Support NeoGAF

[Astral Dog]

They just mention data processing , if the game came out on multiple disks it would work fine (with more loading),it just won't be that easy to simply copy paste without adjustments

 

[PaintTinJr]

Thanks for the insight. I also had the higher SIMD saturation in mind leading to much higher efficiency making 1:1 compute comparisons somewhat misleading (as it tends to happen). And just to confirm the theoretical max; is the official figure for CELL PPU 33.8 GFlops or 25.6 GFlops (just like the SPEs)? Is it the exact same core as the Xenon one? I've had some difficulties to find the official Sony/IBM figures concerning those points.

I'm pretty sure it is the lower figure because it divides correctly by the 3.2Ghz clock. But will read the wiki more thoroughly and update if it isn't the 25.6gflop figure.

 

[Lysandros]

I'm pretty sure it is the lower figure because it divides correctly by the 3.2Ghz clock. But will read the wiki more thoroughly and update if it isn't the 25.6gflop figure.

The thing with wikipedia is it's quite ofen misleading/inaccurate. The article puts CELL's whole computational power at 153 GFlops which is of course false because that is solely the sum of 6 SPEs excluding the PPU, which exists and also has a FPU adding to the computational whole. The 7th SPE for the OS is also there and functioning at 25.6 GFlops but not available to developers. Meanwhile the Xenon article has no problem putting at it 115 GFlops (max/full chip) without excluding any PPU for this instance and without accounting for the OS reserve, odd really.

Edit: By the way, how are we reaching 115 GFlops for Xenon if PPU is at 25.6 GFlops, shouldn't be 76.8 GFlops (3×25.6)?

 

[PaintTinJr]

The thing with wikie is it's quite ofen misleading/inaccurate. The article puts CELL's whole computational power at 153 Glops which is of course false because that is solely the sum of 6 SPEs excluding the PPU (for some reason) which also has a FPU adding to the computational whole. The 7th SPE is also there and functioning at 25.6 GFlops but not available to developers. Meanwhile the Xenon article has no problem putting at it 115 GFlops (max/full chip) without excluding any PPU for this instance and without accounting the OS reserve, odd really.

The Xenon wiki makes no such claim of any performance, and the Cell BE wiki references the Xenon directly below this paragraph

" PPE consists of three main units: Instruction Unit (IU), Execution Unit (XU), and vector/scalar execution unit (VSU). IU contains L1 instruction cache, branch prediction hardware, instruction buffers, and dependency checking logic. XU contains integer execution units (FXU) and load-store unit (LSU). VSU contains all of the execution resources for FPU and VMX. Each PPE can complete two double-precision operations per clock cycle using a scalar fused-multiply-add instruction, which translates to 6.4 GFLOPS at 3.2 GHz; or eight single-precision operations per clock cycle with a vector fused-multiply-add instruction, which translates to 25.6 GFLOPS at 3.2 GHz.[36]"

This text gives a clear statement that the performance number is derived directly from the 3.2GHz clock x 8 flops of the FMA VMX vector unit per clock to give the 25.6GFLOPS.

So I'm pretty sure it is the same for all PPEs and SPEs, just that PPEs operate at single digit Gflops for the main thread core and at best just into double digits for cores doing vector maths, whereas the SPEs when used properly are doing most of their theoretical maximum.

 

[Lysandros]

The Xenon wiki makes no such claim of any performance, and the Cell BE wiki references the Xenon directly below this paragraph

" PPE consists of three main units: Instruction Unit (IU), Execution Unit (XU), and vector/scalar execution unit (VSU). IU contains L1 instruction cache, branch prediction hardware, instruction buffers, and dependency checking logic. XU contains integer execution units (FXU) and load-store unit (LSU). VSU contains all of the execution resources for FPU and VMX. Each PPE can complete two double-precision operations per clock cycle using a scalar fused-multiply-add instruction, which translates to 6.4 GFLOPS at 3.2 GHz; or eight single-precision operations per clock cycle with a vector fused-multiply-add instruction, which translates to 25.6 GFLOPS at 3.2 GHz.[36]"

This text gives a clear statement that the performance number is derived directly from the 3.2GHz clock x 8 flops of the FMA VMX vector unit per clock to give the 25.6GFLOPS.

So I'm pretty sure it is the same for all PPEs and SPEs, just that PPEs operate at single digit Gflops for the main thread core and at best just into double digits for cores doing vector maths, whereas the SPEs when used properly are doing most of their theoretical maximum.

Sorry, that was the Xbox 360 technical specifications article which gave the 115 GFlops figure instead, my mistake. In Central Processing Unit subtitle:

Xbox 360 technical specifications - Wikipedia

en.m.wikipedia.org

So it's 25.6 GFlops for both CELL and Xenon PPU then? How are we reaching 115 GFlops in this case?

