GameCube can achieve 10.5 gigaflops, but that is a measurement of the processor, geometry engine, and hardware lighting combined. Specialized processors like graphics chips do more instructions per second than main processors, but are limited in the functions that they can do. </strong><hr></blockquote>
As corrected earlier by someone else here, the 10.5 refers to the Gecko alone (not in addition to the graphics-oriented Flipper unit). If the entire machine (from processor to GPU) was only capable of only 10.5 GFlops, then I would say that would put the GC far behind in capabilities than any of the current game console choices. A decent GPU around today is good for 40-80 GFlops alone. As such, the GPU Flop rating is typically not added to the CPU Flops rating anyway, because the GPU number so easily swamps the CPU number (not to mention that they aren't exactly the "same" kind of Flops you can add together in the first place). If you wanted to make a comparison like that, then you would have to consider that the PS2 and Xbox would end up with 60-80 GFlops rating for total output.
<strong> [quote]The main processor puts out 1.125 gigaflops as compared to 3.7 gigaflops for a single 500MHz G4. It is an impressive processor that blows XBOX away, but not the G4.</strong><hr></blockquote>
I still have a problem believing a 400-ish Mhz G3 can pull off a GFlop all by itself. Maybe 400-600 MFlops, but not an entire GFlop. If it were even possible, then why didn't Apple throw out their "breaking the GFlop barrier" campaign when the BW G3's were out, instead of waiting for the G4's to come out? So this leads me back to my original question- what else is on the Gecko chip die? If they are vector units, why don't they just say that? If they are special purpose multimedia units, why don't they talk more about them? If it's just a few more instructions added to the chip, how can that increase floating-point performance by such brute levels.
(Going back to some other things said in this thread) I don't believe that super fast RAM or on die cache alone can just magically make a G3 hit a GFlop either. If so, then Apple is surely doing something wrong if this kind of innate performance is getting passed up in their desktops just because of memory architecture (because that would mean the G4's must have another 2 to 4x more performance that remains latent because of SDRAM; that is a bit fanciful, don't you guys think?).
I still have a problem believing a 400-ish Mhz G3 can pull off a GFlop all by itself. Maybe 400-600 MFlops, but not an entire GFlop. If it were even possible, then why didn't Apple throw out their "breaking the GFlop barrier" campaign when the BW G3's were out, instead of waiting for the G4's to come out? So this leads me back to my original question- what else is on the Gecko chip die? If they are vector units, why don't they just say that? If they are special purpose multimedia units, why don't they talk more about them? If it's just a few more instructions added to the chip, how can that increase floating-point performance by such brute levels.
(Going back to some other things said in this thread) I don't believe that super fast RAM or on die cache alone can just magically make a G3 hit a GFlop either. If so, then Apple is surely doing something wrong if this kind of innate performance is getting passed up in their desktops just because of memory architecture (because that would mean the G4's must have another 2 to 4x more performance that remains latent because of SDRAM; that is a bit fanciful, don't you guys think?).
but aren't we missing something here? IBM managed to build a highly specialised/fast PowerPC chip that costs considerably less than a G4 chip. Doesn't that mean that IBM, in it's own right, is quite good ad doing CPUs on order and together with it's facilities could serve Apple pretty well with future chips? I seriously think IBM is a better company than Moto when it comes to chip design and implementation.
Could IBM be doing powerpc chips for Apple branded "digital hub devices"?
I don't think it could render 12 million fully rendered polygons per second, with lighting effects, bump mapping etc.</strong><hr></blockquote>
Remove the AGP bus, improve the system memory, remove the operating system, and yes the geForce 4MX could run Rogue Leader at 640x480 (which is the GameCube's resolution). It has hardware T&L (that means "transform and lighting" for those that don't know), at least 4 hardware lights, and 4-way multitexturing. The GameCube's graphics engine is a bit more capable (8-way multitexturing and 8 lights) but not really that much faster or more capable. The high end geForce4 is definitely more capable, except for still only having 4-way multitexturing.
And why do people keep saying that the Gekko is somehow more amazing than the G3 or G4? It is a G3 with a couple of minor tweaks to add features that game developers can take advantage of (but the G4 has much better alternatives). Its beyond me why anyone would say you "can't compare them".
