Your comment is mostly right, but here are some minor corrections:
First, the Emotion Engine runs at 300Mhz in the PS2. It was initially announced at 250Mhz back in March of 1999, but within 3 weeks of the initial specs (confirming 55 million polygons per second), Sony changed their design to 300Mhz (66 million polygons per second).
You really shouldn't try to do comparisons using cycle counts, since in most cases the operations cost the exact same (integer ADD and SUB cost 1 with a 1-cycle latency on both x86 and MIPS architectures), and in others the x86 has extra concatenated operations (BSWAP, BTEST, etc.) due to its CISC nature that have no direct equivalent on MIPS systems.
As for non-vector performance of the Emotion Engine... The cost of additions, subtractions, and bit operations on a MIPS III architecture processor (like the Emotion Engine) is virtually identical to the cost on an x86. The Emotion Engine has a decent advantage on branch misprediction, since the EE only uses a 6-stage pipeline (as opposed to a 20-stage pipeline on the Wilamette). This basically means that if the processor incorrectly guesses where a program will jump, only 6 operations need to be performed for the Emotion Engine, while 20 will need to be performed by the PC. However, since branch mispredictions will happen more frequently on the Emotion Engine than the PC (PCs have a lot of extra hardware and cache to compensate for their long pipelines), this is basically a net-zero difference. The big difference is that the PC runs at 3x the clock speed of the Emotion Engine, which will basically allow it to perform 3x as many elementary operations in the same amount of time. A better comparison would be to say that for just integer operations, the Emotion Engine at 300Mhz is roughly equivalent to a Pentium II at 300Mhz.
I've debated this in quite a bit of detail on comp.games.development.industry, and basically, after tons of calculations, we came down to the following:
Not too bad, seeing as the EE has 10.7M transistors, dissipates 13W, and has on-board MPEG2 decoding. For use in a workstation environment, where processors can get much warmer, cranking the EE up to 600Mhz wouldn't be out of the question. 2x EE at 600Mhz would best almost every workstation ever designed. Granted, all the tools (Alias, etc.) would need to be rewritten for it.
I pay $80 per year for a 100BT in-room connection on Stanford's network. The fact is, if you're not explicitly paying the school for the hook-up, then the bill is discreetly placed in your tuition somewhere. $80 is far less than what a 100 megabit connection would cost me for a year, so I'm not going to argue.
1) OPEC needs to be broken up by the terms of the Sherman Anti-Trust Act. Their actions undeniably restrict trade, and they do constitute a monopoly on the crude oil industry, and high gas prices have hurt me. Therefore, it's legal to sue them.
2) India's atmospheric nuclear tests spread several particles-per-billion into the air that I breathe. This will probably cause cancer, so just like power companies have been sued, India should be sued.
3) The Soviet Union's belligerence caused taxes to rise into the stratosphere as a result of programs used to balance their military muscle (strategic defense initiative being a primary culprit). If the USSR weren't so damn belligerent back in the 60s, 70s and 80s, I wouldn't be losing 33% of every paycheck. Based on lost wages, interest, and inflation accrued during and after the inauguration of these programs, the American people have lost billions of dollars. This is obviously grounds for a class-action lawsuit against the Kremlin.
Honestly, the US has enough bad laws without passing totally asinine ones. Sue the lawmakers for being paid with our money to debate issues like the legality of suing foreign governments.
If anything, it is a creative way of repaying foreign debts ("What do you mean we owe you $100 billion? We sued your asses and won! We just deducted our $100 billion damages from what we owed you, so we're even!")
Under NT4, you do: HKEY_CURRENT_USER\Software\Microsoft\Command Processor\CompletionChar == 0x0000009
It works a little differently than under Unix, though. Pressing W selects the first entry that starts with W. Pressing again cycles through the other entries.
This will probably sound like an advertisement for Win2K, but I for one am tired of the unbased Win2K bashing on Slashdot. Note that everything I write is based on my own personal experiences with Win2K, as I run it on my personal machine.
Consider me one of the one million nobodies that legally upgraded to Win2K this month. I was previously running NT4SP5, but because of my hard disk partitioning, I had to reformat and do a fresh install.
All the talk of 65k bugs in 2k really means nothing without actual usage backup, so I'll be one of the first people to actually post how Win2K runs, at least on my machine.
First, Win2K is big. I wasn't expecting 900MB for the OS, but to be fair, 60MB is used by DRIVER.CAB (all the included drivers), 192MB by my swap file, and another 70-80MB by the multilingual options (30MB by nihongo alone). Granted, even subtracting out those options, Win2K is far and away the largest OS I have ever seen or used. But, having installed it with more than enough space left over for my programs and MP3s, my real question was how it would compare to NT4SP5, which had run wonderfully for over a year.
First, the bad. Win2K has several minor bugs that I've noticed. Most of these pertain to little graphical nuances in programs - occasionally menu bars don't always repaint themselves correctly, and on some occassions cursors don't alwyas update on applications like ICQ. All in all, nothing I can't live with (or that a service pack won't fix). NT4SP5 didn't have these problems - of coures, NT4SP5 didn't support USB, Plug and Play, AGP texturing, DirectX, or numerous other things that are in Win2K. Another (minor) downside is that it took a little more effort than it should have to enable tab auto-complete in the Command Prompt, but this was also the case with NT4.
Now, the good. Installation was so amazingly painless that I thought I had just installed Unreal Tournament. Win2K boots from the CD, and asks how you want to format your disks (just like NT4). However, dozens of SCSI and mass-storage drivers are included, even my Diamond Fireport 40 (thus saving me the effort of reinstalling a floppy drive to install Fireport 40 drivers). Then, 2K asks you 2 other questions throughout installation - one is what language options you want installed, and the other is what networking options (system name, etc.) you want installed. This can be configured from the command line, so installation can literally just be a matter of inserting the CD, rebooting, and letting it run for an hour. When I returned, *all* of my hardware, sans my DVD Decoder card (Creative Dxr2) had been installed - my RivaTNT, Fireport 40, Sound Blaster Live!, 3C905, everything was in and running just fine. After downloading and installing NVIDIA's new drivers (to enable hardware OpenGL support), all of my OpenGL applications worked flawlessly.
Performance - it is quite apparent that NTFS' disk caching is better under 2000 than it was under NT4. Other things seem just slightly faster than under NT4SP5, and nothing seems slower. Basically, if you liked NT's performance, you won't be disappointed with 2000 (I don't have any data for performance with FAT).
Stability - so far 2000 seems as stable as NT. Granted, this is only something that time can prove, but so far everything is extremely responsive, and I have noticed no hiccups at all. Another major improvement is that you do *not* need to restart anywhere near as often for applications. This is definitely nice; however, unless you install applications 3-4 times daily, probably won't matter much in the long run. Also included is a boot console interface (for major emergencies). I never understood why this wasn't included in NT4; however, while it is definitely on the sparse side, the CLI may prove invaluable, if driver manufacturers start releasing shitty drivers that cause BSODs when starting 2K.
Usage - this is all a matter of taste, but I like some of the Active Desktop features (Quickstart bar, etc). There is lots of alpha blending used in the UI - this will never be useful, but it is pretty cool, at least until the novelty wears off. Having DirectX is really nice; however, I will always be a proponent of OpenGL, so while I will enjoy being able to play the games that are DX-only, I will privately be wishing that they used OpenGL instead. Also, 2K has some very nice support for multimedia files (read: MP3s) built into explorer.
All in all 2K is a worthwhile operating system, and a very nice update to NT4. Not having to deal with the archaic 'Devices' control panel is an extreme relief (stupid NT3.51 hold-over), and a lot of the new features do make doing mundane tasks (installing hardware, managing user accounts, etc.) much more palatable than under NT4. Also included is a nice "auto-login" feature, for anyone interested in using 2K as a work-alike replacement for 98.
I've heard horror (and success) stories about upgrading from 9X->2000, but so far my experience has been very positive. If anyone on Slashdot uses Windows but hates 98's "reliability" or wishes NT played more games, 2000 would be a decent upgrade.
CPU1: NVIDIA NV26 GPU, 600m Matrix x vector operations per second, capable of transforming and lighting 150 million triangles (100 million vertices) per second. Collision detection and vertex blending is integrated into the core at no extra cost.
SPU: Emu E12000 Environmental Extender, real-time 1536kbps DTS-ES 6.1 compression, THX Ultra Certified, 130 dB signal/noise ratio, 96 hardware accelerated 3D voices. Psychoacoustic technologies allow simulating sounds along the vertical axis.
Rasterization Processor: NVIDIA NV26R, 4.8GPixel/sec fill rate, 8 programmable hardware texel units and polygon normal interpolation, yielding 120 million 8-level multi-textured phong shaded 25-pixel polygons / sec. Supports all HDTV resolutions including 1920x1080P, all at 8-bits per channel plus an 8-bit stencil buffer and 24-bit Z Buffer. Also, supports 15 million triangles/second in distributed raytracing mode, with radiosity.
