Slashdot Mirror
Sony's Monster Graphics Chip
GFD writes "EETimes Has an article about a monster (462-mm2!!) graphics chip discussed in a paper at the ISSC. The numbers are astounding such as 256 mbits of on chip memory. Barely manufacturable though..." I'd still love to see what that bugger can do... bet it still can't simulate realistic hair in real time ;)
anymore, as most porn starlets are shaven, or maybe have a thin, sleek landing strip.
PS2, for all its l33t hardware, doesn't seem too impressive. For all that neat stuff, its designed for benchmarks, can't really use it all that well and its too hard to develop for.... when they make use of this thing, will they have the same probles, eg "Hooray, can render up to 65536x65536 res texture maps on over 4 billion polys... but its only got 4 megs of video ram". Or something to that effect. For that matter, when you get to that level how well can a human develop for a platform? Modelling gets tougher and tougher as the renderers get better.... Making more polys, better texture maps, multiple maps (bump, alpha, luminosity, reflection, etc) for layers, blenders, better frame rates for animations.
I'm all for this hardware, but ya gotta wonder: can we even properly use it.... then again, that's been said many times before.
I've just finished reading the article. A few thoughts spring to mind:
First of all, this sounds like the Emotion Engine hype all over again. It might be an amazing chip, but it'll probably just be "decent" when it finally gets here.
Secondly, don't expect to see this in quantity until 0.15/0.13 micron fabs get here. Remember the Emotion Engine. Fabbing a chip that big is a royal pain. It'll get much easier when finer linewidths shrink the die size.
Thirdly, CMOS fabrication processes can be optimized for good quality DRAM, or for good quality logic. Not both (without throwing lots of money at it). The two types of circuit have contradictory requirements for transistor characteristics. In practice, this has meant that DRAM-plus-core chips have either had slow cores or bulky, slow, hot DRAM.
The only saving grace is that most of the chip area will be DRAM. This means that most of it will be tolerant of manufacturing faults (you usually have more DRAM rows than you need, and cut out the faulty ones before packaging). This is the only thing that will let them fab a chip this size at all.
The chip should provide interesting perspective when it arrives (much as the Emotion Engine did), but I don't expect it to take the world by storm.
Toshiba has expressed interest in offering the 128-bit processor for high-end routers and switches.
For....graphics? "Hey, this is great!" "What are you talking about, we lost two whole subnets!?" "Yeah, but look at how beautifully those error messages are rendered"
--
Why pray tell are you running at such a low resolution and why are you satisfied with such a low framerate. I for one will not be happy with anything less than 1280x1024 and ~70-80 fps. So no this chip is far from the final chip.
Cypherpunks: Civil Liberty Through Complex Mathematics. Those who live by the sword die by the arrow.
While it is true that something like 50 million polys/s would be the upper limit for the number of renderable polys on a screen of that size, you are forgetting about all of the hidden polygons necessary to build a realistic scene.
I've seen estimates that figure it would take about 50-200 million polygons to render a modest scene in photo-realism. Now multiply that by 60 frames/s. You are already talking about 3-12 billion polys/s here, and we haven't even started talking about extremely complex surfaces like hair/fur/grass/leaves.
I think we will be building chips for some time before we reach the same clarity with 3d that motion video currently does in 2d.
ah kiss --- such nonsense.
>Therefore, logically, when we reach 50 million
>polygons/second in calculation for a graphics
>chip, it is effectively impossible to make the
>graphics quality any better without improving
>the quality of the screen.
Oh Bollocks. Just spitting a pixel to the screen has nothing to do with the overall quality of the image that is produced. Anti-aliasing. Motion blur. Depth of field. Programmable shading (no more of this gourand/phong with badly mapped textures etc etc). Don't even get me started ---- TONS of effects that can be incorporated. Hair, fur, skin, particles, atmospheric effects, lens effects, volume rendering effects, etc etc etc.