The Gekko is most certainly not capable of 10.5 GFlops all by itself. It runs at 485 MHz, and can issue 1 paired single precision multiply add instruction per clock cycle (peak rate, not sustainable for more than about half a dozen instructions in a row). This means you get 4 "floating point operations" (1 add and 1 multiply on each of 2 floats) per clock, and thus ~1.94 GFlops... except that that rate cannot be sustained due to instruction retirement limitations. About 1 GFlop is believable in some situations, but you have to get doing multiply-add pairs on paired floats and usually you have to load and store them to memory as well. Using GFlops as a benchmark has always been a little fishy. The 7450's AltiVec unit can theoretically issue a 4-way multiply-add every clock cycle and thus is twice as fast as the Gekko at the same clock rate (not to mention the hundred other instructions it has compared to the Gekko).
Game consoles aren't magic, they are computers just like the ones on your desktop, but they are built with a more specific purpose, much lower cost constrains, and no legacy architecture restrictions. They also don't have any other applications running in the background. 5 years ago the dedicated hardware in a game console made it much better at games, but nowadays the same (or better) hardware is available in your PC. It is also fairly normal operating procedure for the hardware manufacturer to take a loss on making the hardware for the first part of the console's life since the important thing is to sell a lot of machines so that people buy a lot of games (where the money is).
BTW: I'm told the "XBox Japan recall" is just an unfounded rumour.
OK, that sounds plausible, but then what else is to account for the other 9.5 GFlops cited for the Gecko (you would think that would get a little more attention, as 9.5 GFlops is certainly not a trivial amount of resources)? More vector units? Something else? If it is a SIMD unit, what in the world is IBM doing with it? I thought SIMD just wasn't IBM's "bag". How in the world did they get away with putting a SIMD on a G3 die, and not come uncomfortably close to licensing conflicts with Motorola's G4 design (not saying they are the same, just that doing so has got to cause unnecessary tension with Motorola no matter what the intentions are)? This just raises so many more questions...
Yes, this is true. The current generation is interesting because (in the case of GC and Xbox) they are using well-known, mainstream chips for their CPU's (so it is inevitably interesting to see how well they work in these systems vs. their desktop PC counterparts). The Gecko in the GC is even more intriguing in that it incorporates an updated G3 core along with some other "mystery" hardware with a rather lofty floating point rating. So that is where my interest stems.
Two things: First, I have to chime in that I've been underwhelmed with all the XBOX games I've seen. AFA graphics power, I have a feeling that the box has considerable power, but the developers are coming nowhere near maxing it out.
2, the reason Quake is the defacto benchmark for games is that it is a CPU/RAM/GPU pig. It will suck up all available resources, and beg for more. Thus, if it flies on a given machine, it means that machine has a lot of muscle.
I mean, Q3A still sucks up your resources, and the games 3-4 years old. Pretty greedy little piece of code, that.
Problem is, the Quake engine was designed from the ground up to run in an x86 processor environment (of 3 years ago?). That means a PIII. That means no vector processing or SIMD type of activity. The engine is clearly focused around a Single-Instruction-Single-Data (vs. SIMD) scenario with regard to how it operates. It's a resource hog because doing a long sequence of essentially the same operation "one-by-one" is extremely exhaustive (vs. utilizing parallelism) so the general CPU naturally has to overcome this via brute force x87 activity and brute force clockspeeds. As mentioned before, it's nontrivial to just revise the engine to take advantage of SIMD w/o having to rewrite major areas of the code (you rewrite one part to handle 4-pieces of data at a time and you find that some place else becomes the bottleneck and needs to be rewritten to handle 4 at a time). You might as well just rewrite the engine from the ground up, and that is exactly what must happen to really take advantage of newer processors such as the P4, G4, and any console that relies extensively on vector processing strategies. Essentially, the Quake engine running on today's processors is like running it on yesteryear's processors but ignoring the videocard as a source of graphics acceleration (albeit, a GPU takes parallelism and dedicated functionality to even more of an extreme than vector units).