Extra: Plays DVD movies, DVD Audio, VCD, CD audio, Playstation, Saturn, and DirectX games. And it can be programmed to download porn while you sleep.
Cost: $300
This does 2 trillion operations per second, so you can get double photo-realistic results. -----------------------
Re:More money = better grade at the end?
on
Laptop Exams?
·
· Score: 1
There are some schools these days (I believe Villanova may be one of them) that require their undergrads to purchase laptops from the 'school source'. I'm not sure if anything but these laptops is allowed onto the school network, either.
Here is a radical idea to take care of vast music piracy, and actually improve the quality of music that we have:
The government (National Association for the Arts is the bureau, I believe) subsidizes professional artists (to some degree), who then sell their original work to collectors for considerable profit. What if music were to stop being treated as an industry, and more as an art form (using a model similar to that used for fine art)? Then the free trading of music via Napster and the like would be completely legal and legitimate, and young'uns wouldn't learn how easy it is to break toothless laws.
A side effect of this would be that the only people remaining in music would be those that actually love making music. The romanticized image of the 'rich rock star that gets lots of girls' would become archaic, and modern industry evils like boy bands would be no more. The good artists, that actually contribute to the medium (there are a few bands (and quite a few rappers) remaining that would fit this mold, but not the Limp Bizkits and Backstreet Boys of the world) would be able to maintain good lifestyles, and culture wouldn't be so overrun with the banality that is pop music.
I doubt that this is realistic, but it would definitely solve quite a few problems - the enjoyment of music would be free, artists wouldn't sacrifice 90% of what is rightfully their profits to industry coffers (granted, they wouldn't be making much more money, if any), and the quality of music would improve considerably.
Actually, all the videos were fake! If you look at NVIDIA's press release (and Microsoft's), it becomes apparent that neither the graphics chip nor the sound chip will exist for another year. Every single video they showed was a mock-up, and none of the high-res screen shots for the mech demo (on Microsoft's home page) had any motion blur to speak of - although antialiasing was in full effect.
Apparently Microsoft has also claimed a 6.4GB/sec bus; however, that's reportedly using PC400 SDRAM, which neither exists nor runs at 6.4GB/sec (PC100 =.8GB/sec, PC200 = 1.6GB/sec, PC400 != 6.4GB/sec).
All in all, the X-Box specs amount to more of what Microsoft would like to achieve than what it actually will achieve. The prototype hardware will not exist for another year, at least, and I suspect that NVIDIA will have some major difficulties producing a processor that is 10x faster than the NV15, in roughly the same amount of time. Given Murphy's law, Microsoft will end up with 2 choices: produce a less-spectacular platform for Christmas 2001, or risk delay until 2002. If they delay until 2002 the X-Box is guaranteed DOA. The Dolphin and PS2 will be on their 2nd and 3rd generation games, respectively, and will cost considerably less than the X-Box. If people have just bought a Dreamcast, a PS2, and a Dolphin, it's unlikely that they'd be interested in also purchasing an X-Box, especially since all the games announced so far will be ports of Playstation/Dreamcast/PS2 games. If Microsoft chooses to release a less powerful console, their specs will probably end up being overestimates by at least a factor of 3 (NVIDIA's track record would pin the NV25 at about 7x faster than the GeForce, or ~110 million triangles/sec), which will not be enough muscle to really technically outshine the PS2 or Dolphin.
AGP is a bus from the graphics card to main system memory, to allow larger texturing areas. It communicates with the chipset, and actual performance is dependent upon the speed of memory in the system, and the processor's demand for memory. PC133 SDRAM has a maximum transfer rate of 1064MB/sec. AGP4X has a maximum transfer rate of 1064MB/sec. That means that for AGP4X to run at full speed, the processor could not be accessing memory. In order to combat this, things the frame buffer and some texture memory are stored on the card itself. This memory has a higher clock speed (on the GeForce it is 4.8GB/sec) thanks to a wider bus (256 bits on the card vs. 64 in main memory).
The XBox specs list a Unified Memory Architecture - that means that the frame buffer, vertex information, system RAM, textures, and sound all reside in the same pool of memory, and all communicate on the same bus (e.g. like AGP, only for all devices, not just the graphics card).
To just read 300 million triangles (~150 million vertices, for sufficiently large models) to the rasterization unit requires 3 floats per vertex, 4 bytes per float, or 1.8 GBytes/sec of memory bandwidth - which will quickly saturate the 3.2GB/sec maximum of dual channel RDRAM (since it is UMA, the memory will need to be external to the processors, and the bus will be limited to 64-bits, yielding a peak bandwidth of 3.2GB/sec). Then you need to add in textures (an additional 2 floats per vertex, or 66% more memory use), texture coordinates, rasterization, sounds, and everything else, and you will have some massive processor stalls. In addition to the fact that I have serious doubts about the 300m number, even if it is a theoretical maximum, as it is implemented in the XBox, you could expect a realistic maximum of ~45m triangles/sec. However, since the NV15 only has 2x the geometry acceleration of the GeForce (or about 30m triangles/sec), I seriously doubt that nVidia is sitting on an accelerator that is 10x as powerful as one that they have yet to release.
In contrast, the Playstation 2 has the chipset integrated into the Emotion Engine, with a 3.2GB/sec pipe from the EE to main memory. There is a 1.2GB/sec pipe from the EE to the Graphics Synthesizer (to stream textures and vertex data), which means that even when the GS is transfering as much as it can, the EE still has over 2GB of bandwidth remaining. Sounds are stored in 2MB of sound RAM (so playing sounds has no effect on system bandwidth), and the frame buffer (and texture cache) are stored on the embedded DRAM in the GS. The eDRAM runs at 48GB/sec, as compared to the 4.8GB/sec on the GeForce, allowing much more innovative effects, such as motion blur and heat warping, to be implemented easily.
And no, a bus for pure graphics could not accomplish what I stated. Almost everything that I mentioned requires massive software FPU capability, in addition to transformations. The Athlon 600 has a peak 2.4 GFlop performance (the P3-600B is even less). If you wanted to evaluate and tesselate bezier surfaces, you wouldn't be able to do that on the graphics card - it would need to be done by the CPU. However, due to congestion in system RAM (due to the UMA nature of the XBox) and minimal instruction decode pipelines (compared to the PS2), x86 processors wouldn't be able to generate anywhere near as many of these as the PS2 (remember - the PS2 has 2 floating point coprocessors, not geometry coprocessors). Physics calculations are the same case.
Realistically, even at best case, the XBox won't be any more powerful than the PS2. Even if the specs for the NVIDIA chip are accurate (I seriously doubt this), the entire design of the XBox will be a huge limitation, and totally limit performance. The XBox will only be marginally faster than a PC with an i840 chipset and an NV15 graphics card.
Basically, the design philosophies for the two systems boils down to: PS2: Work smarter, not harder XBox: Work harder
And the end result is that the XBox will have 10x the transistors of the PS2, and will have equal performance in some areas, and worse performance in others. The fact that it's not even due for another 18 months will be a major deterrent to its success.
Traditionally Tech demos show off the graphics/sound abilities of their hardware - many of the things you mention are things you'd do in software.
What they were showing off was the huge number of polygons that they can throw around, and they are throwing *a lot* of them around in that clip. They also appear to have bump mapping as well as specular reflection on their surfaces. As for realtime curves, it's quite possible that they were doing that on the characters, but I doubt it. Doesn't seem much point.
The point of doing realtime curves is to show off just how powerful the XBox is! Anybody can throw a bunch of triangles on the screen; however, no current game (for PC, Dreamcast, etc.) uses real-time bezier surfacing, because evaluating the curve and then tesselating it is a highly CPU-intensive task. The PS2 is capable of that, and still spitting out 15-20 million triangles (all tesselated from bezier surfaces). Hypertextures, realtime reflection/refraction, physical interaction to hair/cloth level - all are things that have yet to be done in games, and if the XBox is remotely as powerful as they hype it to be, should be able to handle. Honestly, while the R4 girl demo for the PS2 didn't look as nice as the mech demo for XBox, it was considerably more impressive, since it showed facial animation, weighted clothing, weighted hair, and extremely high polygon counts. Tech demos show off how powerful new technology is - radically new effects for radically new technologies - the XBox showed off more of the same.
And then what it showed looked like 50,000 triangles per frame, tops. The martial artist didn't look much better than the characters in Shen Mue. The model that I linked to above has 2 million triangles - the XBox should be able to spit out 2 of those at 60 frames/second easily. That mech did not have terribly complicated geometry, and that room was a textured cylinder, with some cross-beams.
It was *much* more than a Unreal engine - the models were at least an order of magnitude more detailed, and there also appeared to be motion blur on the characters, although that could also be as a result of being put on video... Anyway, the x-box site claims a thoughput of 300 million triangles/second, so at 60 fps, it would have a peak performance of 5 million triangles/frame. In reality you wouldn't see that much in a game, probably closer to a million triangles/frame. perhaps a bit less.
