8 ROPs on the RSX. Bad engineering decision?

Status
Not open for further replies.
I see. However in PC architect the PCIe bottleneck will trash your 7800GTX/GTX(512) performance around 40-50% hit. 16ROPs 7800GTX may perform somewhere around
40-50% performance hit.

ROPs render directly to the graphics memory over that 256bit bus. They don't interact with the PCI-E interface which itself is not a bottleneck at all. Countless benchmarks have proven that point across many games.

However 128-bit mem interface with 8 ROPs on RSX were fit for smart performance
adjustdable. Due the PS3 architect its easy to say RSX will perform better than
PC 7800GTX/GTX(512) side by side.

The GTX512 has 10% more clock speed, twice the number of ROPs and more than double the graphics memory bandwidth and size. So its obvious that the GTX512 would be faster. The standard GTX would be slower under some circumstances though (but not all).

Those number of 7800GTX is more impress but in your PC due the bottleneck around
there. You'll reach only 40-50% remain performance of something you paid for.
I think in the closed box environment like PS3 we'll see something that you all can't
see that 7800GTX/GTX(512) cards perform on PC in soon.

There is no example of the PS3 running a game better than a GTX512 (~7900GTX performance level) at the same settings and resolution. The reason there is no example is because this 50% performance drop you make claim to doesn't exist. There is no bottleneck in the 7800GTX512 that doesn't exist to a greater extent in RSX. Cell may be able to help out in some areas but as yet I see no evidence that its done so to such an extent to make the PS3 exceed a 7800GTX 512 powered PC's capabilities.
 
In PC architect, not only software part produce bottleneck , in fact some part of hardware make some seriously bottleneck, too. In software part, there are many ways to solve its.
By re-coding some OS handle, driver, API, reduce memory taken, ...etc.

By the way, In hardware part nothing can be done except change some components of
your system to solve its. Intel platform the biggest bottleneck was its chipsets other hand
AMD platform its bottleneck was its CPU architect pipeline not wide issue enough as Intel
to perform parallelized. On GPU itself both AMD and nVidia can produce the TITANIC like
GPU (8800GTX aka HD3870 and above etc.) however if you don't match them to the high
performance CPU and Chipset it'll like you drive your speedboats in your swimming pools.

PC CPU's are more than powerful enough to feed GPU's with sufficient graphics data to render most games at 60fps and above.

Unless you would care to show a game running at an unplayable framerate at its minimum settings/resolution on an 8800GTX ?
 
40-50% estimated by what/who? There are different types of load's you can put on a GPU, different engines/game designs put different strains on the GPU.
I wouldn't be surprised if it was a case of trying to do a lot of the same things that people talk about on PS3. Sure, with FlexIO and VRAM bandwidth limitations (and a CPU with the muscle of Cell), it makes sense to texture from or write render targets to main memory as long as you can spare the bandwidth on the CPU side... But if you try to do the same thing over PCIe or even try to copy between VRAM and main RAM, of course you're going to see performance go into the crapper.

When you say PS3 you are talking about the RSX and the Cell, the Cell can and does help out RSX as been stated by some devs in interviews etc. But RSX alone, I doubt it. Not against the Ultra one.
And it's often the things at which Cell helps the RSX that the bottlenecks are found. Something that people often forget is that when you're pixel or texel fillrate-bound, chances are good that you've also got a pretty high amount of geometry and/or number of render passes. And everybody hits on all sorts of use cases that can really make the information look skewed. In general, if you're limited only by pixel fillrate on RSX, that's probably a good sign, considering that there are other limiting factors that you can hit sooner and easier.

Certainly, if Sony was pushing for something like 1080p as a *requirement*, there would be more issue down this road. And on PCs, 16 ROPs is of value because it's so normal for people to push resolutions skywards... 1080p is certainly achievable with 8 ROPs and all, but you'd be hard-pressed to say that the chip would be well-suited to it in the context of PS3.
 
8 ROPs is a good decision given the framebuffer bandwidth.

Assuming 500MHz clock and 650MHz DDR memory, RSX has 42 bits of data access per ROP per clock. You need very good Z-compression, no AA or perfect colour compression, simple pixel shader, high texture magnification (i.e. near-zero texture BW), and optimal memory controller efficiency for ROPs to be a slight bottleneck instead of the BW. You'll rarely if ever meet these conditions, so BW or shader length will limit you to much less than 8 pixels per clock.