I don't think it could render 12 million fully rendered polygons per second, with lighting effects, bump mapping etc.</strong><hr></blockquote>
No clue about the 4MX...but the Ti can do 1.23 trillion operations per second, 4.9 billion textured pixels per second, and 87 million polygons per second.
<strong>The Gecko in the GC is even more intriguing in that it incorporates an updated G3 core along with some other "mystery" hardware with a rather lofty floating point rating. So that is where my interest stems.
</strong><hr></blockquote>
There is no "mystery" hardware in the GameCube! The Gekko "SIMD" unit can only be barely called that... it has a few limited instructions to operate on pairs of numbers. It is the most limited SIMD unit I have ever seen -- although given how little they added to the chip, they chose what to add wisely.
I don't know where the supposed 10.5 GFlops rating came from, but I can only assume they meant the graphics chip (Flipper) -- but I'd be surprised if the Flipper's capability was that low. As mentioned above, modern graphics chips usually have astoundingly high GFlop claims (TFlops in the case of the geForce4).
That contract is only for the G4, when Moto can't meet their production requirement for the G4, then Apple can use IBM to fabricate G4's for Apples use. It does not keep Apple from using G3's, or future processors from IBM.</strong><hr></blockquote>
Again, there is no such contract. A Motorola official took the unusual step of issuing an on the record denial of this last year.
<strong>but aren't we missing something here? IBM managed to build a highly specialised/fast PowerPC chip that costs considerably less than a G4 chip. Doesn't that mean that IBM, in it's own right, is quite good ad doing CPUs on order and together with it's facilities could serve Apple pretty well with future chips? I seriously think IBM is a better company than Moto when it comes to chip design and implementation.
</strong><hr></blockquote>
Erm, no. First of all, price is pretty much determined on how many you can sell - and I suspec that there will be more GameCubes sold than Macs this year What's more, it's not even certain that the processor is that cheap - don't forget that consoles are sold at a big loss, with the profits being made on what are effectively royalties on games.
What's more, the GameCube processor is hardly cutting-edge technology. That's not to say it's not great for the application it's used for, but it's not cutting edge, nowhere near it.
The technology in the Xbox is old aswell, it doesn't have anything near a GeForce Ti4600 in it, thats just ridiculous. All what it has is an nForce motherboard, which has GeForce 2 speeds with a stuck on nFinite FX engine on it, or whatever its called that is meant to impress the average person just by the name.
And anyway, all the sites I have been to, do not have all the Gamecube specifications, so programmer, I would like you to share with is where you get this information from??
<strong>The technology in the Xbox is old aswell, it doesn't have anything near a GeForce Ti4600 in it, thats just ridiculous. All what it has is an nForce motherboard, which has GeForce 2 speeds with a stuck on nFinite FX engine on it, or whatever its called that is meant to impress the average person just by the name.
And anyway, all the sites I have been to, do not have all the Gamecube specifications, so programmer, I would like you to share with is where you get this information from??</strong><hr></blockquote>
Where'd you get your information on the XBox? It is wrong. The XBox does not use the nForce chipset, and it does contain the nv25 which was just released as the geForce4 Ti.
Where do I get my information from? Well, I'm a programmer, so I'll give you 3 guesses as to which 3 machines I've been writing programs for in the last couple of years. Besides, I have seen this information discussed on the web in various places, so perhaps you are just visiting the wrong sites.
I think the X-box GPU is a not quite GF4Ti (but very close).
In ascending (or descending order, or whatever's getting more powerful on the way down) order:
GF2 = nv15
GF4MX = nv17
GF3 = nv20
X-Box GPU = nv2A which I've read described as about nv22 ish.
GF4Ti = nv25
Keep in mind that I know nothing about chips and design but for what I read, and that is that the x-box part and the GF4Ti are 'very' close, but that the 4Ti is a slightly improved/evolved part.
OTOH, I think that anyone who starts to talk tech on these boards ought to state their credentials as I have done -- I use computers, but their insides may as well be magic hampters for all I know.