As for effects - like I said, that is the Unreal Engine. The lights used a precomputed shadow map and a precomputed specular map, and the clothes were environmentally bump mapped. I looked extremely hard for motion blur, and I didn't see any. Basically it amounts to 4 texel units and 8 lights/polygon. Sounds an awful lot like the nv15 specs to me.
God knows how you'd transform that many vertices, but then again, it has a nVidia card in it, and they've got hardware transform and lighting in their most recent cards...
The yet-to-be-released nvidia cards (NV11, NV15), with hardware T&L don't even approach the 600,000,000 matrix x vector ops that you would need to draw 300,000,000 triangles. The Geometry coprocessor would need to be on the order of 15-20Gigaflops to handle that.
If the card actually can do 300 million tris/second (I really doubt this), it appears that all the transformations are handled by a geometry coprocessor, and not a floating point coprocessor. This will make a huge difference, since it means that expensive physical interactions which the PS2 can offload to its vector units must be handled by the Athlon in the XBox, since the geometry coprocessor can only handle geometry. This also means that after manipulating vertices in software, you will need to manipulate them in hardware all over again (triangle setup on the PS2 is much more efficient than on the PC). Perhaps what they meant was that the XBox can draw 300,000,000 particles / second (2x the Graphics Synthesizer). That's believable.
To compare the PS2 has a peak theoretical rate of 60 million tris/second. If you really push it in a tech demo you can get 20-25 million/second. ingame you get closer to 10 million/second max.
From talking with PS2 programmers, if you push it in a game, you can get 20-25. Tech demos are typically 30 or so, with some incredibly cool effects (hypertextures, reflection/refraction, morphing, etc.). After all, that's what a tech demo should be - new things to try, not doing old things better.
If those are indeed the 3D capabilities of the XBox (and typically Tech demos do push a console to graphical levels not regularly reached by games), then the XBox is crap, and their 300 million polygon/second number is a fairy tale.
To truly impress, they should have included any of the following features in the tech demo:
Real-time bezier surfacing real-time reflection/refraction properly weighted clothes properly weighted hair extremely detailed 3D scene dynamic lighting inverse kinematics dynamic soft-body animation hypertextures amazingly detailed characters (e.g. this) 3D morphing Behavior synthesis Depth of field/motion blur/heat warping
As it stands, they merely showed off a nice version of the Unreal Tournament engine. 8 shadow-mapped lights (and the shadow map was pre-computed!), 2 motion-captured (and Shen Mue resolution, at that) characters, environmentally bump-mapped clothes, a texture mapped arena, and minimal facial animation. Every one of Sony's demos were far more impressive.
Geometry acceleration support is ofcourse about the API supporting it, and the hardware being on the way. Ofcourse GL supports it as well, and in it's most basic form has supported it since its existance. But the D3D methods are much more flexible. This will be even more so with DX8.
GL geometry acceleration really couldn't get much easier: glLoadIdentity(); glRotatef(0,1,0,rot_angle);
And that same code will run on any platform, from software-driven Pentium-100s to hardware acceleration on Onyx2 Infinite Reality racks, as quickly as the driver developers choose to optimize their drivers. I haven't worked with DX7's geometry acceleration, but from what I hear, it's considerably more obfuscated, and requires all old code to be rewritten to take advantage of it.
MS indeed is ultimate judge about what gets implemented, but since this interaction process has started it's never really worked out in any bad way as I see it.
The only problem is that Direct3D will permanently be tied down to the current hardware. OpenGL 1.1 still has features (glConvolutionFilter, glTexImage3D, etc.) that have yet to be implemented in hardware on any consumer card, and has supported geometry acceleration since its inception. Plus, if a feature isn't available in hardware, OpenGL has a guaranteed software fall-back renderer for all features in the spec - something Direct3D sorely needs. Sure, a few new features have been added to improve performance (glMultiTextureARB), but that's more a result of card manufacturers optimizing their texel units for write-only, rather than read/write. Multi-texturing can be just as easily implemented with alpha blending two images; however, when writes are 2-3x faster than reads, it's considerably more efficient to pre-alpha blend the pixel in the texel unit and write once, then to write once, read back in, and write again.
And no, there is no reason for D3D to be tied to Windows. It could be implemented on Linux. Just nobody does it. Because it wouldn't be considered very cool:)
Actually, since neither ActiveX nor the Component Object Model exist in a complete enough form for any other OS (MacOS has a partial implementation), it's easily arguable that DirectX will remain tied to Windows. Not too many operating systems have any idea what extending IUnknown is supposed to do. Hell, check the Cygnus Gnu-Win pages - it's difficult just to port the DirectX libraries to a different Win32 compiler, let alone an entirely different operating system!
Actually, although the general public (e.g. people that would be most likely to buy Aptivas and Brios), the graphics lab I work in could always use more power. When single processes can take 24+ hours on 8-way 250Mhz SGI Onyx2s, higher speed processors are definitely useful. Rendering times and other similar massively computationally expensive processes can be sped up considerably with increased processor power.
However, that said, Intel's current push is really pathetic. When processors like Sony's Emotion Engine at 300Mhz are 2x faster than Intel's flagship at 1Ghz, Intel's design is clearly wrong. The fact that Intel has had difficulty reliably producing P3s at 650Mhz or above doesn't help their situation. This is a wonderful marketing move (and will probably convince less techno-savvy consumers away from AMD), but really amounts to nothing more than behaving like an attention-starved child.
If AMD could get SMP Athlon's working, Intel would be in severe trouble. Higher clock speeds are relatively useless - a 4-way 650Mhz Athlon would dust almost anything Intel could muster. I'd love to have a dual 650 Athlon for rendering; unfortunately, it seems as if AMD's methodoly mirrors Intel's - more speed, not better speed. Sigh.
Actually, as a current student of "the art," I can say that there is quite a bit more to CS than one may imagine. Analysis of discrete algorithms, combinatorics, and number theory do play quite a large role in CS (we use Cormen, Leiserson, and Rivest's Introduction to Algorithms, which I find to be an excellent book); however, there is also the formal logic side. My formal logic class used Manna and Waldinger's The Deductive Foundations of Computer Programming. Honestly, I didn't care for the class, and hated the book, but if you truly wish to follow "The Art," your best path would probably be to forget learning computer programming and languages for a while. Honestly, learning a new language is a process that should take at most 2-3 weeks (maybe more if you are moving from structured (C/C++/Pascal) to functional (Lisp) paradigms), but the thrust of computer science is hardly the syntactical usage of one particular programming language, nor has it ever been. When I write programs (currently working on an operating system, and have done 3D games, paint programs, and others in the past), my time division is about 35% design, 15% programming, and 50% realizing that I typed >= rather than >, or + rather than -. Design is entirely language and platform-neutral. I create a pseudocode as I go to explain what my algorithm, data structure, or function should do. That is what CS as "the art" teaches - design independent of all bounds. Programming is then the implementation of that design - it will teach you things like platform-specific memory alignment, cache optimizations, etc., and how to take your elegant design and make it scream on the targeted hardware. It's much easier to learn the facts for a hardware platform and implement a canned algorithm than it is to learn the philosophy behind algorithm design. Computer science will teach you design, computer programming will teach implementation, and experience teaches debugging. If you just want to write cool programs, then Knuth's (or Manna's, or CLR's) books are entirely unnecessary, and you would be better served by a reference manual such as the Red Book. However, if you do find the art interesting, then be prepared for a lot of seemingly useless math, esoteric concepts, and trips to the mental ward. I find most of it very interesting, but there is a reason my friends expect me to end up in a rubber room before I graduate.
OK, I go to a school (with an extremely well-respected Computer Science department) where we don't pre-choose our major, and the same phenomenon exists.
One of the easily identifiable reasons why there has been a decline in computer science majors is due to the rise of the hybrid major. It is a recognizable fact that females have a much higher presence in humanities courses in high school than they do in computer science courses (my AP English class was divided 7/27 (7 guys, 27 girls), whereas my AP CS class was 12/0). While this may be in part due to the "table of [elitist] geeks" that has been mentioned, I distinctly remember seeing quite a bit of enthusiasm among my female classmates during humanities discussions. After reaching colleges, they may uncover a hidden interest and talent for computer science; however, developing operating systems, raytracers, and compilers often comes at the cost of other topics - namely humanities. It's difficult to simply shed an interest that you held for 4 years to embrace a field entirely foreign to you.
Enter then the hybrid major (Symbolic Systems). Rather than dropping all participation in fuzzy classes, you can divide your major among your new CS interest and your old (lifelong?) fuzzy enthusiasm. I know my university has seen a tremendous increase in enrollment in our SymSys major (by both females and males), often by people who were recently exposed to CS, but unwilling to pursue it full-time (I have worked over 90 hours/week for some projects, so it truly is a full-time commitment).