In G7x, the ROPs ran at half speed when alpha blending (i.e. transparency) was enabled, and I assume the same is true here. Since you need to read and write from the framebuffer, you'd be hard pressed to exceed 4 pix/clock even with more ROPs.

Xenos is 8 pix/clk under all 32bpp conditions, including AA and alpha blending (except FP10 with alpha blending, I think, which is half speed). Framebuffer bandwidth is never an issue. Arun is probably right that bandwidth between the GPU and eDRAM was the reason for this limit. Even though the eDRAM has bandwidth for 64 pix/clk without AA or alpha blending, Xenos can't get that much info to the eDRAM. If the fabriction technology was there to economically integrate eDRAM with a complex GPU then the story may have been different.
 
Same to me. 8 ROPs is enough for RSX.

PC CPU's are more than powerful enough to feed GPU's with sufficient graphics data to render most games at 60fps and above.

I think you mean high performace CPU like Core2 Quad. However let combine your 8800GTX
with something like Celeron or Pentium E and try again.
 
Same to me. 8 ROPs is enough for RSX.



I think you mean high performace CPU like Core2 Quad. However let combine your 8800GTX
with something like Celeron or Pentium E and try again.

So either you're saying the PS3 cell is like a Pentium E OR you think that a lot of people with 8800gtx's will put them in computers that have a Pentium E or Celeron???
 
No that not my point. I'd like to show you all that high-end PC GPU need high performance CPU to show up their power,too. If you plug-in its with the low-end PC component these will trash your 8800GTX performance around 30% hit due the bottleneck by CPU , Chipsets and all affect performance issue.

Cell BE design was a lot better than Pentium E.
 
I think you are lost, you cant say any fixed % perfomance penalty due to lower end CPU. It depends on a lot of factors, it depends if the game is CPU limited, is the CPU really weak (Celeron) or just older type, etc.

Chipset affects perfomance but not drastically, we are talking about around 5-10% perfomance difference assuming both mobos has the same RAM. But as said depends on if the game is CPU bound or not. IF yes then what CPU wouldbe needed to achieve 60fps?

Thats a good question but not really a question anymore... all multiplatform games, a Opteron 185 (AMD 2x2.6GHz) achieves it with a G80. And then no other game runs under 30fps due to the CPU.

And in fact I find that many games I play have the GPU maxed out before the CPU.

BTW I consider my CPU to be low-end perfomance wise and price wise. :smile:

What thread was this btw? ;)

Edit: "Around 30%" is still not correct, it could be 0-95% depending on how old the CPU is, if it is single-core or multi-core etc etc. ;)
 
Last edited by a moderator:
I think you are lost, you cant say any fixed % perfomance penalty due to lower end CPU. It depends on a lot of factors, it depends if the game is CPU limited, is the CPU really weak (Celeron) or just older type, etc.

Sorry friend , Please take a look I use " around " for my words not strickly number of percent.
 
Sorry friend , Please take a look I use " around " for my words not strickly number of percent.

I believe you are missing his point. He is stating that there are a plethora of factors that would affect performance under the circumstance. Associating a percentage (regardless of the use of the word "around" or not) is not something that can be estimated outside of a very individualized situation.
 
I think you mean high performace CPU like Core2 Quad. However let combine your 8800GTX
with something like Celeron or Pentium E and try again.

Your example makes no sense. Why would anyone pair a high end GPU with a budget or obsolete CPU.

I'm talking about a normal CPU that you would expect to find paired with a high end GPU.

That ranges from sub $100 Athlon X2's right up to Core 2 duo's and quads.
 
Well one other question in regards to the ROPs, if the chip really is just an off the shelf G71, would've they're really been a point in lowering the number of ROPs considering there would be cost in revising the GPU then putting it production? That's a big thing I think about, if the chip was already in production for 7800 graphics boards, might as well just kept on producing the exact same variant and shoving it into the PS3.
 
