I mean hey its the damn right size for an earth grade atmospherre... its just a little on the warm side... so you have to be careful what you do.
Venus has no angular momentum. Mars does. Venus is also far too close to the Sun. I'm not even sure a water cycle is even possible on Venus anymore, but if it is, it won't be for very long. So if you try to cool down Venus so that water condenses, the Sun's radiation will just dissociate the water vapor into hydrogen and oxygen, and the hydrogen will escape into space.
This is Earth's fate in a few hundred million years. Venus isn't really practical to terraform, unless you move it out to Mars's orbit and smack Mars into it... which sounds a lot like how the Earth was formed.
How so? I would think if backward time travel is going to work, it's still going to be continuous.
Depends on the method. If you multiply connect space (like a wormhole) then no, it will be instantaneous. You'll just multiply connect space - there'll be two ways to get from 1985->2005, for instance - one that involves sitting still and letting time pass (so that path would have a distance of ds^2 = (cdt)^2 where dt is 20 years, and c is the speed of light) and one involves going through the wormhole (so that path would have a distance of ds^2 = (dx^2+dy^2+dz^2) where dx,dy,dz are the physical distance through the wormhole).
Then, technically, both the monkey and the wormhole pop out of thin air, and then a little while later, the monkey goes into the wormhole and it disappears. Mass, energy are all conserved because the wormhole is essentially an anti-monkey.
If you just somehow manage to exceed the speed of light, then it will be continuous. This sort of time travel is what goes on inside the event horizon of a black hole (because the time coordinate now is the same signature as the theta and phi coordinate, and the radial coordinate is the "move only in one direction" i.e. inward).
If you go back, are you physically prevented from firing the gun or will the gun misfire?
No. Nothing weird will happen. It just didn't happen. It won't happen because it didn't happen. The reason for your confusion is that you're assuming that you actually do have control over your future.
It's like trying to stop thunder after you've seen the lightning. Nothing you can do will stop it. It's already happened.
Or if you make a change, does the timeline establish a new universe with the old one running along merrily as a parallel universe.
In the model of the Universe they're proposing, there are no parallel universes.
so is it possible that all those other postulated dimensions (to 11) give the degrees of freedom to allow bifurcations in the timeline?
No - no, no, no, no, no. You do not need an additional dimension to allow the degree of freedom. You already have it. It's proper distance - that is, the (3+1) spatiotemporal dimensions.
Why? Because you had to travel a distance to go there - the distance from your future to the past of this (now 'parallel') universe. There's a causal relation between something you did in universe X to something that happened in universe Y, which means there's a proper distance separation.
Once you realize that, however, the whole concept of changeable timelines basically becomes very ad-hoc - you have two separate universes that have absolutely no reason to be identical but which are. You've also essentially made it so that the total energy content of the Universe is infinite.
We could kill two birds with one stone and suck the CO2 out of the Venus atmosphere, and drop it into the Martian atmosphere. Voila! Thin out Venus, and thicken Mars. Two planets for the price of one.
Or we could just follow the Solar System's first example, and move Venus out to Mars's orbit, and crash Mars into it.
OK, OK, it'd take a few million years, but hell, no work needed afterwards. Nice way to eke a few more billion years out of our Sun after the increasing solar luminosity starts boiling away the seas on Earth in a few hundred million years.
which as I'm sure you know, is the layer of our atmosphere that is the most important in blocking ultraviolet radiation.
If you're capable of increasing the atmospheric density by a factor of a thousand, you're capable of reseeding the upper layers of the atmosphere continually.
Mars also has the nice benefit that ultraviolet radiation is already down by a factor of 3. If you want to be super clever, you could follow Kim Stanley Robinson's suggestion and drop an orbiting mirror to increase the solar insolation. Clever use of reflecting material would mean more sunlight, but the same level of UV.
It will only really work with the tougher Iron asteroids, though, the weaker "rubble piles" won't work.
And big ones, too. Smaller ones will have far too much Coriolis force and too much of a vertical gravity gradient for people not to get nauseous, especially if you want tall buildings. Tens of kilometers is probably the minimum.
And, of course, the bigger you get, the bigger a job hollowing it out is.
Basically all this article is saying is that all time travel must consist of closed timelike loops. That is, you "fulfill" the present, rather than altering it. This isn't news - it's the only kind of timelike loop that can exist in GR anyway. The difference here is that quantum mechanics also forces them to be the only ones that exist.
Point of note, however: as far as I know, we don't actually have the math to deal with the formation of a topological change in a surface (i.e., the "alteration" of a timeline). This is very much akin to a wave crashing - fluid dynamics works up until the exact point when the top of the wave touches the rest of the ocean. After that point, the math breaks down. So it's a little difficult to say "X isn't possible, because the math won't allow it" when theorists are in fact only using math that won't allow it. So it's moderately circular. That's GR. In QM, we don't actually have the math that deals with the collapse of the wavefunction (the 'measurement'), and so again, it's moderately circular. If you instead suppose that the wavefunction doesn't actually collapse, then of course you can change the past - you just end up following a different course in probability the second time around.
Examples of closed timelike loops actually are more common than you think in modern scifi/fantasy. Harry Potter and the Prisoner of Azkaban had an excellent example of a CTL, and the timetravel used in Anne McCaffrey's Pern series are entirely CTLs. This leads to statements like "I know I can do this, because I've done it already."
The problem with CTLs is that they muck with certain people's belief in free will.
If, for example, you knew a picture would be taken, you could reflect light from your body and appear in that picture, thereby altering the future.
Assuming you didn't exist in the picture before you went back in time.
And it just moves up from there for all other physical effects. Nothing touched, no air breathed, no light disturbed, nothing.
Unless it was already disturbed to begin with.
Again, there's no real logical problem here. Just the fact that you would have to disassociate yourself from the fact that all of your future actions are possible.
I don't necessarily agree with time travel. A closed timelike loop is essentially the equivalent of a monkey popping out of thin air, and then disappearing a few seconds later. It seems idiotic, and completely counter to all natural laws. But that wouldn't be the first time nature did that to us.
The only redeeming factor that I see to this hardware is that it's an inexpensive LCD.
Plus a cheap MP3 player. It's probably got just enough oomph to be an Ogg player as well. The LCD screen actually makes it so that you can actually have a decent GUI.
Other than that, let's see... with a serial port, you could easily make a universal remote control with LCD screen, a digital picture frame (as seen above), or a status monitor for X (where X is some other project that spits out serial data).
Of course, if you want to be really stupid, a 66 MHz ARM7 certainly has the capability to emulate an NES.
In my case, it looks like a really, really nice front end to the Motorola GPS unit I have (spits out only serial data).
It's like picking up cool pieces of plastic and cardboard when you're a kid because you swear that you'll find a brilliant use for something so neat. Then it sits in the closet until you're an adult and throw it out.
Bah! You just described my basement, except for the fact that I do find uses for said objects.:)
Otherwise you can custom build a similar device for only slightly more
C'mon, it's $12! No way you could get the parts and board for that little.
Actually, it is. The chip you've just described is an "all-in-one" hardware design. The catch is that it seems like a flexible all-in-one design. And with an LCD screen, what could be better?
We have different opinions of "all in one", apparently. If you've got some sort of embedded PIC with OTP ROM, and it's embedded in epoxy, yah, you're probably done (hence "all-in-one" - this is more of a "all-in-three-or-four"). But this has a lot of hackability in it, as you've got easy access to all of the address and chip selects, and some versions of the board even have a moderately replacable ROM.
Regarding the video: oh, definitely. The video potential of the device sucks. But I wouldn't expect anything else for $12.
The SDCard connector, however, is an add-on that is not all that cheap in of itself.