Maybe I'm using too much hyperbole, but I think it is fair to assume that the majority of females that enter the computer science major didn't intend on pursuing computer science the majority of their lives, and some may have chosen simply based on positive experiences in their first college course or two. I'm basing this assumption simply based on the extremely low enrollment of females in high-school CS courses, so I think it is at least partly justifiable. When given the option of a half-techy/half-fuzzy curriculum division (such as SymSys, Human-Computer Interaction, or any other hybrid major) or a balls-out techy curriculum, I would assume that most of these people would choose the hybrid. Maybe my school is unique since we don't need to declare majors until our junior year; however, I think a truly correct study would compare the declining enrollment in pure computer science (especially among females) with the burgeoning enrollment in hybrid fields (especially among females). Hell, there are times when I'm sitting through formal logic classes where I've thought about the appeal of the hybrid major - all of the pay, none of the bullshit. After all, proving that 0++ is equivalent to (0+)+ and that that is equivalent to 2 is really about as enjoyable as a trip to the dentist's office. I love CS, but there are some things that are just ridiculously painful about majoring in it (e.g. the parts that *aren't* in a hybrid curriculum).
forgetting the fact that vulgar language is an ugly black mark on those who use it. I don't care if you think I'm old-fashioned, I just don't like it!
I'm going to assume that this wasn't flamebait or tongue-in-cheek; however, feel free to correct me.
I've had this discussion several hundred times, with every ethnic and religious combination that I can think of, and I have yet to hear a convincing argument on why vulgar language is an "ugly black mark" on those who use it.
When I read Gravity's Rainbow (one of the best novels of the past century), I never for a second thought that Pynchon's sentences were anything but a comment on his genius, but I'm sure that any dozen pages or so of it have enough profanity to ensure it a solid R rating.
The problem with blanketly saying that profanity is an ugly mark on its users is that it neglects the fact that a profane sentence is not necessarily an obscene one. I can easily take perfectly non-profane words and form them into a sentence that would turn the average sailor green, or I can cuss a blue streak and without anyone would think about it. The problem comes from the context of the words - not the words themselves. I really fail to understand why "fuck" is deemed inappropriate by the majority of American churches (especially seeing as fucking (procreating) is the only legitimate metric for marriage), but a word that carries as many awful denotations as "crucifixion" can be thrown about by the average 3rd grader.
Your choice to not use profanity is just as valid as mine to use it. The metric for intelligence shouldn't be whether an individual uses a very specific subset of the English/Japanese/Spanish/etc. vocabulary, but the entire subset that said individual uses. I don't have much respect for people who can only use obscenities; however, I have just as little respect for people who constantly interject "like, uhm..." and other nonsensical gibberish into their sentences.
While you may find the lack of clean humor unfortunate, I find it vindicating. Old-style conservatism is one of the most authoritarian and fascistic hold-overs from Euro-American history. It caters to the belief that what works for someone important must work for everyone, and if somebody more important than me doesn't like something, than I shouldn't either. However, all religions agree that I am a sentient being - therefore, I am capable of making choices that affect me by myself, and you by yourself. You may not agree with me, and I may not agree with you, but dammit, that's how it is, and nobody can deny us that (I think Voltaire wrote something similar...). South Park may cater to "the lowest common denominator" (I tend to disagree with that assertion), but dammit, somebody sure needs to - how else would I, as an elite member of this LCD, get my weekly ration of comedy?
I doubt what I believe will influence your opinions at all. In fact, I am probably already burning along with Trey Parker, Matt Stone, Adolf Hitler, and the rest because I choose to eschew some of the mores of old-style conservatism. I personally try to paint as few "ugly black marks" on people as possible (you know, it's difficult to find beauty when everything has an ugly black mark on it) - and for that reason I set my "offensive tolerance" quite high (e.g. it'll take more than a few dirty words to offend me), and my "black mark" level even higher. I also think that if everyone started painting significantly fewer ugly black marks, the intolerance and prejudice that culminates in nightmares like Columbine would also disappear, but that's a topic for a different rant.
Finally, if they need a 1 GHz processor to work this thing, then they're doing something wrong. The PS2 achieves 25M pps with only a dedicated 266 processor. A 1 GHz processor sounds like infinitely more than what's needed to power a cutting-edge gaming platform at this time.
Mhz are one of the most meaningless speed measurements these days, *especially* when referring to console processors.
The PS2 has a 300MHz processor, but what is hidden is that fact that it has 2 additional 300MHz vector coprocessors, creating a total of about 12 floating point accumulators and 6 dividers that can be used simultaneously, yielding (yet another meaningless statistic) 6.2Gigaflops. However, since VU0 and VU1 are heavily optimized for vector processing (as opposed to dividing by 0), the majority of the 6.2 billion floating point operations can be utilized in the average case. Add to that the Emotion Engine's fully 128-bit design, and you've got a processor that can outrun any intel processor up to ~1.5Ghz, and a memory bus that can actually supply needed data when it is needed. Granted, a Willamette 1500 would be able to do integer operations much faster than the Emotion Engine, making it a better general-purpose CPU; however, for gaming and signal processing integer operations are minimally important, making the EE a much better processor for multimedia purposes.
What MS has forgotten is that the PC platform is a terrible architecture for gaming. Everything about it is just mere upgrades on 20 year-old technology. Maybe Microsoft has something really sneaky up their sleeves, such as using a 1Ghz SH-4 (Dreamcast processor) and a custom-designed bus to an optimized GeForce 256 processor. That could (theoretically) approach PS2 level performance, but I'm not going to give MS that much credit.
A recent poll on/. showed that most readers were students. Based on the language used in many posts, it can justifiably be assumed that most of these students are still in high school.
Now please, o exalted one, show me a high school that covers not just introductory nuclear physics, but also Cold Fusion 101. We only got to introductory E&M in AP Phycics C in my high school. I consider myself a nerd (engineering student at Stanford), and I know jack about Cold Fusion - it's not something that I read (or research) religiously. Very few people do... proportoinal to the population of the world, there are very few nuclear physicists.
However, if the facts in the article are true, it does suggest something. I can't prove that the article is 100% factually accurate, I can't even prove if it's 50% factually accurate. But I would like it to be true (1.21 gigawatts!). There is a reason Americans (and almost every other modern culture) are suckers for conspiracy theories, and it goes hand-in-hand with increased national cynicism. Cynicism isn't something that justs randomly manifests itself in cultures - there has to be some national reason for it. Americans can no longer believe their government... it's not the responsibility of the Americans to find renewed faith in their administration, it's the responsibility of the administration to prove its honesty. If the article is true, then they've got some explaining to do.
Hewlett-Packard will be releasing a DVD ReWriter (DVD-RW) in September. It will write DVD-RWs at 1.5MB/s (maybe 1.25x DVD speed), and CD-RWs at around 8 or 12 speed. The DVD-RW discs will be 3.0GB single side, single layer, but they may only work in DVD-ROMs... I don't think they will work in DVD players. The street price should be $699, with media being $30 / blank, and the drive will be SCSI only.
This is about as likely as ICQ becoming a commercial app due to AOL's purchase. If anything it will get rid of the $10 shareware fee on WinAMP, since Nullsoft is now backed by the huge pockets of America Online, rather than the 20 or so users that ever registered it.
Of course, if AOL decides to charge for WinAMP (this is about as likely as Microsoft open sourcing Windows), you could just continue to use 2.10, which works wonderfully. I only just upgraded from 2.05, and I really don't notice too much of a difference. I didn't want to use any AOL programs for a while, but they didn't pawn ICQ space to the highest bidder, and I doubt they'd do it to WinAMP. I wouldn't be terribly surprised to see AOL license the Sorensen algorithms next... it looks from their previous purchases (real-time communication and conferencing, audio compression/decompression) that they are looking to become a net tech company. Video compression is the next step.
It depends on what audio format is used. All reels have a 2-channel analog stereo track on them for backup purposes, but additional digital audio tracks are handled differently for each format:
DTS: A seperate CD is included with the reel that has the DTS 5.1 encoded audio on it
SDDS: 2 highly compressed digital tracks (identical, in case one breaks) are run along the rings of the reel, making this format highly susceptible to being damaged. Allows up to 8 channels (5 front, 2 surround, 1 LFE).
Dolby Digital: I think this uses a seperate reel in addition to the movie reel, but I could be mistaken. Allows 5.1 discrete channels, Dolby Digital EX uses the same 5.1 channels, but EX processors matrix decode (a la Dolby Pro Logic) a rear-center channel.
Slashdot Mirror
Alright, here is the easy calculation:
AMD's Athlon at 600Mhz produces 2.4GFlops, according to AMD's white papers.
The Emotion Engine at 300Mhz generates 6.2GFlops, according to Sony's white papers.
I'll leave the calculating up to you.