In G7x, the ROPs ran at half speed when alpha blending (i.e. transparency) was enabled, and I assume the same is true here. Since you need to read and write from the framebuffer, you'd be hard pressed to exceed 4 pix/clock even with more ROPs.
Though there are ways to exceed 4 pixels per clock..;)
 
Your example makes no sense. Why would anyone pair a high end GPU with a budget or obsolete CPU.

You're missing my point. Yes if you've high-end GPU I think at least your CPU isn't bad to
pair them together.right? However the point that I need to show was High-end GPU need
High Performance CPU and Chipset to show up it power
. If your pair some kind of these
configs.

(A) Celeron + GeForce 8800GTX
(B) Core 2 Quad + Intel GMA
(C) Core 2 Qaud + GeForce 8800GTX

Is (A) config perform as well as (C) config ?
Is (B) config perform as well as (A) config ?
Are both (A) and (B) configs better than (C) config ?

If yon didn't have any of those parts to proof. Let's go to computer store near your home
and try it.


I'm talking about a normal CPU that you would expect to find paired with a high end GPU.

That ranges from sub $100 Athlon X2's right up to Core 2 duo's and quads.
 
You're missing my point. Yes if you've high-end GPU I think at least your CPU isn't bad to
pair them together.right? However the point that I need to show was High-end GPU need
High Performance CPU and Chipset to show up it power. If your pair some kind of these
configs.

(A) Celeron + GeForce 8800GTX
(B) Core 2 Quad + Intel GMA
(C) Core 2 Qaud + GeForce 8800GTX

Is (A) config perform as well as (C) config ?
Is (B) config perform as well as (A) config ?
Are both (A) and (B) configs better than (C) config ?

If yon didn't have any of those parts to proof. Let's go to computer store near your home
and try it.

I'll apologize for that then, it seems I misunderstood your original point. I just wanted to make it clear though that its not an inherant weakness of the PC platform that a slow CPU will pull down the performance of a fast GPU. Its simply a situation that should be avoided when designing a PC. Obviously, there is no performance degridation on the GPU side provided you have a decent CPU.

Or in other words, a 7800GTX 512MB that reaches its full potential in the PC because its coupled with say an Athlon X2 should be able to outperform RSX in the PS3. Obviously adding Cells graphics processing abilities may skew things a little.
 
These types of threads remind me of the early days when the PS2 was released, posters laid to waste the hardware of the PS2 as everyone and there brother picked it apart and highlighted it's down faults. Granted the arguments where all worthy and deserved merit, and probably helped the situation by bringing known hardware issues to the fore front. But in time the game devs jumped out of the game code box they where in and developed new software that made great PS2 games happen!!.

Nothings perfect, especially in hardware, and the same could be said for software. The game devs just have to once again, jump out of the box they are in and come up with new gaming code that will work on the PS3.
 
None of this "RSX was poorly designed" "only 8 ROPS" "weak vertex capabilities" type threads are reflected to the average consummer who sees games such as Uncharted, Ratchet, GT5 P, etc. all showcasing pretty much the best graphics available in their respective genre on console. This will only become more pronounced when Killzone 2, Motorstorm2, FFXIII, LPB, Resistance 2 come out in the near future. To the average consumer PS3 will offer the BEST console graphics in all genres of games. Either Sony devs are just aiming higher or things are not as bad as they are made out to be in these "RSX sucks" threads.
 
None of this "RSX was poorly designed" "only 8 ROPS" "weak vertex capabilities" type threads are reflected to the average consummer who sees games such as Uncharted, Ratchet, GT5 P, etc. all showcasing pretty much the best graphics available in their respective genre on console. This will only become more pronounced when Killzone 2, Motorstorm2, FFXIII, LPB, Resistance 2 come out in the near future. To the average consumer PS3 will offer the BEST console graphics in all genres of games. Either Sony devs are just aiming higher or things are not as bad as they are made out to be in these "RSX sucks" threads.

Did I say RSX sucks? No, I was merely questioning the number of ROPs, of which the entire Xenos package has the same, but not in regards to competition but to ease and usage of full 1080p rendering. That and we know Cell can be leveraged, I think Sony fully intended for both chips to be doing graphics work together. Maybe it's a sign of things to come from Sony and IBM?
 
Status
Not open for further replies.
Back
Top