It's just a slot adapter. Grab an SD socket from Molex (either free, $4 inside another order, or $10 from Sparkfun) and wire it up directly, as the pinout is already available.
Plus there's serial input, as well. It's disabled in the default build, so you have to replace the ROM (or via JTAG), but you'd probably want to do that anyway.
There are always a few exceptions, and the JuiceBox looks like it's one of those. Probably for exactly the same reason that it failed in the marketplace as well - a $50 (questionably useful) toy is a bit too much.
The JuiceBox isn't one of the cheap, one-in-all chips: it's an S3C44B0X 66 MHz ARM7TDMI processor with 8M of RAM onboard (some have 2M) and an SD slot (via a custom connector, but you could make an adapter for roughly free by sampling the connector from Molex) with a uClinux distribution. Plus, JTAG pins are available on the far side. Hacking it into anything you want would be relatively simple.
This isn't exactly akin to a "toy". And it wasn't supposed to be $12 - but clearance sales are wonderful things, especially when they involve devices with LCD screens.
People pay into social security with the agreement that they will draw on that income later in life. This is in direct contrast to hand-outs given to the poor and needy, and does not qualify as "financial or other aid," any more than a return on an investment or a collection on insurance would.
I disagree. Regardless of the fact that people pay into Social Security, it's still welfare - just welfare that's very, very carefully budgeted, and you're actually informed how much you're taxed for.
People pay into (low income) welfare, too, by paying income taxes. Yes, there are people who could get money out of welfare without putting any money in - but there are people who get money out of social security who don't put any in (survivor benefits).
By your defintion of social security as "...aid provided to people in need" one could argue that some military roles the government plays (NOT Iraq, of course) is "welfare," in that it is "financial or other aid provided...to people in need." Certainly WW2 and Kosovo would qualify.
That's right. They are welfare efforts. The US press started calling them "relief" efforts (presumedly because of the stigma of the word welfare) but they've been called welfare efforts elsewhere.
welfare Pronunciation (wlfâr) n. 1.a. Health, happiness, and good fortune; well-being. b. Prosperity. 2. Welfare work. 3.a. Financial or other aid provided, especially by the government, to people in need. b. Corporate welfare.
I agree that Social Security should not have the same stigma as low-income welfare, nor the connotation usually associated with it, but it certainly is welfare. It's financial aid provided to people in need (by the government, in fact). The fact that you paid for some of it is irrelevant - it is welfare.
Welfare shouldn't have a negative connotation. Read the definiton - welfare is "giving help to people who need it." Not exactly something deserving of a negative connotation.
Now, the current public low-income welfare system - that might be worthy of a negative connotation. But that's no reason to pervert a perfectly fine word.
You *could* however, actually do that in a unified memory structure but not in a shared memory configuration.
No. The laptop I'm on right now can do that.
(II) I810(0): detected 892 kB stolen memory. (II) I810(0): I830CheckAvailableMemory: 206844 kB available (II) I810(0): Will attempt to tell the BIOS that there is 12288 kB VideoRAM
In the beginning, it allocates 892K to video memory. Later, when I switch to X, it allocates much more - 12M or so. It could allocate up to 202M (out of 256M available). That's just dynamic allocation. Nothing special there. The difference here is that it needs to do a BIOS call to retain compatibility with non-shared memory designs, but you could easily design a BIOS/OS which doesn't need to do that (but which wouldn't work for non-shared memory designs). Most shared memory solutions have dynamic resizing nowadays. The only thing you get with a true unified memory architecture is the ability to have non-contiguous graphics memory blocks, but in general you don't want to do that, as it increases latency and is generally pointless unless you're I/O bound anyway.
It wasn't so much that they designed it poorly, they did the best they could in the short amount of time they had because they had to use components which were already researched and tested.
I disagree. It was designed poorly. They picked a flawed encryption algorithm (so flawed that a high school kid with Google could've found the flaw) for the BIOS encryption. The processor is noticeably overpowered, with far too little cache, putting even more strain on a memory bus that's already overtaxed due to the shared memory design. Hell, they released a console in a country where the controller wouldn't fit in people's hands! This isn't merely quick design. This is simply poor design. OK, maybe it was just quick design - but it definitely wasn't the best they could do given the time they had.
This isn't even hindsight, either. Once the details of the system really began to emerge, many engineers and designers were absolutely baffled by the design choices made.
I hate to keep falling back on Nintendo, but they're a good example of good design, especially in the last-generation case. Near the release of the GameCube, they lowered the clockspeed of the GPU simply because there was no reason for it to be higher - the design actually worked better with a lower clockspeed due to it being matched to the CPU's clockspeed (by an integer fraction).
So you'd rather they pick some other, lower, random number instead? Not gonna happen.
Yes, I would. It's a mistake for them to keep inflating numbers. It hurts their believability - in fact, the believability of the whole industry. Hence the reason we're having this discussion. As time goes on, fewer and fewer people will actually listen to the numbers, because they make no sense.
Do I think it will happen in this case? No, because Microsoft and Sony are both retarded. Nintendo has been constantly trying ever since the SNES to inject sanity into idiotic numbers. It's a breath of fresh air to any engineer that actually reads the specs of the machine to hear Nintendo talk about their consoles.
Hence the entire point of this thread. Microsoft's numbers are inflated, just to keep up with Sony's, whereas Nintendo's are realistic, as they always have been.
Sony is already saying their PS3 is 2x more powerful than the 360 as it is.... thus, proving my point.
It's not like this is a suprise either as we saw this EXACT same crap from them when they compared the PS2 to the Dreamcast.
Exactly. Which is why it would be smarter for Microsoft to let Sony go off and keep spouting crappy numbers while frothing at the mouth, and simply focus on the strength of the games and the appearance of the games, rather than simply saying. "Yours isn't 2X more powerful, OURS is!"
Microsoft is not going to win the next generation contest by convincing the public that their console is more powerful, or even as powerful, as the PS3. It just doesn't matter. So spouting this crap is simply harmful.
Bzzzt! Wrong. It would be silly to not make *textures* contiguous or models or anything that's going to be read all at once, but in a game, you won't be able to line your textures all nice-and-neat in a row in the order they're needed.
Yes, you are. You know when they're needed. If you pull them off of the storage medium, they're needed - so you plop them in memory contiguously. You don't stick one texture at 0-5M, one at 417-419M, and one at 511-512M.
The only time where you wouldn't do that is if you're so strapped on system RAM that you are dragging in textures off of the storage medium as-needed, and if that's the case, you're going to be I/O bandwidth limited anyway.
The GPU is well-designed for a shared-memory design - it offloads a lot of the bandwidth intensive tasks to a high-bandwidth local storage solution. But it is still strictly worse than if the design was 384MB of system RAM and 128MB of GPU RAM with a 10MB eDRAM solution, and some sort of GPU system RAM mapping. That would've been more complicated (and more expensive), however.
And yet, it's proven quite capable despite this.
And Microsoft still lost money on it for virtually the entire time they sold it. Why? Because it was poorly engineered. It was more expensive than it needed to be to produce the graphics it did.
Amazingly enough, most mistakes can be overcome by throwing money at it.
You know, you seem awfully fixated on latencies.
Yah. Because latencies are the bane of modern design. The RDRAM/SDRAM difference is a red herring (not the least of which is because 25% higher latency is *nothing*), as RDRAM only existed on P4 designs, which needed bandwidth more than it needed low latency, and neither of the designs actually were capable of sating it via bandwidth. If you actually had RDRAM now, you'd see a bigger performance gap. A better example is looking at an Athlon 64 vs an Athlon XP (in 32-bit mode), as their main difference is the low latency of the Athlon 64.