Edit: Slide taken from 'Xbox 360 System Architecture, Anderews Baker, Georgia Institute of Technology', deep dive, puts Xenon at ~75 GFlops indeed (couldn't attach pdf file):

 

[The Pleasure]

Looking back at it now, the uncharted games weren't doing anything special. Hell if anything drake's animations were the most impressive aspect. I remember a guy arguing the animations were a bluray feature hahaha

They were with donut drake

 

[Negotiator]

They were with donut drake

 

[The Pleasure]

Reported for donut shaming

 

[PaintTinJr]

... How are we reaching 115 GFlops in this case?

...

Who knows it is sounds like magic for a tech spec win that no one would bother checking, and just illustrates why that whole gen distilled through people with vested interest in so-called faceoff wins was so irksome for anyone that actually cared about technology from a factual angle IMO.

edit:

I think it is derived from the 9.6billion dot products per second figure, where it is implying 12 floats ops per dot product...eg maybe they are think it is FMA being 3 ops on two 4 float vectors [x, y, z, w].

 

[March Climber]

I'd argue that Uncharted 3 felt more impossible than UC 2.

 

[Lysandros]

I'd argue that Uncharted 3 felt more impossible than UC 2.

Very high quality post processing effects, antialiasing, physics and even animation; all CELL's forte in the right hands.

 

[Negotiator]

I'd argue that Uncharted 3 felt more impossible than UC 2.

I'm still waiting for a 2024 video game (either on PC or consoles) to offer a more impressive playable sequence...

I may sound heretic, but for me even UC4 setpieces weren't that impressive, despite PS4's 1.84TF raw power.

 

[March Climber]

I'm still waiting for a 2024 video game (either on PC or consoles) to offer a more impressive playable sequence...

I may sound heretic, but for me even UC4 setpieces weren't that impressive, despite PS4's 1.84TF raw power.

I bet it's because deep down you knew the PS4 was capable of more. TLOU 2 and Horizon FW proved that. With UC 3's release I knew that the PS3 had reached it's peak.

I know that people like to point to TLOU 1 and God of War 3, but I always felt that Uncharted 3 had beaten both of them, graphically and fidelity-wise.

 

[Negotiator]

I bet it's because deep down you knew the PS4 was capable of more. TLOU 2 and Horizon FW proved that. With UC 3's release I knew that the PS3 had reached it's peak.

I know that people like to point to TLOU 1 and God of War 3, but I always felt that Uncharted 3 had beaten both of them, graphically and fidelity-wise.

Of course I know that the PS4 is 5-10 times more capable than the PS3, but you know what?

It's also a matter of having the proper team.

ND of 2011 (UC3) is not the same as ND of 2016 (UC4)... let alone ND of 2024. I don't have high expectations anymore.

 

[March Climber]

ND of 2011 (UC3) is not the same as ND of 2016 (UC4)... let alone ND of 2024. I don't have high expectations anymore

While true there might be some new blood, when it comes to graphical capabilities, they have yet to prove any of us wrong. TLOU 2 was a definite leap, to the point where the remaster on PS5 had almost zero changes. I fully expect their next game to go head-to-head with whatever the greatest UE 5 title will be, when it releases.

 

[Negotiator]

While true there might be some new blood, when it comes to graphical capabilities, they have yet to prove any of us wrong. TLOU 2 was a definite leap, to the point where the remaster on PS5 had almost zero changes. I fully expect their next game to go head-to-head with whatever the greatest UE 5 title will be, when it releases.

Graphics-wise, yeah, TLOU2 is pretty good, but I was talking about setpieces... even TLOU1 wasn't a setpiece-heavy game, it's more "grounded".

Maybe UC4 took some (unwanted) inspiration from TLOU1 to feel more "grounded" and less of a hollywood blockbuster like UC2-3, especially after Amy Hennig left ND.

I remember her saying UC3 was modelled around setpieces (like the ship sequence) and then they wrote the story, not the other way around.

Druckmann's approach is a very different compared to Amy's...

 

[Melfice7]

I remember her saying UC3 was modelled around setpieces (like the ship sequence) and then they wrote the story, not the other way around.

Druckmann's approach is a very different compared to Amy's...

Which explains why U3 feels so disjointed in so many parts in comparison to the fantastic pacing of U2

Druckmans's U4.. half the game is a slog and went the more grounded path which is sad because U4 has the best gunplay and moment to moment combat in the series, but the set pieces are severely lacking in general

 

[Negotiator]

Which explains why U3 feels so disjointed in so many parts in comparison to the fantastic pacing of U2

Druckmans's U4.. half the game is a slog and went the more grounded path which is sad because U4 has the best gunplay and moment to moment combat in the series, but the set pieces are severely lacking in general

It felt disjointed because Graham McTavish had to leave to shoot a movie in New Zealand, IIRC.

UC2 and UC3 are very similar in their approach, I remember some people saying that UC3 felt too "samey" compared to UC2, so maybe that's another reason ND felt compelled to change the recipe.

We'll never know for sure, but Amy's take on UC4 would probably have been more similar to UC2-3, unlike Druckmann's which was definitely more grounded/dramatic.