Your comment is mostly right, but here are some minor corrections:
First, the Emotion Engine runs at 300Mhz in the PS2. It was initially announced at 250Mhz back in March of 1999, but within 3 weeks of the initial specs (confirming 55 million polygons per second), Sony changed their design to 300Mhz (66 million polygons per second).
You really shouldn't try to do comparisons using cycle counts, since in most cases the operations cost the exact same (integer ADD and SUB cost 1 with a 1-cycle latency on both x86 and MIPS architectures), and in others the x86 has extra concatenated operations (BSWAP, BTEST, etc.) due to its CISC nature that have no direct equivalent on MIPS systems.
As for non-vector performance of the Emotion Engine... The cost of additions, subtractions, and bit operations on a MIPS III architecture processor (like the Emotion Engine) is virtually identical to the cost on an x86. The Emotion Engine has a decent advantage on branch misprediction, since the EE only uses a 6-stage pipeline (as opposed to a 20-stage pipeline on the Wilamette). This basically means that if the processor incorrectly guesses where a program will jump, only 6 operations need to be performed for the Emotion Engine, while 20 will need to be performed by the PC. However, since branch mispredictions will happen more frequently on the Emotion Engine than the PC (PCs have a lot of extra hardware and cache to compensate for their long pipelines), this is basically a net-zero difference. The big difference is that the PC runs at 3x the clock speed of the Emotion Engine, which will basically allow it to perform 3x as many elementary operations in the same amount of time. A better comparison would be to say that for just integer operations, the Emotion Engine at 300Mhz is roughly equivalent to a Pentium II at 300Mhz.
I've debated this in quite a bit of detail on comp.games.development.industry, and basically, after tons of calculations, we came down to the following:
Athlon @ 1.6Ghz ~= Emotion Engine @ 300Mhz.
Athlon @ 1.5Ghz EE @ 300Mhz.
Not too bad, seeing as the EE has 10.7M transistors, dissipates 13W, and has on-board MPEG2 decoding. For use in a workstation environment, where processors can get much warmer, cranking the EE up to 600Mhz wouldn't be out of the question. 2x EE at 600Mhz would best almost every workstation ever designed. Granted, all the tools (Alias, etc.) would need to be rewritten for it.
I pay $80 per year for a 100BT in-room connection on Stanford's network. The fact is, if you're not explicitly paying the school for the hook-up, then the bill is discreetly placed in your tuition somewhere. $80 is far less than what a 100 megabit connection would cost me for a year, so I'm not going to argue.
Here are a few accounts that need settling:
1) OPEC needs to be broken up by the terms of the Sherman Anti-Trust Act. Their actions undeniably restrict trade, and they do constitute a monopoly on the crude oil industry, and high gas prices have hurt me. Therefore, it's legal to sue them.
2) India's atmospheric nuclear tests spread several particles-per-billion into the air that I breathe. This will probably cause cancer, so just like power companies have been sued, India should be sued.
3) The Soviet Union's belligerence caused taxes to rise into the stratosphere as a result of programs used to balance their military muscle (strategic defense initiative being a primary culprit). If the USSR weren't so damn belligerent back in the 60s, 70s and 80s, I wouldn't be losing 33% of every paycheck. Based on lost wages, interest, and inflation accrued during and after the inauguration of these programs, the American people have lost billions of dollars. This is obviously grounds for a class-action lawsuit against the Kremlin.
Honestly, the US has enough bad laws without passing totally asinine ones. Sue the lawmakers for being paid with our money to debate issues like the legality of suing foreign governments.
If anything, it is a creative way of repaying foreign debts ("What do you mean we owe you $100 billion? We sued your asses and won! We just deducted our $100 billion damages from what we owed you, so we're even!")
The last line should read W-Tab, and pressing Tab again cycles through entries.
;(.
I forgot that slashdot parses HTML
Tab auto-completion is a well-hidden feature in the registry.
Under 2000, it's HKEY_LOCAL_MACHINE\Software\Microsoft\Command Processor\Completion Char == 0x0000009
and
HKEY_LOCAL_MACHINE\Software\Microsoft\Command Processor\PathCompletion Char == 0x0000009
Under NT4, you do:
HKEY_CURRENT_USER\Software\Microsoft\Command Processor\CompletionChar == 0x0000009
It works a little differently than under Unix, though. Pressing W selects the first entry that starts with W. Pressing again cycles through the other entries.
This will probably sound like an advertisement for Win2K, but I for one am tired of the unbased Win2K bashing on Slashdot. Note that everything I write is based on my own personal experiences with Win2K, as I run it on my personal machine.
Consider me one of the one million nobodies that legally upgraded to Win2K this month. I was previously running NT4SP5, but because of my hard disk partitioning, I had to reformat and do a fresh install.
All the talk of 65k bugs in 2k really means nothing without actual usage backup, so I'll be one of the first people to actually post how Win2K runs, at least on my machine.
First, Win2K is big. I wasn't expecting 900MB for the OS, but to be fair, 60MB is used by DRIVER.CAB (all the included drivers), 192MB by my swap file, and another 70-80MB by the multilingual options (30MB by nihongo alone). Granted, even subtracting out those options, Win2K is far and away the largest OS I have ever seen or used. But, having installed it with more than enough space left over for my programs and MP3s, my real question was how it would compare to NT4SP5, which had run wonderfully for over a year.
First, the bad. Win2K has several minor bugs that I've noticed. Most of these pertain to little graphical nuances in programs - occasionally menu bars don't always repaint themselves correctly, and on some occassions cursors don't alwyas update on applications like ICQ. All in all, nothing I can't live with (or that a service pack won't fix). NT4SP5 didn't have these problems - of coures, NT4SP5 didn't support USB, Plug and Play, AGP texturing, DirectX, or numerous other things that are in Win2K. Another (minor) downside is that it took a little more effort than it should have to enable tab auto-complete in the Command Prompt, but this was also the case with NT4.
Now, the good. Installation was so amazingly painless that I thought I had just installed Unreal Tournament. Win2K boots from the CD, and asks how you want to format your disks (just like NT4). However, dozens of SCSI and mass-storage drivers are included, even my Diamond Fireport 40 (thus saving me the effort of reinstalling a floppy drive to install Fireport 40 drivers). Then, 2K asks you 2 other questions throughout installation - one is what language options you want installed, and the other is what networking options (system name, etc.) you want installed. This can be configured from the command line, so installation can literally just be a matter of inserting the CD, rebooting, and letting it run for an hour. When I returned, *all* of my hardware, sans my DVD Decoder card (Creative Dxr2) had been installed - my RivaTNT, Fireport 40, Sound Blaster Live!, 3C905, everything was in and running just fine. After downloading and installing NVIDIA's new drivers (to enable hardware OpenGL support), all of my OpenGL applications worked flawlessly.
Performance - it is quite apparent that NTFS' disk caching is better under 2000 than it was under NT4. Other things seem just slightly faster than under NT4SP5, and nothing seems slower. Basically, if you liked NT's performance, you won't be disappointed with 2000 (I don't have any data for performance with FAT).
Stability - so far 2000 seems as stable as NT. Granted, this is only something that time can prove, but so far everything is extremely responsive, and I have noticed no hiccups at all. Another major improvement is that you do *not* need to restart anywhere near as often for applications. This is definitely nice; however, unless you install applications 3-4 times daily, probably won't matter much in the long run. Also included is a boot console interface (for major emergencies). I never understood why this wasn't included in NT4; however, while it is definitely on the sparse side, the CLI may prove invaluable, if driver manufacturers start releasing shitty drivers that cause BSODs when starting 2K.
Usage - this is all a matter of taste, but I like some of the Active Desktop features (Quickstart bar, etc). There is lots of alpha blending used in the UI - this will never be useful, but it is pretty cool, at least until the novelty wears off. Having DirectX is really nice; however, I will always be a proponent of OpenGL, so while I will enjoy being able to play the games that are DX-only, I will privately be wishing that they used OpenGL instead. Also, 2K has some very nice support for multimedia files (read: MP3s) built into explorer.
All in all 2K is a worthwhile operating system, and a very nice update to NT4. Not having to deal with the archaic 'Devices' control panel is an extreme relief (stupid NT3.51 hold-over), and a lot of the new features do make doing mundane tasks (installing hardware, managing user accounts, etc.) much more palatable than under NT4. Also included is a nice "auto-login" feature, for anyone interested in using 2K as a work-alike replacement for 98.
I've heard horror (and success) stories about upgrading from 9X->2000, but so far my experience has been very positive. If anyone on Slashdot uses Windows but hates 98's "reliability" or wishes NT played more games, 2000 would be a decent upgrade.
I'd like to take this opportunity to announce the specs for my new console, with an ETA of April, 2001.
-------------------------------
Name: Extrema Bullshit
CPU0: Optimized AMD Athlon Thunderbird-700 processor (2.8GFLOPS).
CPU1: NVIDIA NV26 GPU, 600m Matrix x vector operations per second, capable of transforming and lighting 150 million triangles (100 million vertices) per second. Collision detection and vertex blending is integrated into the core at no extra cost.