Suppose I have a design with a constant latency of 10 ns, running at 100 MHz. That's a latency of 1 clock cycle. Now assume that you have an access pattern that's such that that you get hit with that latency every transfer - so every transfer is 1 latency cycle, then 1 transfer cycle. The effective bandwidth of the design is cut in half - instead of 100 Mtransfers/s, it's 50 Mtransfers/s. So its efficiency is 50%.
Now suppose you have a design with same latency, running at 1 GHz. That's a latency of 10 clock cycles. Assume the same access pattern. The effective bandwidth is now 90.9 Mtransfers/s (11 Hz/transfer, 1 GHz) and the efficiency of the design is now 9%. Note that the effective bandwidth is also not significantly higher than the 100 MHz case - not even twice as high, certainly not 10 times as high. But now assume a different access pattern, such that the latency hit only occurs every 10 transfers (20 Hz/10transfers) and the effective bandwidth is now 500 Mtransfers/s (with an efficiency of 50%). Note that even if I decreased the latency to 5 ns for the 1 GHz case, it's still only 16% efficient! It's not the absolute latency that matters - it's the clock latency.
This is a worst case example, of course. But my point is that you have to take bandwidths like "22 GB/s!" with a grain of salt and that different access patterns can chew up bandwidth like nobody's business. A moderate latency solution can be fine for a design with only 1 thread and 1 GPU clamoring for bandwidth, but it can be much less efficient with 6 threads and 1 GPU as well.
Me I'll stick to a generation gap performance increase of 10x.
The very fact that it's arguable what the gap is is the whole friggin' point. There's no metric to measure it, so when you quote some random number like 10X, or 15X, you do the whole community a disservice, because you're pushing everyone to inflate a number that has no basis in fact. In truth, it's much better to recognize that problem, and instead try to be as conservative as possible when giving those numbers. Had Microsoft said 4-5 times as powerful, no one would be saying they're inflating things.
It'd be silly for it not to be. It would just increase the latencies for graphics reading, and you wouldn't be able to use the full system bandwidth.
The GPU does not need to have the entire buffer in EDRAM to work on it. In fact, the ATI rep mentioned that 1080p was not a problem for the GPU, if MS decided to implement it. It can work on chunks of the screen at the time, buffering texture, z-buffer, fragment programs, whatever.
Of course it doesn't need the frame in eDRAM to work on it, but the fact that you basically can't store the frame in eDRAM should tell you that it's quite small. It's not a texture cache. At least, not for next-generation textures.
You also seem to completely gloss over that the xbox 1 has exactly the same design minus the edram. And yet, this "laughable" design creates games that look better than Sony's PS2 with gobloads of memory bandwidth.
I didn't claim Sony's PS2 as a pinnacle of design. Dear God, it's not. The GameCube is a much more balanced design. Besides, I called the Xbox 360 - if it was lacking the 10 MB eDRAM - laughable. Having a 500 MHz 48-way GPU without any dedicated memory would be laughable. The same isn't true for the Xbox's GPU.
The Xbox was not a good design. It was a poor design that competed solely by brute force. The unified memory architecture was panned greatly when it was announced.
Besides, it's important to remember that the Xbox 360 is a scale up from the Xbox. Latencies that wouldn't hurt the Xbox much will hurt the Xbox 360 much more.
Well, isn't that a little obtuse. So you're saying that a machine that can easily push 6x the number of pixels at 720p isn't really 6x more powerful because thats only 720p vs 480p and 4x MSAA vs no AA.
Maybe you have a differen't definition of segregated than I do. The CPU is in no way limited from writing/reading anywhere in system memory and neither is the GPU. How is that segregated?
Because if you want to have 128M of available memory for graphics-card related objects, you only have 512M-128M available for things that don't involve graphics. And if the GPU is working on something in that 128M, the CPU can't write there (at least, not without causing graphics corruption).
That number can change on the fly depending on memory needs, but that's no different than most modern onboard graphics cards anyway.
The GPU is also the memory controller, though, so it is a bit cheaper. BTW, you still need two busses, one from the CPU to the GPU and one from the GPU to the Memory, so it's not as huge a savings as you make it sound.
Except in the independent bus system you'd need three buses: one from the CPU to the memory controller, one from the memory controller to the memory, and one from the CPU to the GPU. So it's one less bus. If you merge the memory controller+memory into one "block" object, you'll get the two and one I mentioned.
I was thinking dual-ported ram, as would be the straight-forward way of doing this.
It has the advantage of not sacrificing bandwidth, but it heftily complicates system design. You now essentially have to have 2 memory controllers instead of one, as the latency is now unknown.
GART sucked. Notice how both ATI and Nvidia are going back to this now that we have PCIe? Of course, main system bandwidth sucks (6.4GB/s), so it isn't all that usefull, yet.
It will never be useful so long as the graphics card has roughly comparable bandwidth as the main memory bandwidth. Even if main memory is twice as fast, it probably will still suck. That was my point. The "ooh, shared memory is more useful for developers" is a crock - it's a pure limitation, not a benefit.
But this also brings up the 10MB of EDRAM in the GPU you keep ignoring.
I don't ignore it. Without it, the entire design would be laughable. The 10MB is critical, but it's not a texture cache: it's a framebuffer. At the Xbox 360's maximum resolution, the framebuffer will take up basically the entire 10MB.
er... the whole of its parts and the sum of its parts is, well, the same thing...
No, it's not. It's not even the same from a semantic definition. If you have a bundle of 4 oranges and 4 apples, and a bundle of 8 oranges and 8 apples, the second bundle isn't 4X more fruit because it's got 2X the oranges and 2X the apples. It's twice more fruit.
Wow, 3.5x the bandwidth? That's pretty big. More than I thought, really. That's much more than the difference in bandwidth between a GeForce 6800 and a GeForce 3.
That's the main memory bandwidth, not the graphics bandwidth. There are a few issues there - that's now split among 6 front ends, that only have 1M of shared L2 cache between all of them. The latency of GDDR3 is good for high-bandwidth, but not wonderful all-around - sharing it between 7 users will really start to lower its effective bandwidth quite heavily, because instead of streaming data, you're doing a whole lot of commanding.
Either that, or you're going to have to sacrifice on the amount of parallelization that you put in your code, which is probably more likely. I'd bet a lot that several of those front ends will just sit idling with a thread that, I dunno, checks for controller input or something.
But, let's go back to your numbers. If the Xbox 360 could do 400 M/s and the xBox 1 could do 100 M/s (just rounding to make things easier), but could texture them the same, would that make the Xbox 360 4x more powerful? Now add in 4x as many textures... how do you measure that?
4X the polygons, same numbers of textures - that's not 4X more powerful, not in my opinion. With 4X the number of textures and 4X the number of polygons, etc - that's more reasonable. Assuming that those peak fill rates are actually obtainable.
The CPU could twiddle a texture, then the GPU could directly use it for rendering. It won't have to pass over a bus first unlike a MB GPU that segregates a chuck of system ram to become it's own.
Uh... the shared RAM means that the GPU *does* segragate a chunk of system RAM. That's what a unified memory architecture means. If you're saying that that chunk might not be contiguous, as it is with standard shared systems, yah, sure, but that's not that big a benefit. It's still shared bandwidth.
Anyway, the CPU shouldn't be twiddling a texture. That's why the GPU is programmable. Even so, the memory on a GPU could be memory-mapped into the CPU's address space (which... it usually is), and it would seem exactly like the unified case, except that those transfers go out over a different bus.
Unified memory is strictly worse than separate GPU memory, because you can always get all of the "benefits" of unified memory by having something like an AGP GART on the GPU which allows the GPU to access main memory as well as its own local memory. The onboard memory then becomes a cache for the most often-used textures. Incidentally, you can guess how useful this feature actually is by looking at the number of AGP cards that actually *used* the GART to access main memory - as in, basically none. All graphics objects are heavily used. You want them close to the GPU.
but MS chose unified memory because it's more flexible for the developer and not the burden you seem to think it is.
