Beating Emotion Engine

[patroclus02]

This is just curiosity..

Graphics Synthesizer running at 200 MHz
16 pixel pipelines with each pipeline being as robust as an NV4-TNT or NV5-TNT2 pipeline
1 TMU per pixel pipeline
many things done in single pass (not having to redraw geometry) or loopbacks
4x FSAA unit that only demands some of the plentiful bandwidth and half the fillrate
proper mip-mapping
all the expected blending modes
-8 MB eDRAM - 96 GB/sec bandwidth
-32 MB external graphics memory ~12.8 GB/sec
a 4.8 GB/sec connection from GS to EE and/or to the 128 MB main memory

3,200 Mpixels/sec fillrate - 3,200 Mtexels/sec w/ trilinear filtering

1,600 Mpixels/sec fillrate with 4x FSAA

3,200 Mpixels/sec fillrate ??
16 pixel pipelines, with 1 TMU per pipe at 200Mhz leads to 3200 Mpixels/sec for bilinear filtering, but it would take an extra clock cycle (and memory bandwidth) for trilinear filtering... am I wrong?

 

[Lazy8s]

I didn't mean to imply that the PS2 had no performance advantages. However, I presented no opinion regarding PS2 versus DC, just the factual context -- statistics like dates, games, EE's 240 square millimeter area vs. SH-4's 42, etc. -- for the comparison that was being made of them.

How much of a performance increase is needed to justify an area increase of about six fold for a CPU? Technology performance scales much more than 1:1, and the SH-4's floating-point calculation speed is over a quarter faster than that ratio to begin with.

NAOMI2 showed that generic SDRAM had the headroom for even better performance, so Rambus DDR in the PS2 was largely just extra expense to make up for the system's bandwidth wastage.

DC's 100M pix/sec fillrate was enough to keep it competitive. It applied multiple texture layers at 60 fps, it filtered textures as well as PS2, and the image resolution in its back and front buffers was even better than PS2's on average. Depth complexity did average over three, with Early Z systems of the time, even in a conservative example for DC like Quake 3. Post-processing effects, blur, focus and depth of field, etc. -- adjustments that are really the domain of the eye generally -- were certainly better on PS2, yet Dead or Alive 2's blurs, dissolves, and fades proved that the DC was competent.

 

[swaaye]

I didn't mean to imply that the PS2 had no performance advantages. However, I presented no opinion regarding PS2 versus DC, just the factual context -- statistics like dates, games, EE's 240 square millimeter area vs. SH-4's 42, etc. -- for the comparison that was being made of them.

How much of a performance increase is needed to justify an area increase of about six fold for a CPU? Technology performance scales much more than 1:1, and the SH-4's floating-point calculation speed is over a quarter faster than that ratio to begin with.

I think this is a bit too simplistic of a way to compare processors. If you're looking at "FLOPs", well we all know those ratings are almost useless. SH4 was very cool in its day, but it also doesn't seem to have been adopted in many applications and that probably speaks volumes about its speed. Of course one could say the same about EE, but that isn't entirely true because SH4 was more of a off-the-shelf processor, while EE is meant for the PS2 and was designed for it specifically.

Die size doesn't necessarily mean anything either. Look back at the IDT Winchip2, for example. It was a lot smaller than K6-2 and PII, but it didn't clock worth a damn and sure wasn't fast at all.

 

[Crazyace]

I didn't mean to imply that the PS2 had no performance advantages. However, I presented no opinion regarding PS2 versus DC, just the factual context -- statistics like dates, games, EE's 240 square millimeter area vs. SH-4's 42, etc. -- for the comparison that was being made of them.

I think your original quote implied that ( with the graphical downgrade quip ) but it's not really worth a religous argument.

How much of a performance increase is needed to justify an area increase of about six fold for a CPU? Technology performance scales much more than 1:1, and the SH-4's floating-point calculation speed is over a quarter faster than that ratio to begin with.

A lot of chip real estate ends up being used in multiple buses.. ( Just look at the Cell - over 30% of the chip area is internal buses and external interfaces ) and the EE had a lot of individual high bandwidth connections, both internally ( the 128 bit internal bus ) and externally ( rambus / GS direct connection / io direct connection ) There were only 3 times as many transistors in the EE compared to the SH4 after all.

NAOMI2 showed that generic SDRAM had the headroom for even better performance, so Rambus DDR in the PS2 was largely just extra expense to make up for the system's bandwidth wastage.

I think NAOMI2 showed the opposite - especially in the way it required many independant banks of SDRAM. Two for the CLX's , one for the elan(cant remember name) and one for the SH4.

DC's 100M pix/sec fillrate was enough to keep it competitive. It applied multiple texture layers at 60 fps, it filtered textures as well as PS2, and the image resolution in its back and front buffers was even better than PS2's on average. Depth complexity did average over three, with Early Z systems of the time, even in a conservative example for DC like Quake 3. Post-processing effects, blur, focus and depth of field, etc. -- adjustments that are really the domain of the eye generally -- were certainly better on PS2, yet Dead or Alive 2's blurs, dissolves, and fades proved that the DC was competent.