NVidia GPU line is (performance and feature wise):
geforce2
geforce4mx
geforce3
xbox
geforce4
the xbox indeed comes close to the geforce4 in performance simply because it has a more dedicated hardware around it. but then, you can stick 1gb rdram into your overclocked 3ghz p4 system and xbox eats dust.
the nforce chipset on its own is a derivate of the chipsed used in xbox (nvida tought "hey, we got plenty experience now we finished xbox, why not make money out of it and make a chipset for PCs?"). nforce has geforce2mx onboard but the xbox doesn't - it runs a "enhanced geforce3" so to speak.
only what I gathered around anandtech and similiar sites. and I hope i remember it correctly. )
Comments
<strong>
GameCube can achieve 10.5 gigaflops, but that is a measurement of the processor, geometry engine, and hardware lighting combined. Specialized processors like graphics chips do more instructions per second than main processors, but are limited in the functions that they can do. </strong><hr></blockquote>
As corrected earlier by someone else here, the 10.5 refers to the Gecko alone (not in addition to the graphics-oriented Flipper unit). If the entire machine (from processor to GPU) was only capable of only 10.5 GFlops, then I would say that would put the GC far behind in capabilities than any of the current game console choices. A decent GPU around today is good for 40-80 GFlops alone. As such, the GPU Flop rating is typically not added to the CPU Flops rating anyway, because the GPU number so easily swamps the CPU number (not to mention that they aren't exactly the "same" kind of Flops you can add together in the first place). If you wanted to make a comparison like that, then you would have to consider that the PS2 and Xbox would end up with 60-80 GFlops rating for total output.
<strong> [quote]The main processor puts out 1.125 gigaflops as compared to 3.7 gigaflops for a single 500MHz G4. It is an impressive processor that blows XBOX away, but not the G4.</strong><hr></blockquote>
I still have a problem believing a 400-ish Mhz G3 can pull off a GFlop all by itself. Maybe 400-600 MFlops, but not an entire GFlop. If it were even possible, then why didn't Apple throw out their "breaking the GFlop barrier" campaign when the BW G3's were out, instead of waiting for the G4's to come out? So this leads me back to my original question- what else is on the Gecko chip die? If they are vector units, why don't they just say that? If they are special purpose multimedia units, why don't they talk more about them? If it's just a few more instructions added to the chip, how can that increase floating-point performance by such brute levels.
(Going back to some other things said in this thread) I don't believe that super fast RAM or on die cache alone can just magically make a G3 hit a GFlop either. If so, then Apple is surely doing something wrong if this kind of innate performance is getting passed up in their desktops just because of memory architecture (because that would mean the G4's must have another 2 to 4x more performance that remains latent because of SDRAM; that is a bit fanciful, don't you guys think?).
[ 03-09-2002: Message edited by: Randycat99 ]</p>
<strong>
I still have a problem believing a 400-ish Mhz G3 can pull off a GFlop all by itself. Maybe 400-600 MFlops, but not an entire GFlop. If it were even possible, then why didn't Apple throw out their "breaking the GFlop barrier" campaign when the BW G3's were out, instead of waiting for the G4's to come out? So this leads me back to my original question- what else is on the Gecko chip die? If they are vector units, why don't they just say that? If they are special purpose multimedia units, why don't they talk more about them? If it's just a few more instructions added to the chip, how can that increase floating-point performance by such brute levels.
(Going back to some other things said in this thread) I don't believe that super fast RAM or on die cache alone can just magically make a G3 hit a GFlop either. If so, then Apple is surely doing something wrong if this kind of innate performance is getting passed up in their desktops just because of memory architecture (because that would mean the G4's must have another 2 to 4x more performance that remains latent because of SDRAM; that is a bit fanciful, don't you guys think?).
[ 03-09-2002: Message edited by: Randycat99 ]</strong><hr></blockquote>
May be the Gflop come from the limited SIMD unit of the Gekko.(gekko equal G3 and a SIMD unit).
However i will not replace my G4 for a Gekko ...
[ 03-09-2002: Message edited by: powerdoc ]</p>
Compare a P3 733 in the Xbox vs a P3 733 in a PC if you want, as those chips are identical, but the gekko is neither fish nor bird.