Until you can make a CG image indistingishable from a live source at that resolution there is TONS that can be improved.
Have u worked in the graphics biz? I have......
j
Jesus taco, enough pr0n talk already...
--
(snip)
I assume the mean that the wafer is 21.7x21.3mm^2, this is a little under an inch to a side. At first read I thought they meant total package size, which isn't that impressive considering even the size of an old PII.
So this chip has the same fill rate, but 8x the RAM, only 2x the RAM ports, and 7x the complexity?
It sounds to me like Sony have just made this an 8x multitexturing part at *huge* expense. And an 8x multitexturing part with only 2x the internal bus for texture cache reloading. Slow.
And supersampled antialiasing will cost you 75% of your fillrate, since that isn't increased either.
I just don't understand who this chip is for.
"The numbers are astounding such as 256 mbits of on chip memory."
Wow, two hundred fifty six millibits of on chip memory. That's like, what, almost 1/20th of a byte?
Indeed one would normally expect a chip of this size to suffer yield problems as dictated by Murphy's law (that's the REAL Murphy's law, where fractional yield is given by
Y = ((1 - e**(-AD))/AD)**2, where A is area, D is defect density
rather than the other Murphy's law which affects all our lives). However there are remedial measures, especially with DRAMs, which can keep Murphy at bay. These largely have to do with redundancy, which is to say one designs the RAM array with many more rows than can actually be addressed, then one detects dead or malfunctioning rows at device test and substitues in the spares. This is a relatively easy thing to do with a nice regular structre like a RAM.
Also one wonders whether with that many polygons squirting past the eyeballs, is it acceptable to ignore a modest number of defects? After all, human vision is reasonably fault tolerant compared to many computing applications.
Even so, I take my hat off to anyone who gets acceptable yield from a device more than 2cm on a side. RESPECT!
Robert
Nemo me impune lacessit
HDTV doesn't get better then 1024x768 at 60Hz
:-) ) supports resolutions of up to 1080i which equals 1920x1080???
I thought that HDTV (if it will ever exist
Doh!
The X-box won't necessarly be the most powerful console. Nintendo's Game Cube is shaping up to be a pretty powerful machine (400MHz PowerPC Gekko (similiar to G3) processor, 24MB 1-T SRAM main memory, 16MB Graphics/Sound Memory, Art-X video card, etc.) Also, Nintendo is well known for their video game characters (Mario, Luigi, Zelda, Link, and that damn Pikichu), so the machine won't be pure hype like the X-Box...
Doh!
Yeah, but the size in the article was 462 *square* mms. That's not quite so big, and is under a square inch. 462 / (25.4 * 25.4) will give you the approximate size in square inches.
Actually the resolution of a standard TV is far less accurate than that of any pc monitor out there. It's just that you don't notice it because you're too far away, the signal is analog instead of digital, and the pixels are auto-antialiassed. I think somebody once told me the resolution of a TV corresponds to something like 625 lines, 50 Hz, 2:1 interlace, 4:3 aspect ratio. So if you play your game on a regular TV you're wasting an awfull lot of detail (& computational power), still all these powerfull and expensive gaming consoles usually are connected to home TV's.. strange thing, no ?
With great power comes great electricity bills.
Possibly, but the "i" means that you only have to do half the refresh rate. Substantially reducing demand
-Michael
-Michael
Though I agree with you, just wanted to nit-pick:
Hidden surface removal is only really a factor when the card can't handle the current volume. It scales with the complexity of the scene. Plus there are technologies such as ATI's (and now nVida) that help reduce the effect considerably.
FSAA is largely irrelevant when you achieve high enough resolutions. Results may vary though.
Stereo has never been a major factor, nor do I think it'll really catch on; especially on a console, unless you split the output signal.
Still there are plenty of other common sence arguments promoting the continued bleeding edge development. Not least of which is the fact that the intro's are still rendered seperately.