MS chose unified memory because it's far cheaper, and simpler to design. One bus, rather than two.
In my opinion, you don't have to do 15x the resolution to be 15x more powerful. As I see it, 720p is 3x the pixel count of 480p. Add 4x the polly count (have you seen the alpha games like Test Drive?) and 4x the texture quality, where does that leave you?
Uh... about 4X more powerful. That's kinda my point. You're stacking the improvements, but a game is the whole of its parts, not the sum of them. If there are more polygons, but the textures still look crappy, it wouldn't necessarily be better.
I'm sure the "15X" numbers were generated by averaging the peak FLOPs, bandwidths, etc. for the Xbox vs. the Xbox 360, but most of the boost comes from the CPU, which is "80X!!" more powerful than the original (115.2 GFLOP vs 1.5 GFLOP).
The other numbers are much more pedestrian: ~twice as high clock speed on the GPU, though a lot wider, 8 times the RAM - most of the rest of the bandwidth, poly pushing, is about 4-5 times the original.
The Xbox 360 is not 1 TFLOP - the processor is just 115.2 GFLOP, and the rest of the marketroid 1 TFLOP number is from the GPU.
The important numbers, in my mind, are 22.4 GB/s - bandwidth to main memory - which is 3.5X the Xbox's bandwidth, but now it's shared between 6 threads and a GPU rather than 1 thread and a GPU. You can't get 15X performance - can't - with only 3.5X the bandwidth (and equivalent latency).
As for Rare's demos: that's almost exactly the same thing Nintendo did with the GameCube versus the N64 (and almost the same number of characters...). I wouldn't use "number of onscreen characters" as a linear performance metric.
This next-gen leap is about equivalent to each of the previous ones.
I'm pretty sure Xbox 360 could handle 15x as many textured, lit and shaded polies as the Xbox 1.
Last I heard, Xbox 360's peak polygon fill rate is 500M/s, not 6G/s, compared to the Xbox's 116.5M/s. So, uh, no, it can't. It can dump a lot more textures on those polygons - about four times more - but in terms of raw polygons, no, it can't.
So what's *your* definition of more powerful?
It's not my definition of "more" powerful. It's my definition of 15X more powerful. I never said it wasn't more powerful - it's just not "15X" more powerful. That's just silly. It can't run at 15X higher resolution, for instance. It can't have textures with 15X the detail of the original.
I *was* just comparing straight Ghz to Ghz after all.
My bad, I've always been bored by the "watch a blank screen while a box counts to a million 15X faster than an older box" game.
If what they're saying is "they can take old Xbox games (recompiled, obviously) and run then 15X faster on the new Xbox 360", they're lying. It just won't happen. The latency and bandwidth has not improved enough for that to be true, and the architecture is not easily exploitable for game software.
Of course, what they're saying is "the Xbox 360 has a peak MFLOPs 15X higher than the original Xbox", which is true, but utterly meaningless when it comes to games. The PS3 will have an utterly insane peak MFLOPs number (something like twice as high as the Xbox 360) but it certainly won't run games twice as fast.
Re:Massive processor, not much for graphics though
on
The Xbox 360 Unveiled
·
· Score: 1
Not much of a GPU? it has 10MB of EDRAM, and it has 256GB/sec of bandwidth in to it! That's ALOT of bandwidth. No, it's fucking HUGE amount of bandwidth!
It's a huge amount of bandwidth, but it's just 10MB. It's a texture cache, that's all. Especially with them hoping to push its HD capabilities, 10MB is actually a fairly small cache. It'll need to dig into the main memory quite often.
The bandwidth of something to its cache isn't exactly all that impressive. Of course it has to be that high - the damn graphics core is running at 500 MHz.
And in addition, it has the regural GDDR3 at it's disposal as well.
Shared with the processors.
6 front ends. Plus 1 graphics card.
Yup. That's seven users splitting that pipe to the GDDR3. Which is not low latency memory. Which means a huge amount of bandwidth will be lost if all 6 threads are doing anything significant.
This isn't as impressive as it looks. It's about as impressive as any new console looks at its release.
Well, several months before its release, at least.
That's comparing Ghz alone. Add in the more efficient cores, the dual thread execution and take out a bit for the inefficiencies in using multiple cores and 15x raw CPU power doesn't seem all that exaggerated.
Yes it does! You're claiming that 2 cores are 2X as powerful as 1 core, and that a 2 GHz machine is twice as fast as a 1 GHz machine.
Yah, sure, it's twice as fast at counting to a million. But it's not going to be twice as fast at running a generic program, not unless the memory latency halved and the throughput doubled. Anyone who's used a modern computer knows this.
Why is it that most of the world has given up on the MHz myth, but Sony and MS can just parrot it, and we all say "uh, okay".
Nintendo just isn't giving you la-la land numbers. They never have. The fill rate for the N64 was quoted as 150K triangles. The one for the PS1 is quoted as 1.5M triangles.
The PS1 is not 10X more powerful than the N64. Sony quotes raw triangles, Nintendo quoted textured triangles.
Who here is really surprised that xbox 360 isn't completely backwards compatible? If people will dish out $400 for a system, they WILL rebuy games so they can play them on their new system, ergo MS makes more money.
If you go into a store saying "I'm willing to buy an Xbox 360" then it doesn't matter.
But undecided people don't buy systems on hardware. They buy systems on games. I'm not talking about people who know intimate hardware details and hacking potential each box has. I'm talking about most consumers. I'm not talking about people who have an Xbox already and enough disposable income to buy another one. I'm talking about the average person.
Backwards compatibility provides a system with a huge pool of cheap games. It suddenly makes that $300 system with $10 games look less expensive than the $200 system with $50 games, and certainly less expensive than the other $300 system with $50 games.
Keep in mind that the Xbox completely flopped versus the PS2 in the last generation, regardless of what geeks think - at best, they stole 10% market share from Sony. If they want to try the same strategy as last time - that is, just cater to the geeks, have a blast - but they sure as heck didn't make money with that strategy before.
If Microsoft is targeting the people who have $500 + disposable income for games, I don't think they'll make much inroads in the market.
I know for a fact that if a console doesn't have backwards compatability, you have to really convince them otherwise to buy it, because they HATE wasting money on games....
But the parents won't ask "Is this backwards compatible?" just like they won't ask "Does this have the AltiVec vector instruction set for the PowerPC?" (Note that I don't know, nor care, whether it does:) )
They'll say "Can Jimmy play the Xbox games he already has on it?" and they can't say "Yes" if this is true. They'll probably also ask "can he play these games on it?" and point to the $10 preowned Xbox section, and they'll have to say "No". They'll then ask "which ones can he play" and the people will point to the $30-$60 Xbox 360 games. And I'm assuming that rereleased Xbox games will be in the $30 range. Maybe.
That's the point with backward compatibility. It's not about playing your old games. Microsoft didn't demolish the competition in the last generation - hell, it's not even guaranteed that they'll end up in 2nd place, as Microsoft's shifting to Xbox 360, Nintendo's still got Zelda: Twilight Princess, and only 2 million consoles separate the two worldwide.
So not everyone has an Xbox. Backwards compatibility gives people cheap games to buy at launch. Cheap, as in $10 cheap. And when you're asking people to plunk down $300-400 on a console, there better be $10 cheap games. And I don't honestly think that companies could recompile an old game for the Xbox 360, bug test it thoroughly, fix any bugs (because oh, there will be bugs), rinse, repeat, then redo the box art, manufacture the game, repackage, and redistribute, and still sell the game for $10-15.
That's what backwards compatibility really means to consumers. $10-15 games at launch. If they don't have that, it just doesn't matter.
