From a floating point standpoint, the Emotion Engine as a whole, is rightfully more capable than a 733MHz Pentium III. It's in the same vein as the Cell BE vs the Xenon CPU in the 360. On one side you had a GFLOPS monster that went underutilized (EE & Cell) vs a more practical processor (Pentium III & Xenon) with a more general processor setup and therefore easier for devs to grasp.
PS2 EE question
- Thread starter vipa899
- Start date
Except that in the case of EE the VU's were mainly used to do what the NV2A does. I'm sure there's some games that use the VUs more creatively, of course, but the competing consoles had much of the geometry processing done in hardware on the graphics side where PS2 had the VU's to carry the load. I think EE was less underutilized in that at least one of the VUs was used most of the time. I'm not sure most games used half the SPEs in Cell, however.
I think EE was less underutilized in that at least one of the VUs was used most of the time. I'm not sure most games used half the SPEs in Cell, however.[/QUOTE]
VU0 was famously underutilized, but was difficult to use correctly. Unlike the EE, the Cell was less explicit about how devs were to use the SPEs, since they were all symmetrical, with the same cache and capabilities. It's definitely an approach that would've benefited the Emotion Engine, but I'm sure die area, layout, and wattage played a role in the EE's scheme that resulted in the partitioned approach. It seems to me that making both Vector Units direct coprocessors to the MIPS core would've made more sense, but perhaps would've required a wider connection to the Graphics Synthesizer, and perhaps an EFU on each VU as well. On a single die already packing the main core, an FPU, 2x VUs, an MPEG2 Decoder (IPU), and DMA controller............fitting more things was likely a problem.
It was an interesting time for microprocessors and graphics. PS2 was just a victim of how fast graphics rendering pipelines were going towards very specialized silicon (GPUs). I do wonder if Sony was sorta pooping it's pants when the Dreamcast debuted. Yeah it was less capable overall but it had a more full bodied pipeline for graphics rendering, specifically texture mapping units and surely Sony was paying attention to that in reference to the rest of the industry.
As said by others EE is half the graphics for the most. VU0 is about 3.08gflops, 2.44 for vu0 and 0.64 for the FPU, 0.450 gflops for the mips core? That would be around 3.5Gflops for the parts thats used together as a cpu for most games. I really cant find exact numbers for the P3 733, some say a little over 3gflops, some sites claim a 1ghz p3 is around 2Gflops. Anyone know?
Atleast PS2 would have the CPU with more gflops, how important that is or if im counting correct is another matter
Would having VU0 in micromode give it edge over the P3?
@Grall
Are you sure networking for the og xbox was done through the CPU? Found this article, of the many, that the MCP-X performs processing network functions, audio and peripherals.
http://www.nvidia.com/object/IO_20010530_6177.html
Yeah, overall, barring any corner cases it pretty much owned the other consoles' CPUs.
The P3 was much higher clocked than any of the other consoles' CPUs, had a fairly roomy and quite fast L2 cache (256 bit read/write per clock IIRC), it was out-of-order execution capable and superscalar (which out of the others, only the Gamecube's PPC CPU was somewhat out-of-order, but not as advanced as the P3.) Plus, the SSE SIMD FP unit didn't require any jumping through hoops like the vector units of the EE did; it ran ordinary program code, like the integer ALUs or the x87 FPU did.
If you're counting the EE with both vector units it can crunch out over twice as many flops as the pentium(and more than 3x the gamecube). But that's not really a 1:1 is it, since a large portion of, if not all, the VU1 work would be something you'd use the nvidia chip for in the xbox.
Honestly, I've forgotten. It was (what I considered) a reasonable assumption on my part, as hardware-assisted networking wasn't all that common in the pre-gigabit days at least. Then there's the question of what exactly 'networking functions' entail; OSI network model has many layers; you could still claim accelerated networking functions even if only a few of the layers are hardware accelerated.
On paper, yeah. But in real life, the P3 would have been easier to harness. VU0 in particular was infamously difficult to use efficiently (or even at all!), from what people have described.
Paper specs rarely tell the whole story, especially when it comes to wacky architectures like PS2.
@Grall
Yes its a reasonable assumption, have no clue how its done on the xbox mcp, could be Nvidia PR.
Do you have any figures of how many Gflops the P3 733 is? Many say around 3Gflops, yet i find figures around the web that claim 2Gflops for 1Ghz p3. Theres really not much if anything to find about those cpus. Im sure it can be calculated, just dont know how
Also, GS+VU1, i assume NV2A chip was more capable?
How would someone code something like this or zoe2 on xbox, vertex shaders?
Gamasutra did an interview with the SOE devs, mostly about the CPU's of the two systems.
https://www.gamasutra.com/view/feature/131296/porting_a_ps2centric_game_to_the_.php
Yes its a reasonable assumption, have no clue how its done on the xbox mcp, could be Nvidia PR.
Do you have any figures of how many Gflops the P3 733 is? Many say around 3Gflops, yet i find figures around the web that claim 2Gflops for 1Ghz p3. Theres really not much if anything to find about those cpus. Im sure it can be calculated, just dont know how
Also, GS+VU1, i assume NV2A chip was more capable?
How would someone code something like this or zoe2 on xbox, vertex shaders?
Gamasutra did an interview with the SOE devs, mostly about the CPU's of the two systems.
https://www.gamasutra.com/view/feature/131296/porting_a_ps2centric_game_to_the_.php
It's a pretty damn impressive demo, showing the potential of procedural rendering capabilities that came with higher floating/vector performance. I'm gonna go out on a limb and say that NV2A probably doesn't have all the necessary functions to handle this same kind of procedural rendering, at least without more direct involvement of the X-CPU. It would take a new generation of systems to really make procedural rendering practical, and MS had it in mind with the Xenon CPU to save on memory footprint.