-Michael
-Michael
If I remember correctly, our eyes see at ~430 FPS so we still are talking about a long way to go to make a "realistic" image where the eyes physically are slower in refresh compared to the video.
I hardly think our realism barrier consists mainly of faster-than-TV refresh rates. If that was the case, I could get out my CGI-pong video game and run it up to 1,000fps.
Better yet, net hack!!!
-Michael
-Michael
as many others have replied you're missing a lot about what pixel rates are about (hidden pixels, alpha blending, antialiasing etc etc)
However there is a grain of truth in what you're getting at that in the future may eventually result in a whole new generation of hardware. basicly it's this - the number of visible pixels on the screen isn't really changing much, certainly not at the same rate that the ability of silicon to manipulate them is .... which means that rendering techniques that are proportional to the number of pixels (rather than screen complexity) may become more interesting - for example - ray tracing - the number of rays is (to a 1st approximation) is proportional to the number of pixels rather than the scene complexity (however the cost of processing each ray also goes up with scene complexity - but not necessarily always at the same rate if you are carefull) ....
Question:
how is sony expecting to get decent yields on such a big die? The probablity of a chip flaw goes up with the surface area
Answer:
256Mbit = 128Meg Byte = approx 500Meg of HIGHLY symmetric transistors. We're already building chips with 20, 30 and even 100Meg of complex transistor layouts (granted, most is in symmetric caching or register sets). Additionally, single ported memory is a lot simpler than multi-ported LRU-tagged cache. So while Intel, AMD, Alpha, SUN fuss over 4Meg L2 cache sizes, we're all in a similar ball-park.
Next, a little over a year ago, I read an article about a new DRAM memory architecture that was designed for extremely high yields.. Basically you'd have dozens, hundreds or thousands of mostly independant memory cells, then after the testing stage, you marked which cells were good, which then allowed the memory to ignore bad chunks transparently. As long as you met a minimum memory-size, you were golden. If a similar technology is used here, then they'll probably over-allocate it a bit, and allow down to like 70Meg to be considered passing.
However, I've read other interesting questions such as "are they going to optimize this for power consumption / heat dessipation or performance?"
-Michael
-Michael
This story puts everything in bits it seems. As a result the numbers appear much better than their byte counterpart. Here are some things I noticed.
:)
462mm x 462mm? You want large? There are 25.4 mm per inch. This is saying it is almost 18.2 inches x 18.2 inches (can anyone say 1 1/2 feet by 1 1/2 feet?). I sure hope this is a mistype of the actual size of this beast.
256mbit of memory? that comes out to 32MB of memory. I got a geForce2 GTS coming in the mail with 32MB of mem. Granted the memory is embedded in the chip itself but I think that would result in the price being alot more, especially if you want 64MB of mem.
75 million polys/sec. Sure, when the chip has nothing else going on, doesn't have to worry about lightning, textures, and the triangles it is drawing are all touching each other so there are less vertice's to draw. Splitting the triangles up so there are 3 vertices being drawn per triangle will easily drop this number to be 1/3 of it. Throw in some lightning and it drops more. Same with textures.
"the chip can process 75 million polygons per second, has a pixel fill rate between 1.2 and 2.6 gigapixels/s and can draw 75 million polygons/s". Anyone like being redundant? I count 2 things in there and it seems they are searching for features.
A 2,000 bit internal bus means a 250 bytes internal bus. Why 250? why not 256? Most chips have the maximum internal bus the size of how many bits the chip can handle. If the chip is a 128bit chip then it appears to have a bus double of that so it is feeding the chip faster than the chip can empty it. This could be good but it can also be bad.
With all said and done, the sony graphics chip is 4x as big as nVidia's geForce2 GTS and only 2x the power. Yep, lets slap a huge beast into a machine that probably sucks up the power supply and generates more heat than the CPUs
See Shenmue on Dreamcast, particularly on the Passport disc where you can zoom in close to the characters' faces. The game supports the best realtime skin and hair work I've seen (right now, beating the Playstation 2).