Slashdot Mirror
2x cheapest cell plan is still about $60-70.
T-Mobile's Family Time plan is $49.99, and includes 2 lines and unlimited mobile-to-mobile.
Bing, $50 unlimited cell usage.
I mean hey its the damn right size for an earth grade atmospherre... its just a little on the warm side... so you have to be careful what you do.
Venus has no angular momentum. Mars does. Venus is also far too close to the Sun. I'm not even sure a water cycle is even possible on Venus anymore, but if it is, it won't be for very long. So if you try to cool down Venus so that water condenses, the Sun's radiation will just dissociate the water vapor into hydrogen and oxygen, and the hydrogen will escape into space.
This is Earth's fate in a few hundred million years. Venus isn't really practical to terraform, unless you move it out to Mars's orbit and smack Mars into it... which sounds a lot like how the Earth was formed.
How so? I would think if backward time travel is going to work, it's still going to be continuous.
Depends on the method. If you multiply connect space (like a wormhole) then no, it will be instantaneous. You'll just multiply connect space - there'll be two ways to get from 1985->2005, for instance - one that involves sitting still and letting time pass (so that path would have a distance of ds^2 = (cdt)^2 where dt is 20 years, and c is the speed of light) and one involves going through the wormhole (so that path would have a distance of ds^2 = (dx^2+dy^2+dz^2) where dx,dy,dz are the physical distance through the wormhole).
Then, technically, both the monkey and the wormhole pop out of thin air, and then a little while later, the monkey goes into the wormhole and it disappears. Mass, energy are all conserved because the wormhole is essentially an anti-monkey.
If you just somehow manage to exceed the speed of light, then it will be continuous. This sort of time travel is what goes on inside the event horizon of a black hole (because the time coordinate now is the same signature as the theta and phi coordinate, and the radial coordinate is the "move only in one direction" i.e. inward).
If you go back, are you physically prevented from firing the gun or will the gun misfire?
No. Nothing weird will happen. It just didn't happen. It won't happen because it didn't happen. The reason for your confusion is that you're assuming that you actually do have control over your future.
It's like trying to stop thunder after you've seen the lightning. Nothing you can do will stop it. It's already happened.
Or if you make a change, does the timeline establish a new universe with the old one running along merrily as a parallel universe.
In the model of the Universe they're proposing, there are no parallel universes.
so is it possible that all those other postulated dimensions (to 11) give the degrees of freedom to allow bifurcations in the timeline?
No - no, no, no, no, no. You do not need an additional dimension to allow the degree of freedom. You already have it. It's proper distance - that is, the (3+1) spatiotemporal dimensions.
Why? Because you had to travel a distance to go there - the distance from your future to the past of this (now 'parallel') universe. There's a causal relation between something you did in universe X to something that happened in universe Y, which means there's a proper distance separation.
Once you realize that, however, the whole concept of changeable timelines basically becomes very ad-hoc - you have two separate universes that have absolutely no reason to be identical but which are. You've also essentially made it so that the total energy content of the Universe is infinite.
We could kill two birds with one stone and suck the CO2 out of the Venus atmosphere, and drop it into the Martian atmosphere. Voila! Thin out Venus, and thicken Mars. Two planets for the price of one.
Or we could just follow the Solar System's first example, and move Venus out to Mars's orbit, and crash Mars into it.
OK, OK, it'd take a few million years, but hell, no work needed afterwards. Nice way to eke a few more billion years out of our Sun after the increasing solar luminosity starts boiling away the seas on Earth in a few hundred million years.
which as I'm sure you know, is the layer of our atmosphere that is the most important in blocking ultraviolet radiation.
If you're capable of increasing the atmospheric density by a factor of a thousand, you're capable of reseeding the upper layers of the atmosphere continually.
Mars also has the nice benefit that ultraviolet radiation is already down by a factor of 3. If you want to be super clever, you could follow Kim Stanley Robinson's suggestion and drop an orbiting mirror to increase the solar insolation. Clever use of reflecting material would mean more sunlight, but the same level of UV.
It will only really work with the tougher Iron asteroids, though, the weaker "rubble piles" won't work.
And big ones, too. Smaller ones will have far too much Coriolis force and too much of a vertical gravity gradient for people not to get nauseous, especially if you want tall buildings. Tens of kilometers is probably the minimum.
And, of course, the bigger you get, the bigger a job hollowing it out is.
Look up "closed timelike loop."
Basically all this article is saying is that all time travel must consist of closed timelike loops. That is, you "fulfill" the present, rather than altering it. This isn't news - it's the only kind of timelike loop that can exist in GR anyway. The difference here is that quantum mechanics also forces them to be the only ones that exist.
Point of note, however: as far as I know, we don't actually have the math to deal with the formation of a topological change in a surface (i.e., the "alteration" of a timeline). This is very much akin to a wave crashing - fluid dynamics works up until the exact point when the top of the wave touches the rest of the ocean. After that point, the math breaks down. So it's a little difficult to say "X isn't possible, because the math won't allow it" when theorists are in fact only using math that won't allow it. So it's moderately circular. That's GR. In QM, we don't actually have the math that deals with the collapse of the wavefunction (the 'measurement'), and so again, it's moderately circular. If you instead suppose that the wavefunction doesn't actually collapse, then of course you can change the past - you just end up following a different course in probability the second time around.
Examples of closed timelike loops actually are more common than you think in modern scifi/fantasy. Harry Potter and the Prisoner of Azkaban had an excellent example of a CTL, and the timetravel used in Anne McCaffrey's Pern series are entirely CTLs. This leads to statements like "I know I can do this, because I've done it already."
The problem with CTLs is that they muck with certain people's belief in free will.
If, for example, you knew a picture would be taken, you could reflect light from your body and appear in that picture, thereby altering the future.
Assuming you didn't exist in the picture before you went back in time.
And it just moves up from there for all other physical effects. Nothing touched, no air breathed, no light disturbed, nothing.
Unless it was already disturbed to begin with.
Again, there's no real logical problem here. Just the fact that you would have to disassociate yourself from the fact that all of your future actions are possible.
I don't necessarily agree with time travel. A closed timelike loop is essentially the equivalent of a monkey popping out of thin air, and then disappearing a few seconds later. It seems idiotic, and completely counter to all natural laws. But that wouldn't be the first time nature did that to us.
Free. Molex gives away samples.
The only redeeming factor that I see to this hardware is that it's an inexpensive LCD.
:)
Plus a cheap MP3 player. It's probably got just enough oomph to be an Ogg player as well. The LCD screen actually makes it so that you can actually have a decent GUI.
Other than that, let's see... with a serial port, you could easily make a universal remote control with LCD screen, a digital picture frame (as seen above), or a status monitor for X (where X is some other project that spits out serial data).
Of course, if you want to be really stupid, a 66 MHz ARM7 certainly has the capability to emulate an NES.
In my case, it looks like a really, really nice front end to the Motorola GPS unit I have (spits out only serial data).
It's like picking up cool pieces of plastic and cardboard when you're a kid because you swear that you'll find a brilliant use for something so neat. Then it sits in the closet until you're an adult and throw it out.
Bah! You just described my basement, except for the fact that I do find uses for said objects.
Otherwise you can custom build a similar device for only slightly more
C'mon, it's $12! No way you could get the parts and board for that little.
Actually, it is. The chip you've just described is an "all-in-one" hardware design. The catch is that it seems like a flexible all-in-one design. And with an LCD screen, what could be better?
We have different opinions of "all in one", apparently. If you've got some sort of embedded PIC with OTP ROM, and it's embedded in epoxy, yah, you're probably done (hence "all-in-one" - this is more of a "all-in-three-or-four"). But this has a lot of hackability in it, as you've got easy access to all of the address and chip selects, and some versions of the board even have a moderately replacable ROM.