Last edited:
Honestly, I've forgotten. It was (what I considered) a reasonable assumption on my part, as hardware-assisted networking wasn't all that common in the pre-gigabit days at least. Then there's the question of what exactly 'networking functions' entail; OSI network model has many layers; you could still claim accelerated networking functions even if only a few of the layers are hardware accelerated.
On paper, yeah. But in real life, the P3 would have been easier to harness. VU0 in particular was infamously difficult to use efficiently (or even at all!), from what people have described.
Paper specs rarely tell the whole story, especially when it comes to wacky architectures like PS2.
That's micro-mode that was difficult to use. Macro-mode from what i understand was much simpler. It just didn't show up on the performance analyzer, which is where the whole idea of VU0 being extremely underutilized comes from.
https://forum.beyond3d.com/posts/184657/
PS2 also had 2MB sound memory and I/O Memory 2MB EDO DRAM.
PS2 also had 2MB sound memory and I/O Memory 2MB EDO DRAM.
Xbox system management controller has its own ram, rom running at 20mhz. Dont know if theres more ram to be found somewhere, but seeing how important this is to general performance, its not even worth mentioning
http://xboxdevwiki.net/System_Management_Controller
Yes, this is very common for PCs, they typically do the kind of stuff listed on that xbox wiki, IE handle sleep mode and wake on input, handle soft-off mode, monitor buttons, manage fan speeds and so on. These microcontrollers are very slow processors with tiny amounts of on-chip memory/flash, optimized for low power draw. Their function is not really comparable to the PS2 IO processor, which is a full-fledged CPU; although of an old, slow model.
Interestingly, PS2 Slim models had a different, somewhat faster and more recent IOP, but the speed advantage may have been largely eaten up by emulation software layer I dunno; I've never seen any dissection of this difference in the two hardwares.
@Grall
Okej but it isnt direcly something performance enhancing in games or something The IOP is more of a bottleneck since its ultra-slow to transfer over network, it also bottlenecks the internal HDD if i understood right. Its really a pain to copy larger files over ftp, USB1.1 port seems faster, also limits streaming games over network. Tried everything but its almost better dismounting the HDD and use an external drive-case
Do you know if that VUniverse demo would be possible on the xbox? Its not that i dont believe what was said but on the other hand i understand the demo was running on just VU1 according to the description, and im not seeing that much impressive things happening, or is it the beginning with all the stars and planets?
Also, does anyone know how many Gflops the P3 733 is?
Okej but it isnt direcly something performance enhancing in games or something
The IOP is more of a bottleneck since its ultra-slow to transfer over network, it also bottlenecks the internal HDD if i understood right. Its really a pain to copy larger files over ftp, USB1.1 port seems faster, also limits streaming games over network. Tried everything but its almost better dismounting the HDD and use an external drive-case Do you know if that VUniverse demo would be possible on the xbox? Its not that i dont believe what was said but on the other hand i understand the demo was running on just VU1 according to the description, and im not seeing that much impressive things happening, or is it the beginning with all the stars and planets?
Also, does anyone know how many Gflops the P3 733 is?
Tech demos often push hardware as far as it can go in its strongest direction, and it's possible a generally much stronger piece of hardware might not be able to exactly match this in exactly the same ways.
I don't think anyone here can really know if a particular PS2 demo falls into this category, all we can say for sure is that with a similar level of effort, you'd probably see far more Xbox tech showcases greatly exceeding PS2 than vice versa.
And in terms of real games, well ... the procedurally generated utopia is as far away today as it was the early 90s when I first saw "woah!!" demos on Amiga that had no bearing on upcoming games.
I don't think anyone here can really know if a particular PS2 demo falls into this category, all we can say for sure is that with a similar level of effort, you'd probably see far more Xbox tech showcases greatly exceeding PS2 than vice versa.
And in terms of real games, well ... the procedurally generated utopia is as far away today as it was the early 90s when I first saw "woah!!" demos on Amiga that had no bearing on upcoming games.
@function
Ok thats well said, understand it. What part of that demo was procedural, the part with all the planets/stars? I assume thats possible on an xbox, as it isnt that special?
Hows your take on this article, had PS2 advantages, im not fully understanding what they say:
https://www.gamasutra.com/view/feature/131296/porting_a_ps2centric_game_to_the_.php
Asked before but how many Gflops does a P3 733 have/capable of? Also is NV2A more capable then GS+VU1?
Ok thats well said, understand it. What part of that demo was procedural, the part with all the planets/stars? I assume thats possible on an xbox, as it isnt that special?
Hows your take on this article, had PS2 advantages, im not fully understanding what they say:
https://www.gamasutra.com/view/feature/131296/porting_a_ps2centric_game_to_the_.php
Asked before but how many Gflops does a P3 733 have/capable of? Also is NV2A more capable then GS+VU1?
The entire demo was procedurally generated. It had to fit entirely in 16k of memory.
I don't have a "take" on that article. It's explicit. They took a massively PS2 optimised game and assumed they could throw it at the Xbox with unoptimised high level DX code and it'd be faster. It wasn't, and they had to use their talents to optimise. But when they did they had much greater performance..
The article is two pages, very well written, and explicit about their experiences (it also tallies with ERP's findings after which he drastically reduced DX overheads using an approach not available for the PC).
Probably.