- I don't care if they globalize against free speech. All my best free thoughts are done in my head.
So this new Uber-chip does 75 million triangles, and has a fill rate of 1.2 - 2.6 G/pixels. Doesn't that seem familiar to anybody? Those are the box specs for the PSX2. The extra memory will be quite helpful, but this isn't very impressive so far.
That's 400 SQUARE MILLIMETRES which is less than a chip measuring 1 inch x 1 inch.
I don't think this is correct. An HDTV that does 1080i can't do 1080p, because that isn't part of the spec. the highest progressive rez is 720p. Lucas is using special cameras to do 1080p 24fps recording, but that's cause he has money. Perhaps this is where the confusion is coming in?
No, no, no.
256 millibytes = about a quarter of a byte.
That would make it 2 bits.
This chip contains a Shave and a Haircut.
:)
462 mm^2 == (21.5 mm)^2, or about (.85 in)^2.
I hope to clear up a few misconeceptions that people seem to have. I have read some of the replies and it seems that most people are making valid points that are not taking into account all of the factors involved.
1. As many people have noted, raw polygons are only the underlying factor in the visual quality of a scene.
2. There are things that are wasteful both in memory bandwidth and processing power such and redundant pixel redering on the z-buffer. PowerVR, ATI, and NVIDIA are all using techniques to bypass rendering more pixels than necessary. PowerVR in a current card, which is in the middle end because of a lack of hardware T & L (transform and lighting). PowerVR is using what is called tile based rendering, which is a more elegant way to reduce load. ATI has a somewhat less pure technique called hyper Z which decreases memory bandwidth usage (as seen in benchmarks at very high resolutions), and NVIDIA is doing something similiar with the NV20 but doesn't have anything built into their GeForce cards.
3. Yes multi-pass rendering is a factor, but at the same time techniques are being used to render multi-textured polygons in one pass instead of many. PowerVR's has this feature (at least when used with DirectX).
4. Yes, anti-aliasing is a factor also, but 4x4 anti-aliasing doesn't have to require 16 times the rendering power. Only the pixels that have enough contrast to contribute to jaggedness in the first place need to be assesed.
5. The limit how many polygons actually need to be rendered is MUCH less than one per pixel if enough optimization tricks are used. When proper smoothing algorithms are used, nothing distiguishes a highly facted sphere from a regularly facted sphere except for the edges, which will be smoother with increased polygons. A low polygon count object's edges can be smoothed more efficiently with some 2D tricks. Objects far in the distance can be simplified so that the aren't taking up more polygons than necessary.
6. More power can always always always be used. And not just for higher resolutions eighther. One of the things that is so cool about console systems I think, is that they are made to run at 640x480 so they can use plenty of effects and in the end up the visual quality quite a bit.
7. This took me a while, and I didn't preview it.
This Wiki Feeds You TV and Anime - vidwiki.org
think the real push should start moving away from higher polygon rates and more towards greater visualization enhancements for each polygon. We're already dealing with cool things such as environmental bump mapping. I'm still waiting for the fully featured ray-tracing engine. I'd be perfectly happy with a scene that was only 30fps, 800x600, average number of polygons if I could just feel the glimmer of living light.
If you have decent calculation engines on-chip, you can use a silly polygon throughput to emulate nicer features that might be difficult to implement directly. Tesselate large polygons to make NURBS surfaces. Add multiple semitransparent "halos" for fancy lighting effects. Use various sneaky tricks to emulate volume effects like smoke and Ye Canonical Plasma Field. Etc.
You can do all of these in the main CPU, but it bogs down the CPU like crazy and saturates your system bus (sending all of those triangles to the chip). If you can get the chip to do it for you, then it'll look almost as good as real curved surfaces/lighting/etc, without hogging system resources (just rendering resources).