Again I think this isn't something that was true. The DC was competent, but not comparable

 

[Lazy8s]

SH-4s have been adopted into countless consumer electronics, appliances, car information systems and, now, years later, cellphones. They're one of the most popular solutions outside of ARM's practical monopoly.

Die size determines the number of dies that can fit on a wafer in production, the output, and thus ultimately determines the cost of a chip more than any other factor, so it's critical to a chip's value.

Die space is just as expensive whether it's being eaten by bus circuitry or something else.

SDRAM's affordability allowed NAOMI2 to have multiples of memory space for the cost of Rambus DDR space and, despite duplicate data, have much more storage ultimately. The tile-based deferred texture processing architecture which saved enough bandwidth to allow NAOMI2 to use SDRAM also freed even more budget by yielding smaller graphics chips.

 

[Kryton]

NAOMI2 showed that generic SDRAM had the headroom for even better performance, so Rambus DDR in the PS2 was largely just extra expense to make up for the system's bandwidth wastage.

DC's 100M pix/sec fillrate was enough to keep it competitive. It applied multiple texture layers at 60 fps, it filtered textures as well as PS2, and the image resolution in its back and front buffers was even better than PS2's on average. Depth complexity did average over three, with Early Z systems of the time, even in a conservative example for DC like Quake 3. Post-processing effects, blur, focus and depth of field, etc. -- adjustments that are really the domain of the eye generally -- were certainly better on PS2, yet Dead or Alive 2's blurs, dissolves, and fades proved that the DC was competent.

Wasn't that all down to the TBR provided in the graphics chip? The SDRAM didn't have to be amazingly fast because it wasn't being accessed every tile. It's also why DC had some stunning tech. demos for it with hundreds of transparent cards being laid on top of each other running at 60fps (a feat most modern cards would find tricky - they're probably still on the PowerVR site somewhere).

 

[Crazyace]

SH-4s have been adopted into countless consumer electronics, appliances, car information systems and, now, years later, cellphones. They're one of the most popular solutions outside of ARM's practical monopoly.

SH4 is very popular - but not as a single chip.. Most of the SH series being pushed into the market is of the SH+DSP style, and the newer architecture..

Die size determines the number of dies that can fit on a wafer in production, the output, and thus ultimately determines the cost of a chip more than any other factor, so it's critical to a chip's value.

Absolutely true, but packaging and PCB design also costs. Otherwise why would there be the trend for system on chip solutions .. ( such as the various SH4 variants... none of the automotive solutions I've seen use the dreamcast SH-4 after all )

Die space is just as expensive whether it's being eaten by bus circuitry or something else.

I agree, but the EE completely out classes the SH4 in so many ways - much more than the simple difference in flops. It's a bigger chip because it has more capabilities specifically tailered for the PS2 use.
( If you really wanted to make the SH4 seem better just concentrate on double precision )

SDRAM's affordability allowed NAOMI2 to have multiples of memory space for the cost of Rambus DDR space and, despite duplicate data, have much more storage ultimately. The tile-based deferred texture processing architecture which saved enough bandwidth to allow NAOMI2 to use SDRAM also freed even more budget by yielding smaller graphics chips.

I never though of NAOMI2 as 'affordable' in that sense.. I'm sure that if they could have engineered it using faster memory they would have, but for the tiny arcade market the engineering effort wasn't worth it.

 

[Bigus Dickus]

I already say old PS2 design is not so great and that Gamecube and new PS2 slim design is best for low cost manufacture. Elegant is different issue than manufacture cost but Xbox is not so elegant also.

I guess I just don't understand this viewpoint. In six years, the slimline PS2 and GC will be very inelegant compared to what could be accomplished with then current manufacturing processes, storage and drive options, etc. I don't see things in such relative terms, at least not in this case. Engineers do their best with the design solutions they have at their disposal, in the here and now. What improvements come in the future shouldn't negate their efforts of today.

The original PS2 was just as "elegant" for the day it was designed as the slimline PS2 is now. Was a Model T inelegant because it didn't have fuel injection, power brakes, and airbags? I simply don't understand that reasoning. And while Sony was able to enjoy process shrinks, allowing them to offer a slimline PS2, Microsoft was not. Neither of which, IMO, has anything at all to do with how competently designed the original Xbox and PS2 were.

 

[ihamoitc2005]

Maybe

I guess I just don't understand this viewpoint. In six years, the slimline PS2 and GC will be very inelegant compared to what could be accomplished with then current manufacturing processes, storage and drive options, etc. I don't see things in such relative terms, at least not in this case. Engineers do their best with the design solutions they have at their disposal, in the here and now. What improvements come in the future shouldn't negate their efforts of today.

The original PS2 was just as "elegant" for the day it was designed as the slimline PS2 is now. Was a Model T inelegant because it didn't have fuel injection, power brakes, and airbags? I simply don't understand that reasoning. And while Sony was able to enjoy process shrinks, allowing them to offer a slimline PS2, Microsoft was not. Neither of which, IMO, has anything at all to do with how competently designed the original Xbox and PS2 were.

I understand what you say my friend but I look at Gamecube which is very old design and has old manufacturing but is not like old PS2 and Xbox, no? It takes 4 years for PS2 to have same good design.

Also, Xbox was super-quick design with "off-the-shelf" parts so is maybe difficult (without contract problem) to have good shrink like custom design console.