PS3's Cell CPU is by far stronger than new Intel CPUs

I thought the lack of direct memory access was an intentional "feature" added by Sony for the PS3 Linux in order to keep people from playing pirated/homebrew games. If you were in the native console OS, then the software running through that DID have DMA.
Stretching back a lot of years to remember that.
The PS3 hypervisor (OtherOS) does not allow direct hardware acceleration to the RSX GPU, and instead simply uses it as a framebuffer (the CPU [specifically the PPE] does all the graphical work).
However, VRAM is still accessible as a RAMdisk and, in YDL 6.x and above, is auto-mounted by default as a fast SWAP partition, thus allowing 256MB of system RAM and 256MB of fast SWAP (plus whatever SWAP partition was assigned from the disk itself).

The RSX GPU in the PS3 is from the NVIDIA Series 7, which is right before the Series 8 G80 which featured the first programmable unified-shader architecture, so GPGPU applications, GPU compute, and OpenCL/CUDA will not operate with it.
There has been work in recent years to get hardware acceleration working on the PS3's GPU when in Linux, though, which is good news.
 
  • Like
Reactions: Mega6
like this
I recall reading an article a few years ago that stated that the CELL was actually on par or faster than the included nvidia GPU. But it was much easier to code for the RSX since nvidia provided the middleware (libraries, tools, etc)
 
I recall reading an article a few years ago that stated that the CELL was actually on par or faster than the included nvidia GPU. But it was much easier to code for the RSX since nvidia provided the middleware (libraries, tools, etc)
Sony originally intended for Cell to do both graphics duty and regular CPU duty. While Cell can be blazing fast on certain types of code: It did not meet performance requirements for general graphics duty. Which forced them to seek out a GPU contract, late in development. Generally speaking, Cell is not a match for RSX in graphics rendering.

However, it still remains that Cell can do things in the graphics pipeline and many games used it to do some pretty cool stuff, which took some load off the RSX GPU and/or added some flourish which may not have been possible with RSX alone.
 
Sony originally intended for Cell to do both graphics duty and regular CPU duty. While Cell can be blazing fast on certain types of code: It did not meet performance requirements for general graphics duty. Which forced them to seek out a GPU contract, late in development. Generally speaking, Cell is not a match for RSX in graphics rendering.

However, it still remains that Cell can do things in the graphics pipeline and many games used it to do some pretty cool stuff, which took some load off the RSX GPU and/or added some flourish which may not have been possible with RSX alone.

Something about particle effects if my memory serves me.
 
It’s actually because the PPE isn’t a very strong CPU and the other SPE cores are completely incapable of running either game logic or integer math.

Nothing to do with lazy developers, just a terrible architecture for a console CPU.
 
It’s actually because the PPE isn’t a very strong CPU and the other SPE cores are completely incapable of running either game logic or integer math.

Nothing to do with lazy developers, just a terrible architecture for a console CPU.
That isn't true at all, the PPE was moderately powerful CPU (about the equivalent of an IBM 970 FX @ 1.6GHz with SMT and capable of 25 GFLOPS - similar to Pentium 4 and Athlon 64 CPUs of that era), and the SPE (FPU) units were capable of running game logic, physics (specifically NVIDIA PhysX), and one SPE was dedicated to running the PS3 OS.
It was very capable of running games, and The Last of Us (2013) was a mind blowing achievement, especially when it was in an era of Ivy Bridge and Haswell quad-core CPUs, not Pentium 4 and Athlon 64 CPUs.

Software, however, did need to be programmed to specifically take advantage of the SPE units.
A few games didn't use them at all, and just used the PPE; I believe one of The Need for Speed games did this, which suffered frame drops in turn.
 
The PPE has no branch predictor, and very limited out of order execution. The SPE’s can only access main memory using DMA. So yes, a terrible architecture for a gaming machine.
 
The PPE has no branch predictor, and very limited out of order execution. The SPE’s can only access main memory using DMA. So yes, a terrible architecture for a gaming machine.
Yep, the PPE was an in-order execution CPU, and the SPE units had to pass-through the PPE for data and instructions, so there were definitely some clear limitations to the architecture of the Cell.
In 2006 when x86-64 dual-cores were only just emerging (and SMP-software was basically non-existent), it was a real powerhouse with many advantages to general-purpose CPUs of that era.
 
Sony originally intended for Cell to do both graphics duty and regular CPU duty. While Cell can be blazing fast on certain types of code: It did not meet performance requirements for general graphics duty. Which forced them to seek out a GPU contract, late in development. Generally speaking, Cell is not a match for RSX in graphics rendering.

However, it still remains that Cell can do things in the graphics pipeline and many games used it to do some pretty cool stuff, which took some load off the RSX GPU and/or added some flourish which may not have been possible with RSX alone.

That's how I remember it as well. They, and their fanboys, really thought the Cell could do it all. Who could've thought that playing tens of high quality video streams does not equal gaming performance...
 
The RSX GPU in the PS3 is from the NVIDIA Series 7, which is right before the Series 8 G80 which featured the first programmable unified-shader architecture, so GPGPU applications, GPU compute, and OpenCL/CUDA will not operate with it.