SPU: Emu E12000 Environmental Extender, real-time 1536kbps DTS-ES 6.1 compression, THX Ultra Certified, 130 dB signal/noise ratio, 96 hardware accelerated 3D voices. Psychoacoustic technologies allow simulating sounds along the vertical axis.
Rasterization Processor: NVIDIA NV26R, 4.8GPixel/sec fill rate, 8 programmable hardware texel units and polygon normal interpolation, yielding 120 million 8-level multi-textured phong shaded 25-pixel polygons / sec. Supports all HDTV resolutions including 1920x1080P, all at 8-bits per channel plus an 8-bit stencil buffer and 24-bit Z Buffer. Also, supports 15 million triangles/second in distributed raytracing mode, with radiosity.
Memory:
CPU0: 64MB Wide Direct RDRAM, 6.4GB/sec
Rasterization: 32MB Texture RAM, 12MB frame buffer
embedded, 64GB/sec
SPU: 4MB compression cache and 16MB sample cache, 1064MB/sec
Storage: 8X DVD, 75GB hard drive, 8MB memory card
Media: CD-ROM, CD-DA, DVD-A, DVD-Video, DVD-ROM, CD-RW, CD-R
Network: built-in selectable 10/100/1000BT ethernet, and an optional ATM switcher ($75)
OS: Solaris MultiMedia
Development: Open
Output: HDTV, SDTV, NTSC, PAL, VESA. 1x Progressive, 1xSVideo, 1xComposite, 1xVGA, 1xSCART, 1xStereo, 1xTOSLINK
Extra: Plays DVD movies, DVD Audio, VCD, CD audio, Playstation, Saturn, and DirectX games. And it can be programmed to download porn while you sleep.
Cost: $300
This does 2 trillion operations per second, so you can get double photo-realistic results.
-----------------------
There are some schools these days (I believe Villanova may be one of them) that require their undergrads to purchase laptops from the 'school source'. I'm not sure if anything but these laptops is allowed onto the school network, either.
Sad but true.
Here is a radical idea to take care of vast music piracy, and actually improve the quality of music that we have:
The government (National Association for the Arts is the bureau, I believe) subsidizes professional artists (to some degree), who then sell their original work to collectors for considerable profit. What if music were to stop being treated as an industry, and more as an art form (using a model similar to that used for fine art)? Then the free trading of music via Napster and the like would be completely legal and legitimate, and young'uns wouldn't learn how easy it is to break toothless laws.
A side effect of this would be that the only people remaining in music would be those that actually love making music. The romanticized image of the 'rich rock star that gets lots of girls' would become archaic, and modern industry evils like boy bands would be no more. The good artists, that actually contribute to the medium (there are a few bands (and quite a few rappers) remaining that would fit this mold, but not the Limp Bizkits and Backstreet Boys of the world) would be able to maintain good lifestyles, and culture wouldn't be so overrun with the banality that is pop music.
I doubt that this is realistic, but it would definitely solve quite a few problems - the enjoyment of music would be free, artists wouldn't sacrifice 90% of what is rightfully their profits to industry coffers (granted, they wouldn't be making much more money, if any), and the quality of music would improve considerably.
Actually, all the videos were fake! If you look at NVIDIA's press release (and Microsoft's), it becomes apparent that neither the graphics chip nor the sound chip will exist for another year. Every single video they showed was a mock-up, and none of the high-res screen shots for the mech demo (on Microsoft's home page) had any motion blur to speak of - although antialiasing was in full effect.
.8GB/sec, PC200 = 1.6GB/sec, PC400 != 6.4GB/sec).
Apparently Microsoft has also claimed a 6.4GB/sec bus; however, that's reportedly using PC400 SDRAM, which neither exists nor runs at 6.4GB/sec (PC100 =
All in all, the X-Box specs amount to more of what Microsoft would like to achieve than what it actually will achieve. The prototype hardware will not exist for another year, at least, and I suspect that NVIDIA will have some major difficulties producing a processor that is 10x faster than the NV15, in roughly the same amount of time. Given Murphy's law, Microsoft will end up with 2 choices: produce a less-spectacular platform for Christmas 2001, or risk delay until 2002. If they delay until 2002 the X-Box is guaranteed DOA. The Dolphin and PS2 will be on their 2nd and 3rd generation games, respectively, and will cost considerably less than the X-Box. If people have just bought a Dreamcast, a PS2, and a Dolphin, it's unlikely that they'd be interested in also purchasing an X-Box, especially since all the games announced so far will be ports of Playstation/Dreamcast/PS2 games. If Microsoft chooses to release a less powerful console, their specs will probably end up being overestimates by at least a factor of 3 (NVIDIA's track record would pin the NV25 at about 7x faster than the GeForce, or ~110 million triangles/sec), which will not be enough muscle to really technically outshine the PS2 or Dolphin.
Things don't bode well for X-Box.
Huh?
AGP is a bus from the graphics card to main system memory, to allow larger texturing areas. It communicates with the chipset, and actual performance is dependent upon the speed of memory in the system, and the processor's demand for memory. PC133 SDRAM has a maximum transfer rate of 1064MB/sec. AGP4X has a maximum transfer rate of 1064MB/sec. That means that for AGP4X to run at full speed, the processor could not be accessing memory. In order to combat this, things the frame buffer and some texture memory are stored on the card itself. This memory has a higher clock speed (on the GeForce it is 4.8GB/sec) thanks to a wider bus (256 bits on the card vs. 64 in main memory).
The XBox specs list a Unified Memory Architecture - that means that the frame buffer, vertex information, system RAM, textures, and sound all reside in the same pool of memory, and all communicate on the same bus (e.g. like AGP, only for all devices, not just the graphics card).
To just read 300 million triangles (~150 million vertices, for sufficiently large models) to the rasterization unit requires 3 floats per vertex, 4 bytes per float, or 1.8 GBytes/sec of memory bandwidth - which will quickly saturate the 3.2GB/sec maximum of dual channel RDRAM (since it is UMA, the memory will need to be external to the processors, and the bus will be limited to 64-bits, yielding a peak bandwidth of 3.2GB/sec). Then you need to add in textures (an additional 2 floats per vertex, or 66% more memory use), texture coordinates, rasterization, sounds, and everything else, and you will have some massive processor stalls. In addition to the fact that I have serious doubts about the 300m number, even if it is a theoretical maximum, as it is implemented in the XBox, you could expect a realistic maximum of ~45m triangles/sec. However, since the NV15 only has 2x the geometry acceleration of the GeForce (or about 30m triangles/sec), I seriously doubt that nVidia is sitting on an accelerator that is 10x as powerful as one that they have yet to release.
In contrast, the Playstation 2 has the chipset integrated into the Emotion Engine, with a 3.2GB/sec pipe from the EE to main memory. There is a 1.2GB/sec pipe from the EE to the Graphics Synthesizer (to stream textures and vertex data), which means that even when the GS is transfering as much as it can, the EE still has over 2GB of bandwidth remaining. Sounds are stored in 2MB of sound RAM (so playing sounds has no effect on system bandwidth), and the frame buffer (and texture cache) are stored on the embedded DRAM in the GS. The eDRAM runs at 48GB/sec, as compared to the 4.8GB/sec on the GeForce, allowing much more innovative effects, such as motion blur and heat warping, to be implemented easily.
And no, a bus for pure graphics could not accomplish what I stated. Almost everything that I mentioned requires massive software FPU capability, in addition to transformations. The Athlon 600 has a peak 2.4 GFlop performance (the P3-600B is even less). If you wanted to evaluate and tesselate bezier surfaces, you wouldn't be able to do that on the graphics card - it would need to be done by the CPU. However, due to congestion in system RAM (due to the UMA nature of the XBox) and minimal instruction decode pipelines (compared to the PS2), x86 processors wouldn't be able to generate anywhere near as many of these as the PS2 (remember - the PS2 has 2 floating point coprocessors, not geometry coprocessors). Physics calculations are the same case.
Realistically, even at best case, the XBox won't be any more powerful than the PS2. Even if the specs for the NVIDIA chip are accurate (I seriously doubt this), the entire design of the XBox will be a huge limitation, and totally limit performance. The XBox will only be marginally faster than a PC with an i840 chipset and an NV15 graphics card.
Basically, the design philosophies for the two systems boils down to:
PS2: Work smarter, not harder
XBox: Work harder
And the end result is that the XBox will have 10x the transistors of the PS2, and will have equal performance in some areas, and worse performance in others. The fact that it's not even due for another 18 months will be a major deterrent to its success.
Traditionally Tech demos show off the graphics/sound abilities of their hardware - many of the things you mention are things you'd do in software.
What they were showing off was the huge number of polygons that they can throw around, and they are throwing *a lot* of them around in that clip. They also appear to have bump mapping as well as specular reflection on their surfaces. As for realtime curves, it's quite possible that they were doing that on the characters, but I doubt it. Doesn't seem much point.