Regarding the video: oh, definitely. The video potential of the device sucks. But I wouldn't expect anything else for $12.
The SDCard connector, however, is an add-on that is not all that cheap in of itself.
It's just a slot adapter. Grab an SD socket from Molex (either free, $4 inside another order, or $10 from Sparkfun) and wire it up directly, as the pinout is already available.
Plus there's serial input, as well. It's disabled in the default build, so you have to replace the ROM (or via JTAG), but you'd probably want to do that anyway.
There are always a few exceptions, and the JuiceBox looks like it's one of those. Probably for exactly the same reason that it failed in the marketplace as well - a $50 (questionably useful) toy is a bit too much.
The JuiceBox isn't one of the cheap, one-in-all chips: it's an S3C44B0X 66 MHz ARM7TDMI processor with 8M of RAM onboard (some have 2M) and an SD slot (via a custom connector, but you could make an adapter for roughly free by sampling the connector from Molex) with a uClinux distribution. Plus, JTAG pins are available on the far side. Hacking it into anything you want would be relatively simple.
This isn't exactly akin to a "toy". And it wasn't supposed to be $12 - but clearance sales are wonderful things, especially when they involve devices with LCD screens.
People pay into social security with the agreement that they will draw on that income later in life. This is in direct contrast to hand-outs given to the poor and needy, and does not qualify as "financial or other aid," any more than a return on an investment or a collection on insurance would.
I disagree. Regardless of the fact that people pay into Social Security, it's still welfare - just welfare that's very, very carefully budgeted, and you're actually informed how much you're taxed for.
People pay into (low income) welfare, too, by paying income taxes. Yes, there are people who could get money out of welfare without putting any money in - but there are people who get money out of social security who don't put any in (survivor benefits).
By your defintion of social security as "...aid provided to people in need" one could argue that some military roles the government plays (NOT Iraq, of course) is "welfare," in that it is "financial or other aid provided...to people in need." Certainly WW2 and Kosovo would qualify.
That's right. They are welfare efforts. The US press started calling them "relief" efforts (presumedly because of the stigma of the word welfare) but they've been called welfare efforts elsewhere.
1) Social Security isn't "welfare."
welfare Pronunciation (wlfâr)
n.
1.a. Health, happiness, and good fortune; well-being.
b. Prosperity.
2. Welfare work.
3.a. Financial or other aid provided, especially by the government, to people in need.
b. Corporate welfare.
I agree that Social Security should not have the same stigma as low-income welfare, nor the connotation usually associated with it, but it certainly is welfare. It's financial aid provided to people in need (by the government, in fact). The fact that you paid for some of it is irrelevant - it is welfare.
Welfare shouldn't have a negative connotation. Read the definiton - welfare is "giving help to people who need it." Not exactly something deserving of a negative connotation.
Now, the current public low-income welfare system - that might be worthy of a negative connotation. But that's no reason to pervert a perfectly fine word.
No. The laptop I'm on right now can do that.
In the beginning, it allocates 892K to video memory. Later, when I switch to X, it allocates much more - 12M or so. It could allocate up to 202M (out of 256M available). That's just dynamic allocation. Nothing special there. The difference here is that it needs to do a BIOS call to retain compatibility with non-shared memory designs, but you could easily design a BIOS/OS which doesn't need to do that (but which wouldn't work for non-shared memory designs). Most shared memory solutions have dynamic resizing nowadays. The only thing you get with a true unified memory architecture is the ability to have non-contiguous graphics memory blocks, but in general you don't want to do that, as it increases latency and is generally pointless unless you're I/O bound anyway.
It wasn't so much that they designed it poorly, they did the best they could in the short amount of time they had because they had to use components which were already researched and tested.
I disagree. It was designed poorly. They picked a flawed encryption algorithm (so flawed that a high school kid with Google could've found the flaw) for the BIOS encryption. The processor is noticeably overpowered, with far too little cache, putting even more strain on a memory bus that's already overtaxed due to the shared memory design. Hell, they released a console in a country where the controller wouldn't fit in people's hands! This isn't merely quick design. This is simply poor design. OK, maybe it was just quick design - but it definitely wasn't the best they could do given the time they had.
This isn't even hindsight, either. Once the details of the system really began to emerge, many engineers and designers were absolutely baffled by the design choices made.
I hate to keep falling back on Nintendo, but they're a good example of good design, especially in the last-generation case. Near the release of the GameCube, they lowered the clockspeed of the GPU simply because there was no reason for it to be higher - the design actually worked better with a lower clockspeed due to it being matched to the CPU's clockspeed (by an integer fraction).
So you'd rather they pick some other, lower, random number instead? Not gonna happen.
Yes, I would. It's a mistake for them to keep inflating numbers. It hurts their believability - in fact, the believability of the whole industry. Hence the reason we're having this discussion. As time goes on, fewer and fewer people will actually listen to the numbers, because they make no sense.
Do I think it will happen in this case? No, because Microsoft and Sony are both retarded. Nintendo has been constantly trying ever since the SNES to inject sanity into idiotic numbers. It's a breath of fresh air to any engineer that actually reads the specs of the machine to hear Nintendo talk about their consoles.
Hence the entire point of this thread. Microsoft's numbers are inflated, just to keep up with Sony's, whereas Nintendo's are realistic, as they always have been.
Sony is already saying their PS3 is 2x more powerful than the 360 as it is.
It's not like this is a suprise either as we saw this EXACT same crap from them when they compared the PS2 to the Dreamcast.
Exactly. Which is why it would be smarter for Microsoft to let Sony go off and keep spouting crappy numbers while frothing at the mouth, and simply focus on the strength of the games and the appearance of the games, rather than simply saying. "Yours isn't 2X more powerful, OURS is!"
Microsoft is not going to win the next generation contest by convincing the public that their console is more powerful, or even as powerful, as the PS3. It just doesn't matter. So spouting this crap is simply harmful.
Bzzzt! Wrong. It would be silly to not make *textures* contiguous or models or anything that's going to be read all at once, but in a game, you won't be able to line your textures all nice-and-neat in a row in the order they're needed.
Yes, you are. You know when they're needed. If you pull them off of the storage medium, they're needed - so you plop them in memory contiguously. You don't stick one texture at 0-5M, one at 417-419M, and one at 511-512M.
The only time where you wouldn't do that is if you're so strapped on system RAM that you are dragging in textures off of the storage medium as-needed, and if that's the case, you're going to be I/O bandwidth limited anyway.
The GPU is well-designed for a shared-memory design - it offloads a lot of the bandwidth intensive tasks to a high-bandwidth local storage solution. But it is still strictly worse than if the design was 384MB of system RAM and 128MB of GPU RAM with a 10MB eDRAM solution, and some sort of GPU system RAM mapping. That would've been more complicated (and more expensive), however.
And yet, it's proven quite capable despite this.
And Microsoft still lost money on it for virtually the entire time they sold it. Why? Because it was poorly engineered. It was more expensive than it needed to be to produce the graphics it did.
Amazingly enough, most mistakes can be overcome by throwing money at it.
You know, you seem awfully fixated on latencies.
Yah. Because latencies are the bane of modern design. The RDRAM/SDRAM difference is a red herring (not the least of which is because 25% higher latency is *nothing*), as RDRAM only existed on P4 designs, which needed bandwidth more than it needed low latency, and neither of the designs actually were capable of sating it via bandwidth. If you actually had RDRAM now, you'd see a bigger performance gap. A better example is looking at an Athlon 64 vs an Athlon XP (in 32-bit mode), as their main difference is the low latency of the Athlon 64.