G-news
Could IBM be doing powerpc chips for Apple branded "digital hub devices"?
G-News
<strong>Could a GeForce 4MX run Rogue Leader??
I don't think it could render 12 million fully rendered polygons per second, with lighting effects, bump mapping etc.</strong><hr></blockquote>
Remove the AGP bus, improve the system memory, remove the operating system, and yes the geForce 4MX could run Rogue Leader at 640x480 (which is the GameCube's resolution). It has hardware T&L (that means "transform and lighting" for those that don't know), at least 4 hardware lights, and 4-way multitexturing. The GameCube's graphics engine is a bit more capable (8-way multitexturing and 8 lights) but not really that much faster or more capable. The high end geForce4 is definitely more capable, except for still only having 4-way multitexturing.
And why do people keep saying that the Gekko is somehow more amazing than the G3 or G4? It is a G3 with a couple of minor tweaks to add features that game developers can take advantage of (but the G4 has much better alternatives). Its beyond me why anyone would say you "can't compare them".
The Gekko is most certainly not capable of 10.5 GFlops all by itself. It runs at 485 MHz, and can issue 1 paired single precision multiply add instruction per clock cycle (peak rate, not sustainable for more than about half a dozen instructions in a row). This means you get 4 "floating point operations" (1 add and 1 multiply on each of 2 floats) per clock, and thus ~1.94 GFlops... except that that rate cannot be sustained due to instruction retirement limitations. About 1 GFlop is believable in some situations, but you have to get doing multiply-add pairs on paired floats and usually you have to load and store them to memory as well. Using GFlops as a benchmark has always been a little fishy. The 7450's AltiVec unit can theoretically issue a 4-way multiply-add every clock cycle and thus is twice as fast as the Gekko at the same clock rate (not to mention the hundred other instructions it has compared to the Gekko).
Game consoles aren't magic, they are computers just like the ones on your desktop, but they are built with a more specific purpose, much lower cost constrains, and no legacy architecture restrictions. They also don't have any other applications running in the background. 5 years ago the dedicated hardware in a game console made it much better at games, but nowadays the same (or better) hardware is available in your PC. It is also fairly normal operating procedure for the hardware manufacturer to take a loss on making the hardware for the first part of the console's life since the important thing is to sell a lot of machines so that people buy a lot of games (where the money is).
BTW: I'm told the "XBox Japan recall" is just an unfounded rumour.
<strong>
May be the Gflop come from the limited SIMD unit of the Gekko.(gekko equal G3 and a SIMD unit).
However i will not replace my G4 for a Gekko ...
[ 03-09-2002: Message edited by: powerdoc ]</strong><hr></blockquote>
OK, that sounds plausible, but then what else is to account for the other 9.5 GFlops cited for the Gecko (you would think that would get a little more attention, as 9.5 GFlops is certainly not a trivial amount of resources)? More vector units? Something else? If it is a SIMD unit, what in the world is IBM doing with it? I thought SIMD just wasn't IBM's "bag". How in the world did they get away with putting a SIMD on a G3 die, and not come uncomfortably close to licensing conflicts with Motorola's G4 design (not saying they are the same, just that doing so has got to cause unnecessary tension with Motorola no matter what the intentions are)? This just raises so many more questions...
[ 03-09-2002: Message edited by: Randycat99 ]</p>
2, the reason Quake is the defacto benchmark for games is that it is a CPU/RAM/GPU pig. It will suck up all available resources, and beg for more. Thus, if it flies on a given machine, it means that machine has a lot of muscle.
I mean, Q3A still sucks up your resources, and the games 3-4 years old. Pretty greedy little piece of code, that.
[ 03-09-2002: Message edited by: Randycat99 ]</p>
<strong>Could a GeForce 4MX run Rogue Leader??
I don't think it could render 12 million fully rendered polygons per second, with lighting effects, bump mapping etc.</strong><hr></blockquote>
No clue about the 4MX...but the Ti can do 1.23 trillion operations per second, 4.9 billion textured pixels per second, and 87 million polygons per second.
<strong>The Gecko in the GC is even more intriguing in that it incorporates an updated G3 core along with some other "mystery" hardware with a rather lofty floating point rating. So that is where my interest stems.