While a true hardware implementation of nifty features would be more efficient, the brute force approach lets you use mainly well-understood designs, and lets you patch bugs in firmware instead of needing a new chip revision.
No idea what Sony's actually going to do.
Look here (http://www.duhaime.org/dict-b.htm) man, and get that chip away from me!
I can't believe he mentioned hair in the same sentence. Ewwww!
Friends don't help friends install M$ junk.
Money of course, and they see a good market in video games
:-) ), and keeping IE as the dominant browser (which in turn allows them to sell IIS as a viable solution). I think that Microsoft was partially threatend to create a game console
I agree with you to a point that M$ is in the game to make some money. I don't think that Microsoft would have entered if Sony had hyped the PS2 as a game machine, but instead Sony hyped it as a home entertainment center (ie: USB & Firewire ports, theoritical internet capability, DVD, etc), which would slice into Microsoft's business model of selling Windows (heh, and both WebTV units
Doh!
Comment removed based on user account deletion
Before any of you freak out about the chip being 1.5ft^2... the actual size is actually reported at 46.2mm^2. Read more closely.
First off.. The memory size was a typo, but too late for that.
Second, Mega means 1E6. M/Meg often refers to 2^20, but not always.. Depends on what you're talking about (such as hard drive, system memory, etc).
-Michael
The bus runs at CPU speed. However it is still DRAM, hence the latency is a lot higher than your L1 cache. But yes, you have the right idea, its a lot faster then it being off-chip, but not so fast as your L1 (or L2 or even L3 (on a P4)) cache.
And of course, for graphics use, latency is less of an issue then raw bandwidth, because you aren't jumping around looking at a lot of different places in memory (like a PC OS) but you are trying to grab large sections of memory (textures) and keep 'em coming (bus speed).
Spyky
You glossed over one point, and overemphasised another in my mind.
The overemphasis was k vs. K. It's not an extreme mistake to use K instead of k, because K has no meaning of its own. It _is_ still wrong, but it's not as egregious a mistake as confusing b/B or m/M.
The second point isn't one of SI notation, but strictly computer notation. When talking about computers, counting is ALWAYS done in powers of two! So...
k is 2^10 = 1024 NOT 1000!
M is 2^20 = 1024^2 = 1048576 NOT 1000000!
The reason this confusion came about was that drive manufacturers found they could up the advertised size of their disks by nearly five percent, and sell more of an identically sized drive than the competition. The lie of k=1000, M=1000000 in computing was pure and sleazy marketing. No more.
"People who do stupid things with hazardous materials often die." -- Jim Davidson on alt.folklore.urban
When dealing with communication channels, k=1000 and M=1E6. Bytes, or more properly, octets, are a unit for storage devices.
The usage of b=bit and B=byte is not universal. BPS, KBPS and MBPS refer to "bits per second", not "bytes per second". These were in widespread use long before bytes became a common unit.
Mea navis aericumbens anguillis abundat
billion used to be 10^12, but now common UK-english has billion at 10^9, and trillion at 10^12. I,m not sure when change occured, I vaguely think it changed about the same time a shilling became 5p instead of 12d.
Whenever some says a billion pounds they mean 10^9 not 10^12. You must find the financial news very confusing.
http://rareformnewmedia.com/
Oops! I should have said storage systems, rather than computing. Quite right about communication channels--I stand corrected.
As for the BPS/KBPS/MBPS notation, they predate ASCII, and mixed case digital notation in general. I haven't seen them used except by the old guy shuffling off to retirement, for at least a decade. Ethernet, token ring, and modem communications all seem to use mixed case notation now.
"People who do stupid things with hazardous materials often die." -- Jim Davidson on alt.folklore.urban
By writing a driver for POVray and given enough processingpower could I use beowulf-cluster as my graphics card? And play nicely rendered counterstrike? ..and and get more girls?
Yes, no (unless you like playing at 5 beautiful frames per second), and only if they're geeks, respectively.