenos does not have 2 separate busses to fetch textures.

True enough, but unless you just aren't ouputting any pixels at all then it probably has a greater local texture bandwidth in the first place.

 

[sonyps35]

Well it just that both 360 and PS3 are VERY powerful next-gen cosole, so i expect them to be able to run game like Gears of Wars and HS at 60fps without any sacrifice.

They arent THAT powerful. They're mid-high range PC's, with less RAM and imo overrated CPU's (meaning 3X3.2 ghz CPU's in 360 sounds great, but they're not like 3 A64's, same for cell).

 

[bleon]

The article into great detail about how these great looking effects are created. I'm really surprised that Marco was so open about the methods they used, especially since they seem to have done some research and come up with a solution thats pretty ingenious.

Aren't developers concerned about giving away secrets to their competitors? For example, I couldn't imagine Namco revealing their secrets to better IQ and AA on the PS2.

 

[RavenFox]

Great interview and good to see RSX is getting the job done right.

 

[scificube]

True enough, but unless you just aren't ouputting any pixels at all then it probably has a greater local texture bandwidth in the first place.

If pixel fill is not the limit then what does that matter?

Also the local bandwith to Xenos is 22.4Gb/s - whatever the CPU consumes + whatever the textures caches alleviate.

If you're referencing the eDram I would think that has to do with accessing buffers...not texture access unless what you're saying is since buffer access is not something done in system memory Xenos has more free from the 22.4Gb/s available from system memory than RSX has from VRAM (22.4GB/s)+ XDR (measured 26.1GB/s theoretical 35GB/s) - framebuffer usage - Cell comsumption + whatever texture caches alleviate for texturing.

If RSX saturated VRAM with framebuffer access and access to textures in that memory pool it would still have a fair amount of bandwith to draw up from XDR if need be where as with Xenos it is not permitted to saturate GDDR3 bandwith with texture accesses or the CPU would starve as there is no other pool of memory to pull from.

Lastly, it seems to me that pixel shaders often sample textures during intermediate work much more often then they output pixels in the end of course barring blending effects which are fillrate consumers...but then...why do you have to do both at the same time anyway?

Just trying to understand where you're coming from

 

[Cheezdoodles]

I think that does needs some more details, because the only element discussed there, two seprate busses, isn't actually that different between RSX and Xenos.

How so?

 

[Carl B]

The article into great detail about how these great looking effects are created. I'm really surprised that Marco was so open about the methods they used, especially since they seem to have done some research and come up with a solution thats pretty ingenious.

Aren't developers concerned about giving away secrets to their competitors? For example, I couldn't imagine Namco revealing their secrets to better IQ and AA on the PS2.

There's no denying Marco is very open about it, but if you've spent any time here in the past couple of months, then you know that's just how the HS crew is. I don't think they view it as a proprietary secret to keep guarded, but rather something that if other PS3 devs can use, they'd be happy to share their knowledge about. Plus, it's just downright interesting.