</strong><hr></blockquote>
There is no "mystery" hardware in the GameCube! The Gekko "SIMD" unit can only be barely called that... it has a few limited instructions to operate on pairs of numbers. It is the most limited SIMD unit I have ever seen -- although given how little they added to the chip, they chose what to add wisely.
I don't know where the supposed 10.5 GFlops rating came from, but I can only assume they meant the graphics chip (Flipper) -- but I'd be surprised if the Flipper's capability was that low. As mentioned above, modern graphics chips usually have astoundingly high GFlop claims (TFlops in the case of the geForce4).
<strong>
Oh? That's a new one - I didn't know they were forced to use Moto. </strong><hr></blockquote>
You didn't know it because they aren't. I don't know where he's getting this from, but it's high-grade BS.
Come on SpiffyGuyC, where's the source for this?
<strong>
That contract is only for the G4, when Moto can't meet their production requirement for the G4, then Apple can use IBM to fabricate G4's for Apples use. It does not keep Apple from using G3's, or future processors from IBM.</strong><hr></blockquote>
Again, there is no such contract. A Motorola official took the unusual step of issuing an on the record denial of this last year.
<strong>but aren't we missing something here? IBM managed to build a highly specialised/fast PowerPC chip that costs considerably less than a G4 chip. Doesn't that mean that IBM, in it's own right, is quite good ad doing CPUs on order and together with it's facilities could serve Apple pretty well with future chips? I seriously think IBM is a better company than Moto when it comes to chip design and implementation.
</strong><hr></blockquote>
Erm, no. First of all, price is pretty much determined on how many you can sell - and I suspec that there will be more GameCubes sold than Macs this year
What's more, the GameCube processor is hardly cutting-edge technology. That's not to say it's not great for the application it's used for, but it's not cutting edge, nowhere near it.
And anyway, all the sites I have been to, do not have all the Gamecube specifications, so programmer, I would like you to share with is where you get this information from??
<strong>The technology in the Xbox is old aswell, it doesn't have anything near a GeForce Ti4600 in it, thats just ridiculous. All what it has is an nForce motherboard, which has GeForce 2 speeds with a stuck on nFinite FX engine on it, or whatever its called that is meant to impress the average person just by the name.
And anyway, all the sites I have been to, do not have all the Gamecube specifications, so programmer, I would like you to share with is where you get this information from??</strong><hr></blockquote>
Where'd you get your information on the XBox? It is wrong. The XBox does not use the nForce chipset, and it does contain the nv25 which was just released as the geForce4 Ti.
Where do I get my information from? Well, I'm a programmer, so I'll give you 3 guesses as to which 3 machines I've been writing programs for in the last couple of years. Besides, I have seen this information discussed on the web in various places, so perhaps you are just visiting the wrong sites.
In ascending (or descending order, or whatever's getting more powerful on the way down) order:
GF2 = nv15
GF4MX = nv17
GF3 = nv20
X-Box GPU = nv2A which I've read described as about nv22 ish.
GF4Ti = nv25
Keep in mind that I know nothing about chips and design but for what I read, and that is that the x-box part and the GF4Ti are 'very' close, but that the 4Ti is a slightly improved/evolved part.
OTOH, I think that anyone who starts to talk tech on these boards ought to state their credentials as I have done -- I use computers, but their insides may as well be magic hampters for all I know.
[ 03-10-2002: Message edited by: Matsu ]</p>
NVidia GPU line is (performance and feature wise):
geforce2
geforce4mx
geforce3
xbox
geforce4
the xbox indeed comes close to the geforce4 in performance simply because it has a more dedicated hardware around it. but then, you can stick 1gb rdram into your overclocked 3ghz p4 system and xbox eats dust.
the nforce chipset on its own is a derivate of the chipsed used in xbox (nvida tought "hey, we got plenty experience now we finished xbox, why not make money out of it and make a chipset for PCs?"). nforce has geforce2mx onboard but the xbox doesn't - it runs a "enhanced geforce3" so to speak.
only what I gathered around anandtech and similiar sites. and I hope i remember it correctly.