What was the F@H client running on?
 
article writer needs to put down the crack and meth pipe... and go straight for krokodil
 
Is this similar to the argument of Apple PowerPC RISC processors being so much more powerful than Intel and AMD offerings that Apple had to switch to Intel and AMD just to remain competitive?
 
What was the F@H client running on?
It was using the 1 PPE unit and 6 available SPE units, and would achieve around 1500 PPD, which for back then on anything other than 2P or 4P systems, was quite the achievement.
Once GPGPU applications started to become more available as the years went on, those obviously would achieve more.

In 2008 and 2009, an NVIDIA 9500GT would also be able to achieve 1500 PPD (used to use a few of them back then), and were far more power-efficient than the Cell at those tasks.
This is why I've stated that the Cell was a great stop-gap between conventional CPUs and GPGPUs in terms of functionality, and in 2006 it could not be beaten, especially at its price point.
 
That's how I remember it as well. They, and their fanboys, really thought the Cell could do it all. Who could've thought that playing tens of high quality video streams does not equal gaming performance...
This is the exact reason I laugh every time the FLOPS comparisons are prompted by marketing for the latest consoles and next-gen consoles.
FLOPS ≠ gaming performance
 
This is the exact reason I laugh every time the FLOPS comparisons are prompted by marketing for the latest consoles and next-gen consoles.
FLOPS ≠ gaming performance

We all have our naive views on certain topics. Everyone should be put in other people's shoes from time to time before they lose perspective again.

Anyway. A tractor and a Ferrari are both awesome but not really interchangeable
 
Is this similar to the argument of Apple PowerPC RISC processors being so much more powerful than Intel and AMD offerings that Apple had to switch to Intel and AMD just to remain competitive?

This wasn't arch though that was relying on a CPU from a company that stopped bothering to try and keep up in regard to production nodes ect to companies making CPUs for consumer devices.

IBM didn't see consumer devices as a profitable market so they where not bothering to keep up on low end Power chips. IBM is still producing Power chips for server and super computers. It is still a superior Arch. Power 9 powers the 2 fastest super computers in the world right now... and the next 2 are both using Chinese parts. 7nm Power 10 is due out in 2020. No we won't be putting them in a our PCs... but you can bet your ass the fastest computers in the world will be power 10 for probably 4-5 years.

Apple had to dump Power cause at the time they where 2 process nodes behind Intel meaning they where selling chips running less then half the freq of the x86 chips in PCs of the day... and even at that point power was still superior in CPU intense tasks. Apple made a business decision... they where able to buy cheaper chips with bigger marketing numbers, and IBM was just not planning the type of future chips that would sell well in personal computers. From power 5 or so onward there designs trended toward server architectures and technologies aimed at massive core systems... they introduced VVA (virtual vector arch) which allows multiple CPUs to act as a vector node. They also changed to in order instead our out of order operation... much more geared at super computer loads then home computing, in fact it would have required software recompiles that wouldn't have lead to any real major advantage to the type of tasks run on consumer machines. (so if your going to have to force all your software developers to recompile everything anyway why not switch to x86)

These are the same reasons MS and Sony went x86... IBM took power in a direction that largely put the idea of using one chip in a consumer device in the rear view mirror. Apple MS and Sony really had no choice but to switch to x86.
 
This wasn't arch though that was relying on a CPU from a company that stopped bothering to try and keep up in regard to production nodes ect to companies making CPUs for consumer devices.

While that is true as to why Apple had to switch off PowerPC Apple was very much on the "muh architecture" bandwagon when they switched to PPC and during many years of its use. They crowed on like computer science sophomores about the advantages of RISC over CISC, they talked about how the PPC had a positive second derivative (growth of growth) of MHz, where as x86 had a negative one. Up to the end of its lifecycle, they kept on the "It's BETTAR" train, every new release bringing a new set of contrived Photoshop filter benchmarks to show that no really it is the fastest computer ever made by anyone ever.
 
While that is true as to why Apple had to switch off PowerPC Apple was very much on the "muh architecture" bandwagon when they switched to PPC and during many years of its use. They crowed on like computer science sophomores about the advantages of RISC over CISC, they talked about how the PPC had a positive second derivative (growth of growth) of MHz, where as x86 had a negative one. Up to the end of its lifecycle, they kept on the "It's BETTAR" train, every new release bringing a new set of contrived Photoshop filter benchmarks to show that no really it is the fastest computer ever made by anyone ever.

It should be no shock that a company making anything will tell you that what they are making/using is the bestest best thing ever made. Every company does PR... and everyone looks for the examples of what makes them look good.

Photoshop wasn't contrived... its just that ya for things like filters PPC is still superior to x86 arch wise, in the same way a GPU is better suited to doing that type of work. There is a reason why the blenders and photoshops of the world use GPU compute. Nothing contrived about it. You can find benchmark type reviews of Talos Power 9 based systems vs i9s and threadrippers with a bit of google fu... and you will see power vs x86 is actually a pretty close back and forth. But the powerisa risc arch shines in things like Zip compression and the type of math you find in imagine/video filters ect.

No we won't see power chips ever again in workstations or desktops again (in volume anyway) as GPUs can be tasked to crunch those types of workloads better then any general compute CPU. If GPUs where not a thing power would make much more sense for workstations.
 
Last edited:
Back
Top