Suppose I have a design with a constant latency of 10 ns, running at 100 MHz. That's a latency of 1 clock cycle. Now assume that you have an access pattern that's such that that you get hit with that latency every transfer - so every transfer is 1 latency cycle, then 1 transfer cycle. The effective bandwidth of the design is cut in half - instead of 100 Mtransfers/s, it's 50 Mtransfers/s. So its efficiency is 50%.
Now suppose you have a design with same latency, running at 1 GHz. That's a latency of 10 clock cycles. Assume the same access pattern. The effective bandwidth is now 90.9 Mtransfers/s (11 Hz/transfer, 1 GHz) and the efficiency of the design is now 9%. Note that the effective bandwidth is also not significantly higher than the 100 MHz case - not even twice as high, certainly not 10 times as high. But now assume a different access pattern, such that the latency hit only occurs every 10 transfers (20 Hz/10transfers) and the effective bandwidth is now 500 Mtransfers/s (with an efficiency of 50%). Note that even if I decreased the latency to 5 ns for the 1 GHz case, it's still only 16% efficient! It's not the absolute latency that matters - it's the clock latency.
This is a worst case example, of course. But my point is that you have to take bandwidths like "22 GB/s!" with a grain of salt and that different access patterns can chew up bandwidth like nobody's business. A moderate latency solution can be fine for a design with only 1 thread and 1 GPU clamoring for bandwidth, but it can be much less efficient with 6 threads and 1 GPU as well.
Me I'll stick to a generation gap performance increase of 10x.
The very fact that it's arguable what the gap is is the whole friggin' point. There's no metric to measure it, so when you quote some random number like 10X, or 15X, you do the whole community a disservice, because you're pushing everyone to inflate a number that has no basis in fact. In truth, it's much better to recognize that problem, and instead try to be as conservative as possible when giving those numbers. Had Microsoft said 4-5 times as powerful, no one would be saying they're inflating things.
Who says the memory has to be contiguous?
It'd be silly for it not to be. It would just increase the latencies for graphics reading, and you wouldn't be able to use the full system bandwidth.
The GPU does not need to have the entire buffer in EDRAM to work on it. In fact, the ATI rep mentioned that 1080p was not a problem for the GPU, if MS decided to implement it. It can work on chunks of the screen at the time, buffering texture, z-buffer, fragment programs, whatever.
Of course it doesn't need the frame in eDRAM to work on it, but the fact that you basically can't store the frame in eDRAM should tell you that it's quite small. It's not a texture cache. At least, not for next-generation textures.
You also seem to completely gloss over that the xbox 1 has exactly the same design minus the edram. And yet, this "laughable" design creates games that look better than Sony's PS2 with gobloads of memory bandwidth.
I didn't claim Sony's PS2 as a pinnacle of design. Dear God, it's not. The GameCube is a much more balanced design. Besides, I called the Xbox 360 - if it was lacking the 10 MB eDRAM - laughable. Having a 500 MHz 48-way GPU without any dedicated memory would be laughable. The same isn't true for the Xbox's GPU.
The Xbox was not a good design. It was a poor design that competed solely by brute force. The unified memory architecture was panned greatly when it was announced.
Besides, it's important to remember that the Xbox 360 is a scale up from the Xbox. Latencies that wouldn't hurt the Xbox much will hurt the Xbox 360 much more.
Well, isn't that a little obtuse. So you're saying that a machine that can easily push 6x the number of pixels at 720p isn't really 6x more powerful because thats only 720p vs 480p and 4x MSAA vs no AA.
Yah, basically. Guess I'm just demanding.
Maybe you have a differen't definition of segregated than I do. The CPU is in no way limited from writing/reading anywhere in system memory and neither is the GPU. How is that segregated?
Because if you want to have 128M of available memory for graphics-card related objects, you only have 512M-128M available for things that don't involve graphics. And if the GPU is working on something in that 128M, the CPU can't write there (at least, not without causing graphics corruption).
That number can change on the fly depending on memory needs, but that's no different than most modern onboard graphics cards anyway.
The GPU is also the memory controller, though, so it is a bit cheaper. BTW, you still need two busses, one from the CPU to the GPU and one from the GPU to the Memory, so it's not as huge a savings as you make it sound.
Except in the independent bus system you'd need three buses: one from the CPU to the memory controller, one from the memory controller to the memory, and one from the CPU to the GPU. So it's one less bus. If you merge the memory controller+memory into one "block" object, you'll get the two and one I mentioned.
I was thinking dual-ported ram, as would be the straight-forward way of doing this.
It has the advantage of not sacrificing bandwidth, but it heftily complicates system design. You now essentially have to have 2 memory controllers instead of one, as the latency is now unknown.
GART sucked. Notice how both ATI and Nvidia are going back to this now that we have PCIe? Of course, main system bandwidth sucks (6.4GB/s), so it isn't all that usefull, yet.
It will never be useful so long as the graphics card has roughly comparable bandwidth as the main memory bandwidth. Even if main memory is twice as fast, it probably will still suck. That was my point. The "ooh, shared memory is more useful for developers" is a crock - it's a pure limitation, not a benefit.
But this also brings up the 10MB of EDRAM in the GPU you keep ignoring.
I don't ignore it. Without it, the entire design would be laughable. The 10MB is critical, but it's not a texture cache: it's a framebuffer. At the Xbox 360's maximum resolution, the framebuffer will take up basically the entire 10MB.
er... the whole of its parts and the sum of its parts is, well, the same thing...
No, it's not. It's not even the same from a semantic definition. If you have a bundle of 4 oranges and 4 apples, and a bundle of 8 oranges and 8 apples, the second bundle isn't 4X more fruit because it's got 2X the oranges and 2X the apples. It's twice more fruit.
Wow, 3.5x the bandwidth? That's pretty big. More than I thought, really. That's much more than the difference in bandwidth between a GeForce 6800 and a GeForce 3.
That's the main memory bandwidth, not the graphics bandwidth. There are a few issues there - that's now split among 6 front ends, that only have 1M of shared L2 cache between all of them. The latency of GDDR3 is good for high-bandwidth, but not wonderful all-around - sharing it between 7 users will really start to lower its effective bandwidth quite heavily, because instead of streaming data, you're doing a whole lot of commanding.
Either that, or you're going to have to sacrifice on the amount of parallelization that you put in your code, which is probably more likely. I'd bet a lot that several of those front ends will just sit idling with a thread that, I dunno, checks for controller input or something.
But, let's go back to your numbers. If the Xbox 360 could do 400 M/s and the xBox 1 could do 100 M/s (just rounding to make things easier), but could texture them the same, would that make the Xbox 360 4x more powerful? Now add in 4x as many textures... how do you measure that?
4X the polygons, same numbers of textures - that's not 4X more powerful, not in my opinion. With 4X the number of textures and 4X the number of polygons, etc - that's more reasonable. Assuming that those peak fill rates are actually obtainable.
The CPU could twiddle a texture, then the GPU could directly use it for rendering. It won't have to pass over a bus first unlike a MB GPU that segregates a chuck of system ram to become it's own.
Uh... the shared RAM means that the GPU *does* segragate a chunk of system RAM. That's what a unified memory architecture means. If you're saying that that chunk might not be contiguous, as it is with standard shared systems, yah, sure, but that's not that big a benefit. It's still shared bandwidth.
Anyway, the CPU shouldn't be twiddling a texture. That's why the GPU is programmable. Even so, the memory on a GPU could be memory-mapped into the CPU's address space (which... it usually is), and it would seem exactly like the unified case, except that those transfers go out over a different bus.
Unified memory is strictly worse than separate GPU memory, because you can always get all of the "benefits" of unified memory by having something like an AGP GART on the GPU which allows the GPU to access main memory as well as its own local memory. The onboard memory then becomes a cache for the most often-used textures. Incidentally, you can guess how useful this feature actually is by looking at the number of AGP cards that actually *used* the GART to access main memory - as in, basically none. All graphics objects are heavily used. You want them close to the GPU.
but MS chose unified memory because it's more flexible for the developer and not the burden you seem to think it is.