The point of doing realtime curves is to show off just how powerful the XBox is! Anybody can throw a bunch of triangles on the screen; however, no current game (for PC, Dreamcast, etc.) uses real-time bezier surfacing, because evaluating the curve and then tesselating it is a highly CPU-intensive task. The PS2 is capable of that, and still spitting out 15-20 million triangles (all tesselated from bezier surfaces). Hypertextures, realtime reflection/refraction, physical interaction to hair/cloth level - all are things that have yet to be done in games, and if the XBox is remotely as powerful as they hype it to be, should be able to handle. Honestly, while the R4 girl demo for the PS2 didn't look as nice as the mech demo for XBox, it was considerably more impressive, since it showed facial animation, weighted clothing, weighted hair, and extremely high polygon counts. Tech demos show off how powerful new technology is - radically new effects for radically new technologies - the XBox showed off more of the same.
And then what it showed looked like 50,000 triangles per frame, tops. The martial artist didn't look much better than the characters in Shen Mue. The model that I linked to above has 2 million triangles - the XBox should be able to spit out 2 of those at 60 frames/second easily. That mech did not have terribly complicated geometry, and that room was a textured cylinder, with some cross-beams.
It was *much* more than a Unreal engine - the models were at least an order of magnitude more detailed, and there also appeared to be motion blur on the characters, although that could also be as a result of being put on video... Anyway, the x-box site claims a thoughput of 300 million triangles/second, so at 60 fps, it would have a peak performance of 5 million triangles/frame. In reality you wouldn't see that much in a game, probably closer to a million triangles/frame. perhaps a bit less.
As for effects - like I said, that is the Unreal Engine. The lights used a precomputed shadow map and a precomputed specular map, and the clothes were environmentally bump mapped. I looked extremely hard for motion blur, and I didn't see any. Basically it amounts to 4 texel units and 8 lights/polygon. Sounds an awful lot like the nv15 specs to me.
God knows how you'd transform that many vertices, but then again, it has a nVidia card in it, and they've got hardware transform and lighting in their most recent cards...
The yet-to-be-released nvidia cards (NV11, NV15), with hardware T&L don't even approach the 600,000,000 matrix x vector ops that you would need to draw 300,000,000 triangles. The Geometry coprocessor would need to be on the order of 15-20Gigaflops to handle that.
If the card actually can do 300 million tris/second (I really doubt this), it appears that all the transformations are handled by a geometry coprocessor, and not a floating point coprocessor. This will make a huge difference, since it means that expensive physical interactions which the PS2 can offload to its vector units must be handled by the Athlon in the XBox, since the geometry coprocessor can only handle geometry. This also means that after manipulating vertices in software, you will need to manipulate them in hardware all over again (triangle setup on the PS2 is much more efficient than on the PC). Perhaps what they meant was that the XBox can draw 300,000,000 particles / second (2x the Graphics Synthesizer). That's believable.
To compare the PS2 has a peak theoretical rate of 60 million tris/second. If you really push it in a tech demo you can get 20-25 million/second. ingame you get closer to 10 million/second max.
From talking with PS2 programmers, if you push it in a game, you can get 20-25. Tech demos are typically 30 or so, with some incredibly cool effects (hypertextures, reflection/refraction, morphing, etc.). After all, that's what a tech demo should be - new things to try, not doing old things better.
To truly impress, they should have included any of the following features in the tech demo:
Real-time bezier surfacing
real-time reflection/refraction
properly weighted clothes
properly weighted hair
extremely detailed 3D scene
dynamic lighting
inverse kinematics
dynamic soft-body animation
hypertextures
amazingly detailed characters (e.g. this)
3D morphing
Behavior synthesis
Depth of field/motion blur/heat warping
As it stands, they merely showed off a nice version of the Unreal Tournament engine. 8 shadow-mapped lights (and the shadow map was pre-computed!), 2 motion-captured (and Shen Mue resolution, at that) characters, environmentally bump-mapped clothes, a texture mapped arena, and minimal facial animation. Every one of Sony's demos were far more impressive.
GL geometry acceleration really couldn't get much easier:
glLoadIdentity();
glRotatef(0,1,0,rot_angle);
And that same code will run on any platform, from software-driven Pentium-100s to hardware acceleration on Onyx2 Infinite Reality racks, as quickly as the driver developers choose to optimize their drivers. I haven't worked with DX7's geometry acceleration, but from what I hear, it's considerably more obfuscated, and requires all old code to be rewritten to take advantage of it.
MS indeed is ultimate judge about what gets implemented, but since this interaction process has started it's never really worked out in any bad way as I see it.
The only problem is that Direct3D will permanently be tied down to the current hardware. OpenGL 1.1 still has features (glConvolutionFilter, glTexImage3D, etc.) that have yet to be implemented in hardware on any consumer card, and has supported geometry acceleration since its inception. Plus, if a feature isn't available in hardware, OpenGL has a guaranteed software fall-back renderer for all features in the spec - something Direct3D sorely needs. Sure, a few new features have been added to improve performance (glMultiTextureARB), but that's more a result of card manufacturers optimizing their texel units for write-only, rather than read/write. Multi-texturing can be just as easily implemented with alpha blending two images; however, when writes are 2-3x faster than reads, it's considerably more efficient to pre-alpha blend the pixel in the texel unit and write once, then to write once, read back in, and write again.
And no, there is no reason for D3D to be tied to Windows. It could be implemented on Linux. Just nobody does it. Because it wouldn't be considered very cool :)
Actually, since neither ActiveX nor the Component Object Model exist in a complete enough form for any other OS (MacOS has a partial implementation), it's easily arguable that DirectX will remain tied to Windows. Not too many operating systems have any idea what extending IUnknown is supposed to do. Hell, check the Cygnus Gnu-Win pages - it's difficult just to port the DirectX libraries to a different Win32 compiler, let alone an entirely different operating system!
However, that said, Intel's current push is really pathetic. When processors like Sony's Emotion Engine at 300Mhz are 2x faster than Intel's flagship at 1Ghz, Intel's design is clearly wrong. The fact that Intel has had difficulty reliably producing P3s at 650Mhz or above doesn't help their situation. This is a wonderful marketing move (and will probably convince less techno-savvy consumers away from AMD), but really amounts to nothing more than behaving like an attention-starved child.
If AMD could get SMP Athlon's working, Intel would be in severe trouble. Higher clock speeds are relatively useless - a 4-way 650Mhz Athlon would dust almost anything Intel could muster. I'd love to have a dual 650 Athlon for rendering; unfortunately, it seems as if AMD's methodoly mirrors Intel's - more speed, not better speed. Sigh.
Actually, as a current student of "the art," I can say that there is quite a bit more to CS than one may imagine. Analysis of discrete algorithms, combinatorics, and number theory do play quite a large role in CS (we use Cormen, Leiserson, and Rivest's Introduction to Algorithms, which I find to be an excellent book); however, there is also the formal logic side. My formal logic class used Manna and Waldinger's The Deductive Foundations of Computer Programming. Honestly, I didn't care for the class, and hated the book, but if you truly wish to follow "The Art," your best path would probably be to forget learning computer programming and languages for a while. Honestly, learning a new language is a process that should take at most 2-3 weeks (maybe more if you are moving from structured (C/C++/Pascal) to functional (Lisp) paradigms), but the thrust of computer science is hardly the syntactical usage of one particular programming language, nor has it ever been. When I write programs (currently working on an operating system, and have done 3D games, paint programs, and others in the past), my time division is about 35% design, 15% programming, and 50% realizing that I typed >= rather than >, or + rather than -. Design is entirely language and platform-neutral. I create a pseudocode as I go to explain what my algorithm, data structure, or function should do. That is what CS as "the art" teaches - design independent of all bounds. Programming is then the implementation of that design - it will teach you things like platform-specific memory alignment, cache optimizations, etc., and how to take your elegant design and make it scream on the targeted hardware. It's much easier to learn the facts for a hardware platform and implement a canned algorithm than it is to learn the philosophy behind algorithm design. Computer science will teach you design, computer programming will teach implementation, and experience teaches debugging. If you just want to write cool programs, then Knuth's (or Manna's, or CLR's) books are entirely unnecessary, and you would be better served by a reference manual such as the Red Book. However, if you do find the art interesting, then be prepared for a lot of seemingly useless math, esoteric concepts, and trips to the mental ward. I find most of it very interesting, but there is a reason my friends expect me to end up in a rubber room before I graduate.
One of the easily identifiable reasons why there has been a decline in computer science majors is due to the rise of the hybrid major. It is a recognizable fact that females have a much higher presence in humanities courses in high school than they do in computer science courses (my AP English class was divided 7/27 (7 guys, 27 girls), whereas my AP CS class was 12/0). While this may be in part due to the "table of [elitist] geeks" that has been mentioned, I distinctly remember seeing quite a bit of enthusiasm among my female classmates during humanities discussions. After reaching colleges, they may uncover a hidden interest and talent for computer science; however, developing operating systems, raytracers, and compilers often comes at the cost of other topics - namely humanities. It's difficult to simply shed an interest that you held for 4 years to embrace a field entirely foreign to you.