MS chose unified memory because it's far cheaper, and simpler to design. One bus, rather than two.
In my opinion, you don't have to do 15x the resolution to be 15x more powerful. As I see it, 720p is 3x the pixel count of 480p. Add 4x the polly count (have you seen the alpha games like Test Drive?) and 4x the texture quality, where does that leave you?
Uh... about 4X more powerful. That's kinda my point. You're stacking the improvements, but a game is the whole of its parts, not the sum of them. If there are more polygons, but the textures still look crappy, it wouldn't necessarily be better.
I'm sure the "15X" numbers were generated by averaging the peak FLOPs, bandwidths, etc. for the Xbox vs. the Xbox 360, but most of the boost comes from the CPU, which is "80X!!" more powerful than the original (115.2 GFLOP vs 1.5 GFLOP).
The other numbers are much more pedestrian: ~twice as high clock speed on the GPU, though a lot wider, 8 times the RAM - most of the rest of the bandwidth, poly pushing, is about 4-5 times the original.
The Xbox 360 is not 1 TFLOP - the processor is just 115.2 GFLOP, and the rest of the marketroid 1 TFLOP number is from the GPU.
The important numbers, in my mind, are 22.4 GB/s - bandwidth to main memory - which is 3.5X the Xbox's bandwidth, but now it's shared between 6 threads and a GPU rather than 1 thread and a GPU. You can't get 15X performance - can't - with only 3.5X the bandwidth (and equivalent latency).
As for Rare's demos: that's almost exactly the same thing Nintendo did with the GameCube versus the N64 (and almost the same number of characters...). I wouldn't use "number of onscreen characters" as a linear performance metric.
This next-gen leap is about equivalent to each of the previous ones.
I'm pretty sure Xbox 360 could handle 15x as many textured, lit and shaded polies as the Xbox 1.
Last I heard, Xbox 360's peak polygon fill rate is 500M/s, not 6G/s, compared to the Xbox's 116.5M/s. So, uh, no, it can't. It can dump a lot more textures on those polygons - about four times more - but in terms of raw polygons, no, it can't.
So what's *your* definition of more powerful?
It's not my definition of "more" powerful. It's my definition of 15X more powerful. I never said it wasn't more powerful - it's just not "15X" more powerful. That's just silly. It can't run at 15X higher resolution, for instance. It can't have textures with 15X the detail of the original.
I *was* just comparing straight Ghz to Ghz after all.
My bad, I've always been bored by the "watch a blank screen while a box counts to a million 15X faster than an older box" game.
If what they're saying is "they can take old Xbox games (recompiled, obviously) and run then 15X faster on the new Xbox 360", they're lying. It just won't happen. The latency and bandwidth has not improved enough for that to be true, and the architecture is not easily exploitable for game software.
Of course, what they're saying is "the Xbox 360 has a peak MFLOPs 15X higher than the original Xbox", which is true, but utterly meaningless when it comes to games. The PS3 will have an utterly insane peak MFLOPs number (something like twice as high as the Xbox 360) but it certainly won't run games twice as fast.
Not much of a GPU? it has 10MB of EDRAM, and it has 256GB/sec of bandwidth in to it! That's ALOT of bandwidth. No, it's fucking HUGE amount of bandwidth!
It's a huge amount of bandwidth, but it's just 10MB. It's a texture cache, that's all. Especially with them hoping to push its HD capabilities, 10MB is actually a fairly small cache. It'll need to dig into the main memory quite often.
The bandwidth of something to its cache isn't exactly all that impressive. Of course it has to be that high - the damn graphics core is running at 500 MHz.
And in addition, it has the regural GDDR3 at it's disposal as well.
Shared with the processors.
6 front ends. Plus 1 graphics card.
Yup. That's seven users splitting that pipe to the GDDR3. Which is not low latency memory. Which means a huge amount of bandwidth will be lost if all 6 threads are doing anything significant.
This isn't as impressive as it looks. It's about as impressive as any new console looks at its release.
Well, several months before its release, at least.
That's comparing Ghz alone. Add in the more efficient cores, the dual thread execution and take out a bit for the inefficiencies in using multiple cores and 15x raw CPU power doesn't seem all that exaggerated.
Yes it does! You're claiming that 2 cores are 2X as powerful as 1 core, and that a 2 GHz machine is twice as fast as a 1 GHz machine.
Yah, sure, it's twice as fast at counting to a million. But it's not going to be twice as fast at running a generic program, not unless the memory latency halved and the throughput doubled. Anyone who's used a modern computer knows this.
Why is it that most of the world has given up on the MHz myth, but Sony and MS can just parrot it, and we all say "uh, okay".
Nintendo just isn't giving you la-la land numbers. They never have. The fill rate for the N64 was quoted as 150K triangles. The one for the PS1 is quoted as 1.5M triangles.
The PS1 is not 10X more powerful than the N64. Sony quotes raw triangles, Nintendo quoted textured triangles.
Don't let them keep fooling you.
Who here is really surprised that xbox 360 isn't completely backwards compatible? If people will dish out $400 for a system, they WILL rebuy games so they can play them on their new system, ergo MS makes more money.
If you go into a store saying "I'm willing to buy an Xbox 360" then it doesn't matter.
But undecided people don't buy systems on hardware. They buy systems on games. I'm not talking about people who know intimate hardware details and hacking potential each box has. I'm talking about most consumers. I'm not talking about people who have an Xbox already and enough disposable income to buy another one. I'm talking about the average person.
Backwards compatibility provides a system with a huge pool of cheap games. It suddenly makes that $300 system with $10 games look less expensive than the $200 system with $50 games, and certainly less expensive than the other $300 system with $50 games.
Keep in mind that the Xbox completely flopped versus the PS2 in the last generation, regardless of what geeks think - at best, they stole 10% market share from Sony. If they want to try the same strategy as last time - that is, just cater to the geeks, have a blast - but they sure as heck didn't make money with that strategy before.
If Microsoft is targeting the people who have $500 + disposable income for games, I don't think they'll make much inroads in the market.
I know for a fact that if a console doesn't have backwards compatability, you have to really convince them otherwise to buy it, because they HATE wasting money on games....
:) )
But the parents won't ask "Is this backwards compatible?" just like they won't ask "Does this have the AltiVec vector instruction set for the PowerPC?" (Note that I don't know, nor care, whether it does
They'll say "Can Jimmy play the Xbox games he already has on it?" and they can't say "Yes" if this is true. They'll probably also ask "can he play these games on it?" and point to the $10 preowned Xbox section, and they'll have to say "No". They'll then ask "which ones can he play" and the people will point to the $30-$60 Xbox 360 games. And I'm assuming that rereleased Xbox games will be in the $30 range. Maybe.
That's the point with backward compatibility. It's not about playing your old games. Microsoft didn't demolish the competition in the last generation - hell, it's not even guaranteed that they'll end up in 2nd place, as Microsoft's shifting to Xbox 360, Nintendo's still got Zelda: Twilight Princess, and only 2 million consoles separate the two worldwide.
So not everyone has an Xbox. Backwards compatibility gives people cheap games to buy at launch. Cheap, as in $10 cheap. And when you're asking people to plunk down $300-400 on a console, there better be $10 cheap games. And I don't honestly think that companies could recompile an old game for the Xbox 360, bug test it thoroughly, fix any bugs (because oh, there will be bugs), rinse, repeat, then redo the box art, manufacture the game, repackage, and redistribute, and still sell the game for $10-15.
That's what backwards compatibility really means to consumers. $10-15 games at launch. If they don't have that, it just doesn't matter.