Enter then the hybrid major (Symbolic Systems). Rather than dropping all participation in fuzzy classes, you can divide your major among your new CS interest and your old (lifelong?) fuzzy enthusiasm. I know my university has seen a tremendous increase in enrollment in our SymSys major (by both females and males), often by people who were recently exposed to CS, but unwilling to pursue it full-time (I have worked over 90 hours/week for some projects, so it truly is a full-time commitment).
Maybe I'm using too much hyperbole, but I think it is fair to assume that the majority of females that enter the computer science major didn't intend on pursuing computer science the majority of their lives, and some may have chosen simply based on positive experiences in their first college course or two. I'm basing this assumption simply based on the extremely low enrollment of females in high-school CS courses, so I think it is at least partly justifiable. When given the option of a half-techy/half-fuzzy curriculum division (such as SymSys, Human-Computer Interaction, or any other hybrid major) or a balls-out techy curriculum, I would assume that most of these people would choose the hybrid. Maybe my school is unique since we don't need to declare majors until our junior year; however, I think a truly correct study would compare the declining enrollment in pure computer science (especially among females) with the burgeoning enrollment in hybrid fields (especially among females). Hell, there are times when I'm sitting through formal logic classes where I've thought about the appeal of the hybrid major - all of the pay, none of the bullshit. After all, proving that 0++ is equivalent to (0+)+ and that that is equivalent to 2 is really about as enjoyable as a trip to the dentist's office. I love CS, but there are some things that are just ridiculously painful about majoring in it (e.g. the parts that *aren't* in a hybrid curriculum).
I'm going to assume that this wasn't flamebait or tongue-in-cheek; however, feel free to correct me.
I've had this discussion several hundred times, with every ethnic and religious combination that I can think of, and I have yet to hear a convincing argument on why vulgar language is an "ugly black mark" on those who use it.
When I read Gravity's Rainbow (one of the best novels of the past century), I never for a second thought that Pynchon's sentences were anything but a comment on his genius, but I'm sure that any dozen pages or so of it have enough profanity to ensure it a solid R rating.
The problem with blanketly saying that profanity is an ugly mark on its users is that it neglects the fact that a profane sentence is not necessarily an obscene one. I can easily take perfectly non-profane words and form them into a sentence that would turn the average sailor green, or I can cuss a blue streak and without anyone would think about it. The problem comes from the context of the words - not the words themselves. I really fail to understand why "fuck" is deemed inappropriate by the majority of American churches (especially seeing as fucking (procreating) is the only legitimate metric for marriage), but a word that carries as many awful denotations as "crucifixion" can be thrown about by the average 3rd grader.
Your choice to not use profanity is just as valid as mine to use it. The metric for intelligence shouldn't be whether an individual uses a very specific subset of the English/Japanese/Spanish/etc. vocabulary, but the entire subset that said individual uses. I don't have much respect for people who can only use obscenities; however, I have just as little respect for people who constantly interject "like, uhm..." and other nonsensical gibberish into their sentences.
While you may find the lack of clean humor unfortunate, I find it vindicating. Old-style conservatism is one of the most authoritarian and fascistic hold-overs from Euro-American history. It caters to the belief that what works for someone important must work for everyone, and if somebody more important than me doesn't like something, than I shouldn't either. However, all religions agree that I am a sentient being - therefore, I am capable of making choices that affect me by myself, and you by yourself. You may not agree with me, and I may not agree with you, but dammit, that's how it is, and nobody can deny us that (I think Voltaire wrote something similar...). South Park may cater to "the lowest common denominator" (I tend to disagree with that assertion), but dammit, somebody sure needs to - how else would I, as an elite member of this LCD, get my weekly ration of comedy?
I doubt what I believe will influence your opinions at all. In fact, I am probably already burning along with Trey Parker, Matt Stone, Adolf Hitler, and the rest because I choose to eschew some of the mores of old-style conservatism. I personally try to paint as few "ugly black marks" on people as possible (you know, it's difficult to find beauty when everything has an ugly black mark on it) - and for that reason I set my "offensive tolerance" quite high (e.g. it'll take more than a few dirty words to offend me), and my "black mark" level even higher. I also think that if everyone started painting significantly fewer ugly black marks, the intolerance and prejudice that culminates in nightmares like Columbine would also disappear, but that's a topic for a different rant.
Mhz are one of the most meaningless speed measurements these days, *especially* when referring to console processors.
The PS2 has a 300MHz processor, but what is hidden is that fact that it has 2 additional 300MHz vector coprocessors, creating a total of about 12 floating point accumulators and 6 dividers that can be used simultaneously, yielding (yet another meaningless statistic) 6.2Gigaflops. However, since VU0 and VU1 are heavily optimized for vector processing (as opposed to dividing by 0), the majority of the 6.2 billion floating point operations can be utilized in the average case. Add to that the Emotion Engine's fully 128-bit design, and you've got a processor that can outrun any intel processor up to ~1.5Ghz, and a memory bus that can actually supply needed data when it is needed. Granted, a Willamette 1500 would be able to do integer operations much faster than the Emotion Engine, making it a better general-purpose CPU; however, for gaming and signal processing integer operations are minimally important, making the EE a much better processor for multimedia purposes.
What MS has forgotten is that the PC platform is a terrible architecture for gaming. Everything about it is just mere upgrades on 20 year-old technology. Maybe Microsoft has something really sneaky up their sleeves, such as using a 1Ghz SH-4 (Dreamcast processor) and a custom-designed bus to an optimized GeForce 256 processor. That could (theoretically) approach PS2 level performance, but I'm not going to give MS that much credit.
There are only 2 things wrong with your post:
/. showed that most readers were students. Based on the language used in many posts, it can justifiably be assumed that most of these students are still in high school.
A recent poll on
Now please, o exalted one, show me a high school that covers not just introductory nuclear physics, but also Cold Fusion 101. We only got to introductory E&M in AP Phycics C in my high school. I consider myself a nerd (engineering student at Stanford), and I know jack about Cold Fusion - it's not something that I read (or research) religiously. Very few people do... proportoinal to the population of the world, there are very few nuclear physicists.
However, if the facts in the article are true, it does suggest something. I can't prove that the article is 100% factually accurate, I can't even prove if it's 50% factually accurate. But I would like it to be true (1.21 gigawatts!). There is a reason Americans (and almost every other modern culture) are suckers for conspiracy theories, and it goes hand-in-hand with increased national cynicism. Cynicism isn't something that justs randomly manifests itself in cultures - there has to be some national reason for it. Americans can no longer believe their government... it's not the responsibility of the Americans to find renewed faith in their administration, it's the responsibility of the administration to prove its honesty. If the article is true, then they've got some explaining to do.
Hewlett-Packard will be releasing a DVD ReWriter (DVD-RW) in September. It will write DVD-RWs at 1.5MB/s (maybe 1.25x DVD speed), and CD-RWs at around 8 or 12 speed. The DVD-RW discs will be 3.0GB single side, single layer, but they may only work in DVD-ROMs... I don't think they will work in DVD players. The street price should be $699, with media being $30 / blank, and the drive will be SCSI only.
This is about as likely as ICQ becoming a commercial app due to AOL's purchase. If anything it will get rid of the $10 shareware fee on WinAMP, since Nullsoft is now backed by the huge pockets of America Online, rather than the 20 or so users that ever registered it.
Of course, if AOL decides to charge for WinAMP (this is about as likely as Microsoft open sourcing Windows), you could just continue to use 2.10, which works wonderfully. I only just upgraded from 2.05, and I really don't notice too much of a difference. I didn't want to use any AOL programs for a while, but they didn't pawn ICQ space to the highest bidder, and I doubt they'd do it to WinAMP. I wouldn't be terribly surprised to see AOL license the Sorensen algorithms next... it looks from their previous purchases (real-time communication and conferencing, audio compression/decompression) that they are looking to become a net tech company. Video compression is the next step.
It depends on what audio format is used. All reels have a 2-channel analog stereo track on them for backup purposes, but additional digital audio tracks are handled differently for each format:
DTS: A seperate CD is included with the reel that has the DTS 5.1 encoded audio on it
SDDS: 2 highly compressed digital tracks (identical, in case one breaks) are run along the rings of the reel, making this format highly susceptible to being damaged. Allows up to 8 channels (5 front, 2 surround, 1 LFE).
Dolby Digital: I think this uses a seperate reel in addition to the movie reel, but I could be mistaken. Allows 5.1 discrete channels, Dolby Digital EX uses the same 5.1 channels, but EX processors matrix decode (a la Dolby Pro Logic) a rear-center channel.